Analysis of the Impact of Policies and Meteorological Factors on Industrial Electricity Demand in Jiangsu Province

Abstract

Under the strategic background of “carbon peak by 2030 and carbon neutrality by 2060”, the impact of energy policy on China’s industrial electricity demand is increasingly significant. This study focuses on the industrial electricity demand in Jiangsu Province, comprehensively considering the impact of policy and meteorological factors, and uses multivariate regression analysis to systematically explore the impact mechanisms of policy adjustments and climate change on industrial electricity demand. First, by analyzing the policy background and climate characteristics of Jiangsu Province, relevant policy and meteorological indicators are extracted, followed by a correlation analysis and the establishment of an industrial electricity multivariate regression prediction model. Finally, the evolution of the industrial electricity load in Jiangsu Province under different socio-economic pathways is forecasted. The results show the following: (1) Policy factors such as the electrification rate and self-generated electricity show significant correlation with electricity demand, as do meteorological factors such as temperature. (2) The future industrial electricity level in Jiangsu Province is expected to show a fluctuating upward trend, with industrial electricity consumption reaching 767.51 to 794.32 billion kWh by 2035. Accordingly, the forecast results are expected to guide future planning of the industrial electricity system in Jiangsu Province under the carbon neutrality scenario.

1. Introduction

The accurate forecasting of electricity demand is crucial for the sustainable development of governments, enterprises, and societies, especially in the complex and dynamic environment of China’s electricity market. Reasonable predictions of electricity demand can help ensure the reliability of power supply. In recent years, with the introduction of China’s “Carbon Peak by 2030 and Carbon Neutrality by 2060” strategic background, China issued the “14th Five-Year Plan for the National Economic and Social Development and the Long-Range Objectives Through the Year 2035” in March 2021. It is evident that carbon emissions are increasingly emphasized, and various energy policies are being introduced around carbon emissions. The power sector is a major source of carbon emissions in China, accounting for 47.39% of the total carbon emissions in 2019 [1]. Renewable energy serves as a substitute for fossil fuels in the decarbonization process of the power industry. To support the development of renewable energy and restrict carbon emissions, China has implemented various regulatory policies (such as carbon pricing) and support policies (such as feed-in tariffs, R&D subsidies, and tradable green certificates) in the power sector [2]. By 2020, China had become the world’s largest producer of renewable energy, thanks to these energy policies [3]. In this context, energy policies increasingly affect electricity load forecasting. Additionally, climate change has become a crucial concern for power market managers. This underscores the need to understand its impact on electricity demand for future power systems development.

Historical models of electricity demand have primarily relied on historical data and meteorological factors. However, in the context of a national emphasis on carbon neutrality, policy factors are playing an increasingly significant role in the electricity market. Traditional forecasting models have failed to fully consider the potential impact of policy macro-control on electricity demand. Additionally, past studies have lacked simultaneous consideration of both policy and meteorological factors, potentially leading to omissions and errors in existing electricity demand forecasting models. This study aims to fill this research gap by employing multivariate regression analysis to deeply explore the complex mechanisms of the impact of policy variables and climate variables on electricity demand. By thoroughly investigating the dual influences of policies and weather on electricity demand, this study, using Jiangsu Province as a case study, aims to provide a more forward-looking and sustainable electricity demand forecasting model for China’s power market and to offer decision-making support for achieving carbon neutrality goals.

2. Literature Review

According to the forecasting horizon, electric load forecasting can be roughly divided into three different categories [4]. These categories are short-term load forecasting (STLF), medium-term load forecasting (MTLF), and long-term load forecasting (LTLF).

STLF aims to predict the load from the next thirty minutes up to fifteen days into the future [5]. It facilitates the planning of future power supply in the electricity market and demand-side management (DSM) [6]. MTLF typically involves forecasting from one month up to one year into the future. Such forecasts aid in revenue assessment, unit maintenance planning, and energy trading, among others [7]. For LTLF [8], the forecasting range covers one to ten years, and its purpose is to estimate load to assist energy planners in expansion planning, future investments, and distribution planning in the power system [9].

He et al. [10] proposed a power demand forecasting method based on system dynamics, specifically targeting long-term urban power demand under the new economic normal. Additionally, the study notes that current research on long-term load forecasting (LTLF) primarily concentrates on two aspects: (1) analysis of factors influencing power consumption; and (2) analysis of electricity consumption forecasting methodology.

2.1. Analysis of Influencing Factors

In terms of the statistical analysis of influencing factors, traditional long-term electricity forecasting factors mainly include economic [11], social [12], and meteorological factors [13]. Economic factors encompass industry output, relevant industry investments, and electricity consumption prices, while social factors mainly include population and Gross Domestic Product (GDP). Meteorological factors typically include temperature, rainfall, humidity, and wind speed. In terms of the accuracy of forecasting methods based on historical data, measured weather parameters and load data are the most effective parameters [14]. Moral-Carcedo and Pérez-García [12] considered historical load, historical electricity supply capacity and meteorological data over various time scales to enhance the flexibility of external inputs affecting load forecasting in the Spanish electricity market. De Felice et al. [15] considered regional economic and climate data over different time scales to address the issue of single-source data in the European Centre for Medium-Range Weather Forecasts (ECWMF) electricity impact studies. Ang et al. [16] considered average temperature, time trends, and seasonal variations (holidays) to improve load forecasting accuracy in Singapore and Hong Kong. Chabouni et al. [17] considered cooling degree days (CDD), heating degree days (HDD), fixed seasonal variables, and movable seasonal variables for industrial load forecasting in Algeria. Zheng et al. [18] considered GDP, population, electricity prices, CDD, and HDD for urban load forecasting in Guangzhou, China. In recent years, energy and carbon emission targets have become key factors in medium and long-term electricity forecasting, with the macro-control of energy policies being strengthened. Energy policy objectives directly affect electricity consumption and energy structure reorganization in different economies [19]. Researchers, based on the evolution of the current economic and energy systems, construct various development assumptions by setting business-as-usual scenarios, policy planning, and green scenarios to comprehensively examine future changes in electricity demand [20,21].

Current short and long-term electricity demand forecasting models often treat policy factors as static scenario settings rather than dynamic parameters or independent variables. This approach neglects the fact that the actual impact of policies has already occurred in the past and will continue to affect the future. This bias towards considering policies as static conditions may limit the models’ ability to accurately capture real market dynamics, as the substantial effects of policies should be seen as changing factors, not constant background conditions. On the other hand, they overlook the quantitative and qualitative relationship between energy policy decisions and other factors influencing electricity demand, such as meteorological factors.

2.2. Methodological Analysis

From a methodological perspective, load forecasting techniques are categorized into traditional load forecasting and machine learning algorithm-based load forecasting [22]. Traditional load forecasting methods include the following: Time series forecasting, for example, Li et al. [23] applied the Seasonal Autoregressive Integrated Moving Average (S-ARIMA) for electricity load forecasting. They proposed a modeling scheme that preprocesses data effectively within an acceptable computational complexity, yielding satisfactory results. Louzazni et al. [24] developed an estimation method using the ARMAX model to determine the power generation of Cairo’s grid-connected photovoltaic systems in Egypt. According to statistical standards, this model outperforms others in forecasting load and electricity prices. Econometrics, such as He et al. [10], who designed a mid- to long-term electricity portfolio forecasting model based on econometrics and system dynamics to address issues of low accuracy and unstable forecasting errors in load forecasting. Pérez-García and Moral-Carcedo [25] employed a growth rate decomposition approach to solve the problem of the mid- to long-term power demand forecasting necessary for the rational planning of Spain’s electricity production and transmission system dimensions. Vu et al. [26] used methods of multicollinearity reduction and stepwise regression elimination to select appropriate climatic variables for forecasting electricity demand in Australia, thus improving the model’s prediction accuracy. Regression analysis, for instance, Şahin [27] used “Non-linear Regression (NLR)” to estimate the share of renewable energy and hydropower in Turkey’s future total power generation. Durmaz et al. [28] estimated Hong Kong’s monthly electricity consumption using the “Autoregressive Distributed Lag (ARDL) model”, mentioning ways to reduce electricity consumption based on the climate impact of new residential electric fees. Torrini et al. [29] predicted residential, commercial, and industrial electricity consumption in Brazil using fuzzy logic and regression-like techniques. Aslan et al. [30] forecasted the monthly electricity consumption of China’s industrial sector in the coming years using multiple regression, quadratic regression, and Artificial Neural Networks (ANN).

