A Novel Energy Management Optimization Method for Commercial Users Based on Hybrid Simulation of Electricity Market Bidding

Abstract

Energy management and utilization for commercial users is becoming increasingly intelligent and refined, fostering a closer and growing connection with the electricity market. In this paper, a novel energy management optimization theoretical framework for commercial users is proposed based on the hybrid simulation of electricity market bidding. The hybrid simulation model based on Multi-Agent Simulation (MAS) with reinforcement learning and System Dynamic Simulation (SDS) is established to solve the problem using a single simulation method: it cannot adjust the clearing price when considering the whole market; considering the uncertainty of Electric Vehicles (EVs) travel and Lighting Loads (LLs), the multi-objective optimization model of energy management for commercial users is constructed to minimize the total energy cost of commercial users, as well as maximize the lighting comfort of indoor office staff, which compensates for the lack of the single-objective optimization of the power consumption for commercial users. A multi-objective optimization model of energy management for commercial users is established based on the hybrid simulation of electricity market bidding. By running the multi-objective optimization model based on hybrid simulation, the results show that the proposed method can realize the optimization of energy management for commercial users considering electricity market bidding.

1. Introduction

At present, the construction of the smart grid and Energy Internet has become the strategic direction of global energy development [1]. As an essential part of the smart grid, intelligent power consumption realizes the flexible two-way interaction between the grid and its users, significantly changing the users’ power consumption mode [2]. As a powerful eco-friendly initiative, the number of EVs with good interoperability regarding intelligent power consumption is rapidly rising worldwide. If well-integrated with the urban environment, they will become a key component of the smart city concept [3]. Vehicle-to-grid technology breaks through the mode in which EVs can only be charged as load, and realizes the two-way energy flow between EVs and the power grid. The disorderly charging/discharging of EVs will lead to increased power consumption expenditure and low energy utilization among power users [4].

Additionally, the urbanization process has led to the continuous emergence of commercial buildings, and the proportion of energy consumption in commercial buildings continues to rise. Especially under the severe situation of the energy crisis and environmental pollution, it is particularly important to implement power management on the commercial user side. With the increasingly intelligent and refined energy management and utilization in commercial buildings, managing various electrical equipment to reduce total energy costs becomes a key issue. Additionally, a close relationship between user-side power management and electricity market bidding should be established to form a benign interaction mechanism. Compared with residential and industrial users [5], commercial users have the characteristics of large power consumption, a high degree of automation and significant participation in power grid demand response [6]. At the same time, the business hours determine that the electricity load is an important part of the peak load of the power system. However, the literature on its management optimization is limited [7]. Therefore, research on the optimization strategy of power consumption management for commercial users considering EV travel is of great significance in guiding users to use electricity responsibly, reduce their electricity expenditure and improve their energy utilization [6].

An overview of recent energy management methods is provided in [8]. An unconstrained, large-scale, global energy management optimization problem was effectively solved through the adaptive particle swarm optimization algorithm in [9]. Moreover, a mixed integer linear programming model, which can optimize the forecast uncertainty intervals of wind power, was proposed in [10]. In [11], the research group proposed a User Dominated Demand Side Response (UDDSR)-based energy management optimization method for Integrated Energy Systems (IES), taking the willingness of consumers into account. A real-time algorithm for cost optimization was presented in [12] to achieve the demand-side energy management of a renewable energy source-integrated microgrid. The above ideas can effectively guide research on optimizing power management for commercial power consumption.

The overall operation of electricity markets is a complex system, involving a large number of different entities. Therefore, it is necessary to study the electricity market based on establishing the complex system model and using the complex system simulation method. The main complex system simulation methods, Multi-Agent Simulation (MAS) and System Dynamics Simulation (SDS) [13] are often used to simulate the electricity market. As the key to the electricity market, electricity price is used by Ciferri et al. to integrate multiple electricity markets [14]. System Dynamics Simulation (SDS) studies the behavior of the system at the macro level by analyzing the feedback relationship between the variables in the complex system [15]. While simulating the dynamic environment, in order to reflect the homogeneity and heterogeneity of complex socio-economic systems, a hybrid simulation method based on the SDS top-down modelling method and MAS bottom-up modelling method are proposed. Over the past two decades, an increasing number of scholars have begun to explore the integration of SDS and MAS for modelling and simulation [16].