Machine learning algorithms for load forecasting are widely used in short-term load predictions. For instance, Luo et al. [31] proposed a multi-objective forecasting framework based on machine learning (ANN, SVM, LSTM) to enhance the prediction accuracy of multiple building energy loads and BIPV power generation. The results showed that the model based on ANN exhibited the smallest mean absolute percentage error, while the SVM-based model had the shortest computation time. Aly [32] improved the accuracy of power prediction in wind power grid integration using a high-precision hybrid deep learning cluster model. Wen et al. [33] developed a deep recurrent neural network model with long short-term memory units to predict the combined power load and photovoltaic generation in China’s power grid. Guo et al. [34] explored electricity usage patterns based on feature importance, proposing a probability density forecasting method using deep learning, quantile regression, and kernel density estimation to address power dispatch issues in China’s electricity grid. Pillai et al. [35] generated a composite benchmark power load curve using Artificial Neural Networks based on publicly available power load and weather data. Guo et al. [36] designed a power consumption forecasting method for Thailand using a “causal network” based on vector error correction models and an adaptive screening method. Jin et al. [37] assessed China’s dynamic energy market as well as load and price forecasts using “Self-Organizing Maps (SOM)”. Zhang and Wang [38] proposed a “Nonlinear Autoregressive Model with Exogenous Inputs-Neural Network (NARX-NN)” to forecast short-term load time series with multiple seasonal patterns in China. Hamedmoghadam et al. [39] used ANN and deep neural networks to predict long-term electricity demand in Australia, noting that deep neural networks performed better than ANN. T. Ashfaq and Javaid [40] used K-nearest neighbors (KNN) for load forecasting, noting that this method is sensitive to data scaling and could pose high computational costs and data storage issues. Borthakur and Goswami [41] presented an implementation using Naive Bayesian Classification for load forecasting, which may face zero-frequency problems depending on the predictive variables.

Currently, neural networks and support vector machines excel in short-term load forecasting but struggle with mid- to long-term forecasts due to difficulties in incorporating macroeconomic policies. Effective mid- to long-term forecasting, essential for electricity system planning, requires considering diverse and long-term factors. Additionally, some machine learning algorithms (like K-nearest neighbors and Naive Bayes) have limitations such as high computational costs and sensitivity to data scaling, which may pose challenges when dealing with large-scale, diverse electricity data.

3. The Framework of the Electricity Consumption Forecasting Method Proposed in This Paper

While extensive modeling has addressed electricity demand influenced by economic and meteorological factors, previous studies have largely overlooked their subtle interplay with evolving energy policies, especially in rapidly changing regulatory environments. Against the strategic backdrop of achieving carbon peaks by 2030 and carbon neutrality by 2060, the impact of energy policy indicators on China’s industrial electricity demand has become increasingly significant. Yet, current research often still relies on traditional factors, which are inadequate for accurately predicting electricity demand in new carbon neutrality scenarios and do not effectively address the discontinuity and non-linearity of policy factor releases. This paper innovatively combines energy policy indicators with traditional meteorological factors by embedding policy analysis directly into the demand forecasting framework. This model predicts how legislative changes, regulatory updates, and government initiatives related to energy will affect industrial electricity consumption. This includes anticipated enhancements in renewable energy adoption, improvements in energy efficiency, and adjustments in subsidies or penalties related to carbon emissions. Moreover, the integration of Shared Socioeconomic Pathways (SSPs) and Representative Concentration Pathways (RCPs) addresses the complexities of future energy scenarios. SSPs outline potential societal developments that influence energy use and policy, ranging from sustainable growth (SSP1) to high dependency on fossil fuels (SSP5). RCPs quantify the projected radiative forcing by 2100, covering a range from low (RCP2.6) to high (RCP8.5) impacts. This detailed approach enables more robust predictions of industrial electricity demand across various future scenarios.

This paper employs correlation analysis, multivariate regression analysis, and establishes a hierarchical analysis model to forecast industrial electricity demand in Jiangsu Province under policy macro-control and meteorological changes. The key steps of the proposed method are illustrated in Figure 1, starting with the extraction of impact indicators from policy and meteorological data, analyzing their correlation and collinearity, and selecting the key indicators. Subsequently, a multivariate regression model for electricity demand is established. Finally, by calculating the adjustment rates and regression coefficients for each hierarchical indicator, the industrial electricity demand in Jiangsu Province under different SSPs and RCPs is predicted, providing critical decision support for policy-making and energy planning. Despite the extensive modeling of electricity demand influenced by economic and meteorological factors, prior research has largely overlooked the nuanced interplay between these factors and evolving energy policies, particularly in a rapidly changing regulatory environment. This study addresses this gap by incorporating comprehensive policy analysis into the forecasting model, thereby providing a more robust framework that aligns with China’s dynamic policy landscape aimed at carbon neutrality.

4. Structural and Correlation Analysis of Load Forecasting Influencing Factors in the Chinese Energy Market

4.1. Extraction of Influencing Factors

4.1.1. Meteorological Indicators

This study focuses on Jiangsu Province, situated in China’s central coastal region (Figure 2). This area, characterized by a subtropical monsoon climate, is one of China’s wealthiest regions. Its industrial sector, including heavy industries like steel, chemicals, and manufacturing, has a high electricity demand. With the introduction of China’s carbon neutrality goals, industrial enterprises in Jiangsu face stricter carbon emission limits and management. Therefore, forecasting the industrial electricity demand in Jiangsu Province is crucial for ensuring the reliability of the power supply and achieving the carbon neutrality objectives.

Meteorological variables like temperature, rainfall, potential evaporation, wind speed, relative humidity, and sunshine duration significantly impact electricity demand [42]. The complex interactions among these variables make it difficult to describe their relationships. Generally, the more input variables selected, the greater the predictive advantage. However, too many variables can lead to disadvantages, such as possibly causing the model to overfit the training data, thereby reducing the model’s generalization ability [43].

In addition, degree days are commonly used to assess the impact of climate change on electricity demand. Degree days are a meteorological index used to evaluate energy demand and play an important role in short-term load forecasting [44]. Degree days represent the deviation of the temperature from a balance point. The balance point temperature is a threshold temperature frequently used in the calculation of degree days. At this point, the per capita electricity demand is minimal. Since this point is neither too cold nor too hot, the demand for electricity for cooling and heating is the least. According to studies by Ge et al. [45] and Fan et al. [46], 18 degrees Celsius is generally used as the balance point temperature, a value that has been employed in research in East China. However, due to different geographical settings, this value often varies in different regions. In the proposed study, the balance point temperature for the Jiangsu area in China was investigated based on per capita demand and temperature. Ideally, if demand is plotted against temperature on a scatter plot, a “V” shaped trend curve should be observed, as shown in Figure 3.

The best fit line can be obtained by drawing a bisector that passes through the data points, minimizing the sum of squared errors between this line and the data points. The left side of the “V” shaped curve has a negative slope, corresponding to heating demand, while the right side has a positive slope, corresponding to cooling demand. The intersection of these two lines represents the theoretical balance point temperature. This paper adopts the same method to study the balance point temperature in the Jiangsu region of China. As electricity users respond differently to heat demands generated by temperature changes, degree days are represented as two separate variables, namely cooling degree days (CDD) and heating degree days (HDD) [47]. These variables should be treated separately in the regression analysis [48], assigning individual coefficients to each variable. CDD represents the degree of cooling required due to temperatures above the reference level, while HDD represents the degree of heating required due to temperatures below the reference level. Formulas (1) and (2) are used to calculate CDD and HDD, respectively [47]:

$$CD{D}_i=\left\{\begin{array}{cc}(T_i-T_b)\hfill & \mathrm{if}(T_i>T_b)\hfill \\ 0\hfill & \mathrm{if}(T_i\le T_b)\hfill \end{array}\right.,CDD=\sum _{i=1}^n\sum _{i=1}^{m}CD{D}_i$$

(1)

$$HD{D}_i=\left\{\begin{array}{cc}(T_b-T_i)\hfill & \mathrm{if}(T_i<T_b)\hfill \\ 0\hfill & \mathrm{if}(T_i\ge T_b)\hfill \end{array}\right.,HDD=\sum _{i=1}^n\sum _{i=1}^{m}HD{D}_i$$

(2)

where n is the number of days in the year, m is the number of temperature data samples considered per day, $T_i$ is the temperature at the m hour of the n day, and $T_b$ is the balance point temperature.

Degree days are commonly used to study residential electricity consumption, but are seldom used to predict industrial electricity consumption. However, degree days have unique advantages in processing temperature data. Compared to direct temperature measurements, degree days provide a cumulative and more sensitive indicator of temperature fluctuations. Direct temperature measurements, while capable of providing data at specific times or as averages, often fail to fully account for the cumulative impact of temperature changes on energy consumption. Degree days, by calculating the cumulative difference above or below a certain baseline temperature, can more comprehensively reveal the actual impact of temperature changes on energy consumption. Furthermore, considering that energy policies also have a cumulative effect on long-term industrial electricity consumption, degree days are particularly suitable for predicting industrial electricity consumption. They effectively integrate energy policy factors to provide more accurate forecasts of industrial electricity consumption, and

Table 1. Regression analysis of the impact of degree days on electricity consumption in the residential and industrial Sectors.

Sectors	National Level		Jiangsu Province Level
	CDD	HDD	CDD	HDD
Household sector	0.038	0.020	0.060	0.052
Industrial sector	0.010	0.004	0.017	0.011

displays the sensitivity changes of monthly electricity consumption in the residential and industrial sectors to degree days.

The table above shows that both CDD and HDD coefficients for the industrial sector are lower than those for the residential sector, indicating that the secondary sector’s sensitivity to temperature changes is less than that of the residential sector. However, its total electricity consumption, accounting for a significant 73.4% (sample average level) of total societal electricity use, should not be overlooked. Moreover, the CDD coefficient is about twice that of the HDD, suggesting that the increase in electricity consumption in the industrial sector due to temperature effects is primarily for coping with hot weather, including the enhanced use of air conditioning systems, cooling of industrial equipment, and utilizing condensate water for cooling.

The temperature data utilized in this analysis was sourced from the Jiangsu Provincial Meteorological Bureau [49]. To calculate future adjustment rates for CDD and HDD, it is assumed that all baseline temperature values will shift upwards to equal the future expected temperature change. The sample data used in this paper include 24 data points for daily temperatures, assuming a baseline year of 365 days, which results in 8760 temperature data points for the baseline year. According to the assumption, all 8760 points in the baseline year will shift up by $\Delta T$. If $T_b$ is the balance point temperature as shown in Figure 4, there could be five scenarios. First, a temperature point $T1$ that will fall between $T_b$ and ($T_b-\Delta T$ ) in the future, thus not affecting CDD but causing a reduction in HDD by $\Delta T/24$; a temperature point $T_2=(T_b-\Delta T)$ will overlap with $T_b$, not affecting CDD but causing a reduction in HDD by $\Delta T/24$; a temperature point $T_3$ located between $T_b$ and $T_b-\Delta T$ will cross the $T_b$ line, and the temperature difference compared to the reference temperature $T_b$ is $(T_1+\Delta T-T_b)$, thus increasing CDD by $(T_1+\Delta T-T_b)/24$; and $T_4$ and $T_5$ will update the CDD value by $\Delta T/24$ but will not change HDD. Thus, knowing the temperature change $\Delta T$ over the coming months can help estimate future adjustment rates for CDD and HDD.

4.1.2. Energy Policy Indicators

This section first reviews the relevant policies issued by the National Development and Reform Commission [50] and the Jiangsu Provincial Development and Reform Commission [51] from 2015 to 2024, extracting policy factors that affect the electricity demand in Jiangsu Province’s energy market.

Table 2. Relevant policies issued at the provincial level in Jiangsu.

Reference Policies	Policy Objectives	Extraction of Policy Impact Factors
Jiangsu Province’s Technology Support Carbon Peak and Carbon Neutrality Implementation Plan	A special fund for technological innovation for carbon peak and carbon neutrality has been set up. This fund will be used to support low-carbon technology innovation projects and has compiled this year’s project guide for the carbon peak and carbon neutrality technological innovation special fund.	R&D investment in low-carbon technology as a proportion of GDP $R_{LC,t}$
The 14th Five-Year Plan for the National Economic and Social Development of Jiangsu Province and the Long-Range Objectives Through the Year 2035	Jiangsu Province will increase investments in energy-saving retrofits for industrial facilities, through technological upgrades and equipment updates, to improve energy efficiency in industrial production processes and reduce energy consumption and carbon emissions. Key support will be given to energy-saving retrofits for high-energy-consuming industries and key enterprises, promoting green transformation and upgrading of the industry.	Energy-saving retrofit investments in facilities as a proportion of GDP $R_{ES,t}$

and

Table 3. Relevant policies issued at the national level.

Reference Policies	Policy Objectives	Extraction of Policy Impact Factors
“14th Five-Year” Plan for Renewable Energy Development; (2021) No. 1445	Currently, the national electricity consumption accounts for about 27% of the terminal energy consumption, which is higher than the global average. It is expected that by 2025, the proportion of electricity in terminal energy consumption will increase to 31.2%. By 2035, renewable energy is projected to account for more than 50% of the incremental consumption of primary energy (i.e., electricity substitution rate).	Electrification rate $E_t$ Electricity substitution rate $R{E}_{sub,t}$
	Setting clear targets for energy consumption intensity is crucial. China’s energy intensity continues to decrease but still remains 1.5 times higher than the global average. At the same time, a target has been set to reduce energy consumption per unit of GDP by 13.5% by 2025.	Gross Domestic Product $P_t$ Energy consumption intensity $E{C}_t$
Modern Energy System “14th Five-Year” Plan; (2022) No. 210	Demand for electricity continues to grow rigidly, and energy consumption intensity continues to decline. By 2025, the proportion of non-fossil fuel power generation is expected to reach about 39%, and the proportion of non-fossil energy consumption will increase significantly by 2030, with renewable energy generation becoming the main power source. New power system construction achieves substantial results.	Renewable energy penetration rate $R_{Dp,t}$ Energy consumption intensity $E{C}_t$
	Promote the power grid’s proactive adaptation to large-scale centralized new energy sources and the widespread development of distributed energy; support the rational allocation of energy storage systems for distributed new energy; enhance the nearby development and utilization of energy resources, actively develop distributed energy, and encourage priority local consumption of wind and solar power; and improve new mechanisms for the development of distributed power sources and promote fair access to the grid.	Proportion of distributed power installed capacity $I{R}_{Dic,t}$ Distributed power installation subsidy $E{S}_{Dis,t}$
Opinions on Promoting Future Industrial Innovation Development by the Ministry of Industry and Information Technology and other seven departments; (2024) No. 12	According to the National Development and Reform Commission, the secondary industry consumes the most electricity per unit of GDP among the three major industries, and it usually shows a strong positive correlation with the overall electricity consumption intensity. By the end of the “14th Five-Year Plan”, the GDP proportion of the secondary industry is expected to drop to about 35.5%.	Gross Domestic Product $P_t$ Proportion of secondary industry in GDP $R_{S,t}$
Implementation Plan for the Green Low-Carbon Advanced Technology Demonstration Project; (2023) No. 1093	Implementing “Green Hydrogen Carbon Reduction Demonstration Projects”, including the research and development, manufacturing, and scaled demonstration application of hydrogen fuel cells.	Proportion of fuel cell installations $I{R}_{Fc,t}$ Fuel cell installation subsidy $E{S}_{FC,t}$
“14th Five-Year” Plan for New Energy Storage Development; (2022) No. 209	By 2025, the installed capacity of new energy storage is expected to exceed 30,000 megawatts. And for user-side energy storage over 1000 kWh, after commissioning, a one-time subsidy of 200 yuan/kWh will be provided by provincial finance.	Proportion of energy storage installed capacity $I{R}_{ES,t}$ Energy Storage Installation Subsidy $E{S}_{ES,t}$
Notification on Further Promoting New Energy Storage to Participate in Electricity Markets and Dispatch Applications; (2022) No. 475	Research to establish a new energy storage pricing mechanism, adapted to the needs of the new power system.	New energy storage prices $E{S}_{price,t}$
Guidance on Promoting the Integrated Development of Power Source-Grid-Load Storage and Multi-Energy Complementarity; (2021) No. 280	In areas with high industrial load and favorable new energy conditions, support the development and construction of distributed power sources and their nearby connection and consumption, in conjunction with the construction of incremental distribution networks and other initiatives, to develop integrated green power supply parks that combine source, grid, load, and storage.	Investment in power distribution network construction as a proportion of GDP $R_{DN,t}$
Further Improvement of Pricing Mechanisms; (2018) No. 943	Reduce the subsidy standards for new distributed photovoltaic power generation, encouraging industrial and commercial users to adopt the "self-generation and surplus-to-grid" model.	Self generating electricity $Q_{self,t}$

list the selected indicators related to energy policies.

4.2. Hierarchical Division of Influencing Indicators

In practical situations, there is sometimes a discrepancy between the official data and observed data, which may be related to data instability and the time lag in the release of government data. For instance, in the parameter of self-generated electricity ($Q_{self,t}$), the value given in 2021 was the same as that in 2020, but by 2022, it had increased by 19.5% compared to 2020 and 2021. Specifically, some indicators might remain relatively stable for several consecutive years before experiencing a significant increase or decrease in a particular year. This phenomenon may stem from inconsistent data update frequencies. Government agencies or data providers might release data irregularly, resulting in significant updates in some years. Additionally, changes in data collection methods or definitions can cause data discontinuity and nonlinearity.

To address issues such as policy effect lag, lack of historical data support, and data discontinuity in influencing factors, as well as to determine the quantitative relationship between long-term electricity demand, policy objectives, and meteorological factors, we organize the hierarchy of policy influencing factors by tracing the paths to policy objectives and calculating the relationships among influencing factors, following the classification of meteorological influences by Ahmed et al. [13].

(1) Primary Indicators

Primary indicators are factors closely related to electricity demand. Although their data may not be frequently updated, their relationship with electricity demand is more pronounced. The electrification rate is chosen as an indicator of the proportion of electricity in end-use energy consumption, reflecting the trend of energy consumption shifting towards electricity, directly influenced by energy conservation and emission reduction policies. Gross Domestic Product (GDP) and energy consumption intensity are selected to assess energy efficiency in economic activities. Energy consumption intensity, focusing on the amount of energy used per unit of GDP, is a key indicator for measuring the effectiveness of policies like energy conservation and emission reduction. The amount of self-generated electricity reflects the acceptance of self-generation systems by consumers and their capacity to interact with the grid, directly related to the effectiveness of distributed energy and demand response policies.

(2) Secondary Indicators

Secondary indicators, as an intermediate layer, link primary and tertiary indicators. They are reasonably feasible in terms of data updates and have a certain relationship with electricity demand. For instance, the electricity substitution rate describes the proportion of electricity substituting other energy sources (such as coal, oil), reflecting the depth of the energy structure adjustment driven by policies. The renewable energy penetration rate shows the share of renewable energy in the total energy supply. The proportion of the secondary industry in GDP reflects the impact of industrial upgrading policies and measures to enhance energy consumption intensity. Proportions such as distributed power installation, fuel cell installation, and power storage installation capacity serve as manifestations of self-generated electricity in policy objectives. According to research by Ahmed, temperature is the most significant factor affecting electricity consumption, impacting evaporation, humidity, and atmospheric pressure. Therefore, correlations exist between temperature and these three weather variables.

Consequently, the electricity substitution rate, renewable energy penetration rate, proportion of the secondary industry in GDP, proportion of distributed power installed capacity, proportion of fuel cell installations, proportion of energy storage installed capacity, humidity, and atmospheric pressure are selected as secondary micro indicators.

(3) Underlying/Tertiary Indicators

Tertiary indicators are more specific and operational factors, with relatively frequent data updates, and they significantly capture changes in electricity demand. The price of new energy storage, a key factor influencing the proliferation of new energy storage devices, profoundly affects the distributed power installation proportion, fuel cell installation proportion, and power storage installation capacity. The proportion of the facility’s energy-saving retrofit investment in GDP, the proportion of the power distribution network construction investment in GDP, and the proportion of low-carbon technology R&D investment in GDP are major measures to promote low-carbon transition and industrial structure optimization and upgrading. Power storage installation subsidies, distributed power installation subsidies, and fuel cell installation subsidies directly relate to enhancing the renewable energy coverage rate and are specific manifestations of government policies to promote energy transformation. According to Ahmed’s research, humidity affects rainfall, and atmospheric pressure affects wind speed.

Therefore, new energy storage prices, energy-saving retrofit investments in facilities as a proportion of GDP, investments in power distribution network construction as a proportion of GDP, R&D investments in low-carbon technology as a proportion of GDP, energy storage installation subsidies, distributed power installation subsidies, fuel cell installation subsidies, rainfall, and sunshine duration are selected as underlying indicators.

This multilevel analytical framework helps us to understand changes in policy indicators and examine data anomalies from a more comprehensive perspective, reducing fluctuations caused by single factors.

4.3. Historical Patterns of Total Electricity Demand and Selected Key Indicators

Historical monthly electricity demand was plotted against selected primary indicators. If we found a linear relationship between demand and any primary indicator, a simple linear regression would be adequate, making multiple regression unnecessary. Using historical data from 2015 to 2020, the Pearson correlation coefficients [52] between the primary indicators and electricity demand were calculated. To calculate the Pearson correlation coefficients, we ranked these indices based on their impact on electricity demand, as detailed in Section 4.2. We used these rankings to assess the interconnectivity of the indices, focusing on relationships beyond mere numerical values. This helps in understanding how these factors interact with each other and how they evolve over time without overly focusing on their specific numerical values.

The relationships between electricity demand and various primary indicators are depicted in Figure 5. Using bivariate correlation analysis, we highlighted the relationships between different variables, noting that none displayed a linear relationship with demand. However, cooling degree days (CDD), heating degree days (HDD), electrification rates, energy consumption intensity, and gross product value seem to have a positive correlation with electricity demand. On the other hand, self-generated electricity appears to have a negative correlation with electricity demand.

Table 4. Correlation coefficients between electricity demand and primary indicators.

Indicators	Demand	Electrification Rate	Energy Consumption Intensity	Gross Product Value	Self-Generation Electricity	HDD	CDD
Demand	1	0.991	0.985	0.886	−0.928	0.568	0.456
Electrification rate	0.991	1	−0.156	0.189	−0.136	−0.041	0.037
Energy consumption intensity	0.985	−0.156	1	−0.367	−0.018	−0.171	0.231
Gross Domestic Product	0.886	0.189	−0.367	1	0.345	0.132	0.018
Self-generated electricity	−0.928	−0.136	−0.018	0.345	1	−0.023	−0.013
HDD	0.568	−0.041	−0.171	0.132	−0.023	1	−0.349
CDD	0.456	0.037	0.231	0.018	−0.013	−0.349	1

presents the specific Pearson correlation coefficient matrix. If the correlation coefficient between two factors is close to 1, it indicates a strong positive correlation between them; if the correlation coefficient is close to −1, it indicates a strong negative correlation, meaning one factor increases as the other decreases. If the correlation coefficient is close to 0, it suggests a weak relationship, possibly with no significant linear correlation. Based on the results of the correlation coefficients, those factors with strong linear correlations are removed to avoid the issue of multicollinearity. From Table 4, it can be seen that multiple variables are needed to establish a relationship with electricity demand. Multiple regression analysis will help in connecting only those variables that are significantly related to demand and in excluding non-significant variables. It will also assist in quantifying the impact of individual variables on the demand pattern.

In addition to the correlation coefficients between the main variables and power demand, the pairwise correlation coefficients among the variables are less than 0.4, indicating that the correlations among the variables are not high. Therefore, multicollinearity is avoided in the model. Furthermore, multivariate regression analysis needs to be conducted on power demand, followed by the construction of a multivariate linear regression model. This study employs logarithmic variables, and the model is defined by the following equation:

$$ln\left(Q_t\right)=c+{\beta }_1\times ln\left(E_t\right)+{\beta }_2\times ln\left(E{C}_t\right)+{\beta }_3\times ln\left(P_t\right)+{\beta }_4\times ln\left({\mathrm{HDD}}_t\right)+{\beta }_5\times ln\left({\mathrm{CDD}}_t\right)-{\beta }_6\times ln\left(Q_{\mathrm{self},t}\right)+{ϵ}_t$$

(3)

Here, $Q_t$ represents the power demand of the industrial sector in year t, $P_t$ represents the gross output value, $E_t$ represents the electrification rate, $E{C}_t$ represents the energy consumption intensity, $HD{D}_t$ represents heating degree days, $CD{D}_t$ represents cooling degree days, $Q_{self,t}$ represents self-generated electricity, c is the intercept, and $ϵ$ is the random error term in the model, reflecting the impact of other factors not considered.

Subsequently, a difference equation that reflects the changes over time points in the dynamic system is constructed as follows [53]:

$$X_t=\left(X_{t-1}+\int U_{X,t}dt\right)$$

(4)

where $X_t$ is the state of the system at time t, and $X_{t-1}$ is the state of the system at the previous time $t-1$. The integral represents the cumulative effect of changes in the state X over time, where $U_{X,t}$ is the change in X at time t. This integral term can be considered as the cumulative effect of the system state in the time series.

According to Equation (3), introducing the difference equations for each indicator, the calculation formula for power demand can be further expressed as:

$$\begin{array}{c}ln\left(Q_t\right)=c+{\beta }_1\times ln\left(E_{t-1}+\int U_{E,t}dt\right)+{\beta }_2\times ln\left(E{C}_{t-1}+\int U_{EC,t}dt\right)+{\beta }_3\times ln\left(P_{t-1}+\int U_{P,t}dt\right)\hfill \\ \hfill +{\beta }_4\times ln({\mathrm{HDD}}_t\times f\left({HDD}_t\right))+{\beta }_5\times ln({\mathrm{CDD}}_t\times f\left({CDD}_t\right))-{\beta }_6\times ln\left(Q_{\mathrm{self},t-1}+\int U_{Q_{\mathrm{self},t}}dt\right)+{ϵ}_t\end{array}$$

(5)

where,

$$\begin{array}{c}\hfill \left\{\begin{array}{c}{U}_{E,t}=E_{t-1}\times f\left(E_t\right)\hfill \\ U_{P,t}=P_{t-1}\times f\left(P_t\right)\hfill \\ U_{EC,t}=E{C}_{t-1}\times f\left(E{C}_t\right)\hfill \\ U_{Q_{\mathrm{self}},t}=Q_{\mathrm{self},t-1}\times f\left(Q_{\mathrm{self},t}\right)\hfill \end{array}\right.\end{array}$$

(6)

For policy and meteorological factors, their modes of influence might differ; policy factors have a cumulative effect, meaning their impact accumulates over time, while meteorological factors have immediacy and transience, with their impact occurring immediately as meteorological conditions change. Therefore, meteorological factors are considered instantaneous variables and are not subjected to cumulative treatment. Here, $U_{E_t}$, $U_{P_t}$, $U_{E{C}_t}$, and $U_{Q_{\mathrm{self}},t}$ represent the adjustment amounts for the primary indicators in year t, while $f\left(\right)$ represents the adjustment rate of each factor, which will be derived through the established secondary and underlying regression equations.

4.4. Analysis of Correlation of Influencing Indicators

To ensure the effectiveness of the influencing indicators and the electricity demand forecasting model, considering that past studies often did not incorporate both policy factors and meteorological factors into the power demand forecasting model, we need to conduct a correlation analysis experiment to further explore the relationships between these indicators. The results of the correlation coefficients between the secondary indicators and primary indicators are shown in

Table 5. Correlation coefficients between secondary indicators and primary indicators.

Indicators	Energy Consumption Intensity	Electrification Rate	Gross Product Value	Self-Generated Electricity	HDD	CDD
Proportion of secondary industry in GDP	0.928	0.031	0.997	−0.134	0.001	0.146
Proportion of distributed power installed capacity	−0.948	0.977	−0.107	0.998	0.023	0.012
Proportion of fuel cell installations	−0.943	0.978	−0.102	0.997	0.012	0.019
Proportion of energy storage installed capacity	−0.952	0.977	−0.103	0.999	0.002	0.023
Electricity substitution rate	0.189	0.966	−0.019	−0.178	−0.001	0.008
Renewable energy penetration rate	−0.587	−0.018	−0.689	0.821	0.000	0.018
Evaporation	0.012	−0.003	−0.002	0.005	−0.314	0.865
Wind speed	0.108	0.015	0.012	0.119	0.043	0.367
Humidity	−0.004	−0.013	−0.005	−0.015	−0.062	−0.716

, where all secondary indicators are significantly correlated with at least one primary indicator.

The results of the correlation coefficients between the underlying indicators and the selected secondary indicators are shown in

Table 6. Correlation coefficients between underlying indicators and selected secondary indicators.

Indicators	Proportion of Secondary Industry in GDP	Proportion of Distributed Power Installed Capacity	Proportion of Energy Storage Installed Capacity	Electricity Substitution Rate	Renewable Energy Penetration Rate	Evaporation	Humidity	Wind Speed
New energy storage prices	−0.351	−0.167	−0.912	−0.915	0.562	0.008	0.000	−0.198
Energy-saving retrofit investments in facilities as a proportion of GDP	−0.198	0.878	0.891	0.328	0.476	0.000	0.000	0.000
R&D investments in low-carbon technology as a proportion of GDP	−0.165	0.994	0.981	0.234	0.196	0.000	0.000	0.000
Investments in power distribution network construction as a proportion of GDP	−0.183	0.817	−0.132	0.179	0.798	0.000	0.000	0.000
Distributed power installation subsidy	0.831	0.992	0.432	0.391	0.003	0.000	0.000	0.000
Energy storage installation subsidy	−0.238	0.371	0.943	0.391	0.012	0.000	0.000	0.000
Fuel cell installation subsidy	−0.241	0.373	0.483	0.321	0.006	0.000	0.000	0.000
Rainfall	−0.004	−0.019	−0.023	0.003	−0.023	−0.789	0.746	0.319
Sunshine duration	−0.007	0.002	0.003	0.012	0.389	0.871	−0.716	−0.031

, where all underlying indicators are significantly correlated with at least one secondary indicator.

In multivariate regression analysis, in addition to calculating correlations, it is also necessary to compute collinearity. Indicators at the same level should not exhibit collinearity, as it could increase the variance of the predictor variables, affecting the accuracy of the model parameters. To quantitatively assess this collinearity, we used the variance inflation factor (VIF) to measure the degree of multicollinearity for each variable. The VIF of each variable is calculated as follows:

$$\begin{array}{c}\hfill {\mathrm{VIF}}_{x_i}={\displaystyle \frac{1}{1-R_{x_i}^2}}\end{array}$$

(7)

where $R_{x_i}^2$ is the coefficient of determination for the regression model, where variable $x_i$ is treated as the dependent variable and all other variables as independent variables.

Through the VIF analysis of the indicators, we found that the VIF values for the fuel cell installation proportion and energy storage installed capacity proportion, as well as for fuel cell installation subsidies and energy storage installation subsidies, are significantly higher than 5. This indicates a strong collinearity between these pairs of variables.

Based on these findings, we decided to exclude the fuel cell installation proportion and fuel cell installation subsidy from the regression model. This decision is based on reducing collinearity in the model to improve the accuracy of the model estimates. The filtered hierarchy of electricity demand indicators is shown in Figure 6.

4.5. Electricity Demand Regression Model

The hierarchical structure of the influencing factors can be observed from Figure 6, where upper-level influencing factors are influenced by multiple underlying factors. Based on historical data, multivariate regression analysis can eliminate multicollinearity among influencing factors and quantify the impact weights of different influencing factors, thus obtaining a quantitative predictive equation with explanatory power for reality. The quantitative relationship model is as follows:

$$\begin{array}{c}\hfill \begin{array}{cc}& ln\left[f\left(Y_{i,t}\right)\right]=\sum _i{\lambda }_{i}ln\left[f\left(X_{i,t}\right)\right]+{\delta }_i\hfill \end{array}\end{array}$$

(8)

where $Y_{i,t}$ represents the initial data of the i-th indicator in year t, $X_{i,t}$ represents the initial data of the i-th indicator of the next-level indicator in year t, ${\lambda }_i$ represents the regression coefficient of the i-th indicator, and ${\delta }_i$ represents the constant term of the multivariate regression equation. Here, $f\left(X_{i,t}\right)$ represents the calculation function of the rate of change of $X_{i,t}$ in year t, with the formula as follows:

$$\begin{array}{c}\hfill \begin{array}{c}\hfill f\left(X_{i,t}\right)={\displaystyle \frac{X_{i,t}-X_{i,t-1}}{X_{i,t-1}}}\end{array}\end{array}$$

(9)

Taking the electrification rate regression equation as an example, based on the indicator correlation screening in Section 4.4, it is found that the primary indicator, electrification rate, is correlated with the secondary indicators electricity substitution rate, proportion of energy storage installed capacity, and proportion of distributed power installed capacity. Therefore, the regression equation for the electrification rate can be expressed as follows:

$$\begin{array}{c}\hfill \begin{array}{c}\hfill ln\left[f\left(E_t\right)\right]=x_{1}ln\left[f\left(R_{\mathrm{Esub},t}\right)\right]+x_{2}ln\left[f\left(I{R}_{\mathrm{Dic},\mathrm{t}}\right)\right]+x_{3}ln\left[f\left(I{R}_{\mathrm{ES},\mathrm{t}}\right)\right]+{\delta }_1\end{array}\end{array}$$

(10)

where $E_t$ represents the electrification rate of the industrial sector in year t, $R_{Esub,t}$ represents the electricity substitution rate of the industrial sector in year t, $I{R}_{Dic,t}$ represents the proportion of distributed power installed capacity in the industrial sector in year t, and $I{R}_{ES,t}$ represents the proportion of energy storage installed capacity in the industrial sector in year t.

5. Experiments and Results

5.1. Linear Regression Results

During the regression process, we used a linear adaptive random model for stepwise regression, and the results are shown in

Table 7. Stepwise multivariate regression results for primary indicators.

	(1)	(2)	(3)	(4)
Variables	$Ln\left[f\left(Q_t\right)\right]$	$Ln\left[f\left(Q_t\right)\right]$	$Ln\left[f\left(Q_t\right)\right]$	$Ln\left[f\left(Q_t\right)\right]$
$Ln\left[f\left(E{C}_t\right)\right]$	0.321 **	0.321 ***	0.315 ***	0.314 ***
$Ln\left[f\left(E_t\right)\right]$	0.159 **	0.154 **	0.164 **	0.162 ***
$Ln\left[f\left(P_t\right)\right]$	0.783 **	0.764 ***	0.753 ***	0.749 ***
$Ln\left[f\left(Q_{self,t}\right)\right]$		−1.123 ***	−1.131 ***	−1.145 ***
$Ln\left[f\left(HD{D}_t\right)\right]$			0.045 ***	0.042 ***
$Ln\left[f\left(CD{D}_t\right)\right]$				0.095 ***
Constant	−1.313 ***	−1.243 ***	−1.876 ***	−2.134 ***
R-squared	0.914	0.934	0.956	0.984

*** p < 0.01, ** p < 0.05.

. From the regression results, it can be observed that as control variables are gradually added, the fitting coefficients of each variable are statistically significant at least at the 5% significance level. Furthermore, the stability of regression coefficients during the inclusion of new variables suggests robustness in our empirical model. The model’s goodness of fit, at 98.4% as shown in column (4) of the results, effectively captures most electricity demand fluctuations in Jiangsu Province. The regression results show that increases in HDD, CDD, electrification rate, energy consumption intensity, and gross output value lead to an increase in industrial electricity demand in Jiangsu Province, while an increase in self-generated electricity leads to a decrease in industrial electricity demand. This is consistent with the findings in Section 4.3.

Subsequently, a multivariate regression was conducted using the primary and secondary indicators identified in Section 4.4, and the results are presented in

Table 8. Multivariate regression results between primary indicators and secondary indicators.

Indicators	$\mathit{Ln}\left[\mathit{f}\left({\mathit{EC}}_{\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{E}}_{\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{P}}_{\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{Q}}_{\mathit{self},\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{HDD}}_{\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{CDD}}_{\mathit{t}}\right)\right]$
$Ln\left[f\left(R{S}_t\right)\right]$	0.718 ***	-	0.246 ***	-	-	-
$Ln\left[f\left(I{R}_{Dic,t}\right)\right]$	0.027 ***	0.008 ***	-	0.323 ***	-	-
$Ln\left[f\left(I{R}_{Es,t}\right)\right]$	0.032 **	0.013 ***	-	0.293 **	-	-
$Ln\left[f\left(R{E}_{sub,t}\right)\right]$	-	3.123 ***	-	-	-	-
$Ln\left[f\left(R_{Dp,t}\right)\right]$	-	-	-	0.017 **	-	-
$Ln\left[f\left(Evaporatio{n}_t\right)\right]$	-	-	-	-	0.039 ***	0.123 ***
$Ln\left[f\left(WindSpee{d}_t\right)\right]$	-	-	-	-	0.079 ***	0.023 **
$Ln\left[f\left(Humidit{y}_t\right)\right]$	-	-	-	-	0.021 ***	0.031 ***
Constant	0.611 ***	0.015 **	1.321 ***	−1.812 ***
R-squared	0.932	0.933	0.936	0.932	0.913	0.948

*** p < 0.01, ** p < 0.05.

.

Similarly, a multivariate regression was conducted between the secondary indicators and the underlying indicators, with the results shown in

Table 9. Multivariate regression results between secondary indicators and underlying indicators.

Indicators	$\mathit{Ln}\left[\mathit{f}\left({\mathit{RS}}_{\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{IR}}_{\mathit{Dic},\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{IR}}_{\mathit{Es},\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{RE}}_{\mathit{sub},\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}\left({\mathit{R}}_{\mathit{Dp},\mathit{t}}\right)\right]$	$\mathit{Ln}\left[\mathit{f}(\mathit{Evapor}-{\mathit{ation}}_{\mathit{t}})\right]$	$\mathit{Ln}\left[\mathit{f}(\mathit{Humid}-{\mathit{ity}}_{\mathit{t}})\right]$	$\mathit{Ln}\left[\mathit{f}(\mathit{Wind}-{\mathit{Speed}}_{\mathit{t}})\right]$
$Ln\left[f\left(E{S}_{price,t}\right)\right]$	-	-	−0.323 **	−0.412 **	-	-	-	-
$Ln\left[f\left(R_{ES,t}\right)\right]$	-	0.134 **	0.221 **	-	-	-	-	-
$Ln\left[f\left(R_{LC,t}\right)\right]$	-	0.353 ***	0.198 ***	-	-	-	-	-
$Ln\left[f\left(R_{DN,t}\right)\right]$	0.271 **	0.532 **	-	-	0.443 ***	-	-	-
$Ln\left[f\left(E{S}_{dis,t}\right)\right]$	-	0.561 **	-	-	-	-	-	-
$Ln\left[f\left(E{S}_{ES,t}\right)\right]$	-	-	0.451 **	-	-	-	-	-
$Ln\left[f\left(Rainfal{l}_t\right)\right]$	-	-	-	-	-	−0.032 **	0.121 ***	0.093 ***
$Ln\left[f(Sunshine-duratio{n}_t)\right]$	-	-	-	-	-	0.445 ***	−0.154 ***	−0.109 ***
Constant	0.932 **	1.365 ***	−0.134 **	−0.634 ***	−0.113 **	−0.541 ***	0.321 **	−0.127 ***
R-squared	0.923	0.912	0.936	0.941	0.961	0.913	0.932	0.981

*** p < 0.01, ** p < 0.05.

.

5.2. Scenario Hypothesis

The empirical study of this research is based on panel data from 2011 to 2021, acknowledging that climate change is a long-term process. To reflect the future impact of climate change on China’s electricity demand, this study extends the empirical findings by forecasting the changes in electricity demand in China due to alterations in temperature and precipitation under three socioeconomic development scenarios (relative to the 2011 baseline) [54]. These scenarios include various SSP–RCP combinations proposed under the CMIP6 framework [55], representing different Shared Socioeconomic Pathways (SSPs) and Representative Concentration Pathways (RCPs). Specifically, SSP1, SSP2, SSP3, SSP4, and SSP5 represent sustainable development, moderate development, local development, unbalanced development, and conventional development, respectively [56,57,58]. The study utilizes three scenario combinations, including a high-carbon scenario (SSP5-8.5), a medium-carbon scenario (SSP2-4.5), and a low-carbon scenario (SSP1-2.6). SSP5-RCP8.5 (SSP585) relies on high fossil fuel emissions for economic growth, leading to the highest greenhouse gas emissions forecasted for this century, with limited implementation of low-carbon energy policies. SSP2-RCP4.5 (SSP245) features relatively stable economic conditions with a slight increase in greenhouse gas emissions and moderate implementation of low-carbon energy policies. SSP1-RCP2.6 (SSP126) emphasizes global cooperation and technological innovation, aiming to control greenhouse gas emissions with significant effects from low-carbon policies. Among these, SSP5-8.5 results in the highest temperature increase, followed by SSP2-4.5 and SSP1-2.6. SSP1-2.6 has the least impact on global warming. A comparison of the characteristics of different SSP–RCP pathways is presented in

Table A1. Comparison of pathway characteristics of different SSP–RCPs.

Scenario Hypothesis	SSP1-2.6	SSP2-4.5	SSP5-8.5
Characteristic	Very low greenhouse gas emission	Intermediate stabilization	Very high greenhouse gas emissions
Radiative forcing	2.6 W/m2	4.5 W/m2	8.5 W/m2
Scenario definition	This scenario emphasizes the global community’s efforts toward sustainable development by incorporating economic, social, and environmental goals. There is a significant increase in investments in low-carbon technologies and energy-saving retrofits of facilities to ensure long-term prosperity and sustainability.	This scenario describes a future development trend that generally follows historical patterns, with stable economic growth, but it is not sufficient to achieve sustainability. Existing policy measures gradually expire, and no new policies are introduced. The power sector has largely achieved the industry-specific goals that have been announced. Power sector-related technologies have seen some development but are unlikely to achieve significant breakthroughs in the short to medium term.	Under this scenario, environmental protection is not a primary consideration; economic growth is seen as the most important goal, and factors such as environmental protection and carbon emissions are neglected. Carbon emissions will continue to increase, causing severe impacts on the world due to climate change.
Supply-Side Policies	Intensify the scale-up of renewable energy electricity generation, encourage self-generation, and by 2050, ensure that non-hydro renewable energy accounts for at least 39.0% of electricity generation through renewable energy quota policies.	Modest development of renewable energy, aiming for non-hydro renewable energy to make up at least 31.0% of electricity generation by 2050 through renewable energy quota policies.	No development of renewable energy.
Demand-Side Policies	Efficient demand-side management, such as emphasizing the construction of distribution networks and distributed power sources, and large-scale development of energy storage technologies.	Moderate development of demand-side management and moderate promotion of the development of energy storage technologies.	No development of demand-side management and energy storage technologies.

.

The changes in meteorological parameters in Jiangsu under these three climate change scenarios from 2025 to 2035 are shown in Figure 7.

As shown in Figure 7a, the peak temperature rise observed around 2033 to 2035 under the SSP3-8.5 scenario coincides with the sharp decrease in heating degree days (HDD) in Figure 7b. This phenomenon further supports the concept of milder winters with shorter cold periods. Meanwhile, Figure 7c shows a significant increase in cooling degree days (CDD) during the same period, where rising temperatures directly lead to the increase in CDD, indicating longer and more intense hot periods. The simultaneous rise in temperature and CDD suggests an increased demand for cooling energy, especially under high-emission scenarios. As observed in Figure 7d,e, the wind speed peaks around 2027 and 2033 coincide with changes in the precipitation peaks during the same period. This indicates that changes in wind speed may be influenced by variations in atmospheric pressure systems, which in turn affect precipitation patterns. The strong wind conditions during these periods may also explain the increase in rainfall variation. The lower wind speeds in SSP3-8.5 may reduce evaporation rates, which could contribute to more stable but lower precipitation levels. In Figure 7f, the peak in solar irradiance around 2029 in the SSP2-4.5 scenario correlates with increased precipitation. This suggests that clearer skies during periods of lower rainfall allow for higher solar radiation. The significant peaks in 2029 under SSP1-2.6 and SSP3-8.5, as well as the notable low point in SSP2-4.5 around 2033, may be linked to higher concentrations of atmospheric aerosols (such as pollutants or natural dust) during this stage.

Table A2. Future planning of certain policy indicators.

Indicator Level	Indicator	Policy Basis	Specific Description
Secondary Indicators	Proportion of distributed power installed capacity	Jiangsu Province Carbon Peak Implementation Plan	The installed capacity of distributed power sources in Jiangsu is expected to maintain a high growth rate, with installations projected to reach 33–35 million kilowatts by the end of 2024 and 40–45 million kilowatts by the end of 2025.
Secondary Indicators	Renewable energy penetration rate	Jiangsu Province’s “14th Five-Year Plan” Special Plan for Renewable Energy Development	By 2025, renewable energy is expected to account for more than 15% of the total energy consumption in Jiangsu Province.
Secondary Indicators	Proportion of energy storage installed capacity	Jiangsu Province’s “14th Five-Year Plan” Implementation Plan for New Energy Storage Development	By 2025, the province aims to achieve an installed capacity of energy storage of 2.6 million kilowatts, with a clear target of reaching approximately 5 million kilowatts for new energy storage projects province-wide by 2027.
Underlying Indicators	R&D investment in low-carbon technology as a proportion of GDP	Jiangsu Province Statistical Bulletin on National Economic and Social Development	Under the SSP2-4.5 scenario, there is slow annual growth (approximately 0.05% per year), while under the SSP1-2.5 scenario, there is rapid annual increase (approximately 0.075% per year).
Underlying Indicators	Energy-saving retrofit investments in facilities as a proportion of GDP	Jiangsu Province Statistical Bulletin on National Economic and Social Development	Under the SSP2-4.5 scenario, there is slow annual growth (approximately 0.03% per year), whereas under the SSP1-2.5 scenario, there is rapid annual increase (approximately 0.05% per year).
Underlying Indicators	Distributed power installation subsidy	Jiangsu Province "14th Five-Year" Plan for Science and Technological Innovation	In Jiangsu Province, the subsidy for electricity generation special projects is up to 30% of the total amount of subsidies actually issued annually; in other regions, the subsidy is up to 15%.
Underlying Indicators	Energy-saving retrofit investments in facilities as a proportion of GDP	Notification on Several Measures to Accelerate the High-Quality Development of New Energy Storage Projects in Our Province	From January 2023 to January 2026, during the peak summer (winter) periods (January, July–August, December), peak cost support will be provided based on the electricity discharged to the grid, with peak costs gradually decreasing as follows: 0.5 yuan/kWh from 2023 to 2024, and 0.3 yuan/kWh from 2025 to January 2030.

shows the changes in certain policy indicators in Jiangsu, with data sourced from the China Electric Power Research Institute. Due to confidentiality reasons, some policy indicators are not disclosed here.

5.3. Expected Changes in Future Industrial Electricity Demand in Jiangsu

Based on the multivariate regression relationships derived from previous studies on factors affecting electricity consumption, combined with the forecast results of meteorological factors and future energy policy planning indicators for Jiangsu Province, we can predict the changes in electricity consumption in Jiangsu Province by 2035.

In the SSP1-2.6 scenario, policies focus on low-carbon, sustainable development with an emphasis on efficient demand management and renewable energy promotion. This scenario aims to enforce strict energy efficiency improvements and widespread adoption of renewable energy. For example, plans include increasing investments in low-carbon technology R&D and promoting energy-saving renovations to reduce industrial electricity demand. This explains the steady growth in electricity demand around 2030 seen in the graph. Climatic factors such as temperature increases may lead to higher cooling demands, but due to efficient energy use and widespread application of renewable energies, the increase in electricity demand is effectively controlled.

The SSP2-4.5 scenario assumes a moderate level of policy intervention and technological development. In this scenario, policies will moderately encourage the development of distributed energy sources, such as solar and wind power, which will lead to a peak in industrial electricity demand around 2025. Furthermore, with the achievement of distributed energy storage targets, the installed capacity peak of energy storage is expected by 2027, which will help stabilize the electrical grid load and thus impact electricity demand trends. Climatic factors, especially the rise in CDD, will drive the demand for air conditioning and refrigeration, thereby increasing electricity demand in the short term.

In the SSP3-8.5 scenario, the forecast for industrial electricity demand in Jiangsu Province shows significant fluctuations, primarily because this scenario emphasizes economic growth above all, neglecting environmental protection and sustainable energy development. With the rapid increase in economic activities and the warming climate leading to higher cooling needs, electricity demand sharply increases around 2030. However, due to a lack of investment in renewable energies and efficient energy management systems, combined with the uncertainty and cost issues brought by fossil fuel dependence, electricity demand begins to decline after reaching a peak. This trend reflects the sustainability challenges faced by the energy system under high carbon emissions and low environmental policy support.

From Figure 8, it can be concluded that the forecast results under different SSP–RCP pathways all indicate that the level of electricity consumption in Jiangsu will fluctuate and rise in the future. By 2035, the industrial electricity consumption in Jiangsu Province will reach between 767.51 and 794.32 billion kilowatt-hours.

6. Conclusions and Future Work

This study has developed a multivariate regression model that comprehensively considers policy and meteorological factors, aimed at providing decision support for China’s goals of achieving a carbon peak and carbon neutrality. The model stratifies policy factors influencing electricity demand and incorporates differential equations to describe the cumulative effects of these factors, thus rationally integrating policy impacts into the forecast of electricity demand. By introducing policy factors, this paper significantly supplements the existing literature methodologically, especially on how to directly map policy changes into the electricity demand forecasting model. This innovation not only enhances the model’s adaptability to future policy changes but also improves the accuracy of forecasts in complex policy environments, providing a scientific basis for formulating energy policies aligned with China’s carbon neutrality strategy.

The empirical analysis results show that policy factors such as the electrification rate, energy consumption intensity, and industrial structure adjustments, as well as meteorological factors like temperature, humidity, and duration of sunlight, all have a significant impact on industrial electricity demand. Policy factors influence electricity demand through promoting electricity substitution and the development of distributed power sources, while meteorological factors mainly affect demand levels through changes in cooling and heating loads. This presents a clear distinction from traditional models that only consider a single factor.

The model further predicts that by 2035, industrial electricity demand in Jiangsu Province will exhibit varying degrees of fluctuating upward trends under different development scenarios, reflecting the dual impacts of future policies and climate change. Under low-carbon scenarios, the growth in electricity demand is relatively moderate, whereas under high-carbon scenarios, demand growth accelerates significantly. This provides a forward-looking reference for future planning and investment decisions in the electricity systems of Jiangsu Province and China at large.

Overall, the multivariate regression model established in this study can comprehensively capture the main drivers affecting electricity demand and offers some guidance for supporting energy transformation and achieving climate change goals.

To further deepen the research and improve the accuracy of model predictions, future work will primarily focus on the following: (1) Further exploring the interplay between population growth and GDP growth and energy policy. In particular, analyzing how energy policies are adjusted to meet the needs of socio-economic development under different population structures and economic growth backgrounds, and how these adjustments affect the long-term trends in industrial electricity demand. (2) Integrating population dynamics and GDP development trends with specific energy policies in Jiangsu Province to research and simulate future changes in population and GDP with electricity demand under energy policies, especially in the face of economic fluctuations and adjustments to population policies, ensuring effective planning and sustainable management of the power system.