The interaction between the electricity market and user-side power consumption management represents a popular research direction in recent years. Current research results mainly focus on the interaction between fixed day-ahead electricity prices and the user-side management, without considering the impact of dynamic price changes in the electricity market on the optimization of user-side energy management. Using the proposed industrial energy management scheme, in [17], electricity demand is transferred during peak hours to off-peak hours to save energy costs of industrial equipment, based on day-ahead hourly electricity prices. In [18], the peak–valley difference of the user-side power grid is taken as the objective function, optimizes the user-side power consumption behavior using the particle swarm optimization algorithm and obtains a reasonable time-sharing electricity price strategy. In [19], a multi-energy transaction decision-making model for a commercial park operator was constructed, considering electric energy substitution, as well as a game, which formulates the energy sales price and mobilizes the user-side interaction. Considering the interaction between the real-time electricity price and the electricity demand between the electricity retailer and the user, a real-time pricing mechanism that is updated with the electricity demand forecast can be developed [20], to increase the electricity retailer’s profit. In the process of forecasting and pricing, the asymmetry in the information between power suppliers and consumers cannot be ignored. In [21], a practical framework for this feature is proposed. Linking electricity market bidding with demand response, that is, guiding commercial users to optimize their power consumption behavior through electricity price signals, can strengthen the role of the demand side in the balance of supply and demand in the market. Fully introducing competition on the power generation side and the power sales side can improve the production efficiency of the power generation side and reduce the electricity price on the user side. Users can employ intelligent power technology to control power equipment, change power consumption patterns to participate in demand response and thus interact with the power system. While saving on power costs, they can also alleviate power supply shortages, promote intermittent renewable energy and optimize power generation and consumption allocation. Therefore, this paper studies energy management optimization for commercial users based on the hybrid simulation of electricity market bidding.

The contributions of this paper are as follows: (1) A novel energy management optimization theoretical framework for commercial users is proposed based on the hybrid simulation of electricity market bidding, and energy management for commercial users considering electricity market bidding is realized by using a simulation-based optimization method. (2) A hybrid simulation model of electricity market bidding is established based on Multi-Agent Simulation (MAS) with Reinforcement Learning (RL) and System Dynamic Simulation (SDS). The model solves the problem where a single simulation method cannot adjust the clearing price considering the whole market. (3) For commercial users, considering the travel and load uncertainty of Electric Vehicles (EVs) and Lighting Loads (LLs), a multi-objective optimization model of energy management for commercial users is proposed, which compensates for the lack of the single-objective optimization of commercial power consumption. (4) A multi-objective optimization model for commercial users based on the hybrid simulation of electricity market bidding is established, and power management for commercial users considering electricity market bidding is realized through the simulation-based optimization method.

2. Energy Management Optimization Theoretical Framework Based on Hybrid Bidding Simulation

Based on the research results of the authors’ research group on energy management and electricity market bidding simulation [6,22], referring to the operation mode of the Texas power market in the United States [23] and the trading mode of China’s spot market [24], we propose a novel energy management optimization theoretical framework for commercial users based on the hybrid simulation of electricity market bidding, as shown in Figure 1. Different from the previous research, this paper adds the commercial user agent and commercial power management optimization on the basis of the hybrid simulation model established in the authors’ previous research paper [22]. In addition, our research focuses on the combination of electricity market bidding simulation and intelligent power consumption optimization management. A simulation and optimization-based analysis of the impact of the electricity market on the management of commercial electricity consumption is studied in our manuscript. The electricity market directly provides hourly updated price information for commercial buildings, shortens the transaction cycle, and provides more refined management optimization for commercial buildings, enabling them to achieve goals such as power consumption optimization, and cost reduction. It is worth mentioning that there is currently no literature that considers the visual comfort of commercial office buildings.

Figure 1 shows the electricity market environment in which information can be exchanged among multi-agents composed of power plant agents, the Independent System Operator (ISO) agent and the commercial user agent, which jointly realize electricity market bidding and energy management optimization for commercial users. The power plant agent is used for intelligent quotation, the ISO agent performs the function of market clearing and the commercial user agent optimizes energy management. In this process, not only is ISO uniform allocation required to avoid the formation of a monopoly, and to instead achieve a reasonable allocation of resources, but government supervision and macro control are also required. In particular, government regulation is included in the ISO agent, represented by the variable HHI (Herfindahl–Hirschman Index). Two electricity market bidding simulation methods, System Dynamics-based Simulation (SDS) and Multi-Agent Simulation (MAS) are included in the ISO agent and power plant agents, respectively. SDS controls the operation of the entire system from a macro perspective, observes changes in the system by monitoring the main variables and weakens individual behaviors. In contrast, MAS simulates the behavior of individuals in a system from a microscopic perspective. The two are combined using some variables, such as market clearing price (MCP) and market demand. It is worth noting that the method for calculating market clearing prices in this paper follows typical market operations in [22], see [22] for more details.

In the ISO agent, the causality diagrams in SDS consist of four basic causality circles: the declared electricity generation can affect the declared quantity of electricity purchased by affecting the declared electricity supply–demand ratio. In addition, the declared electricity generation can also affect the market clearing price by affecting the declared electricity supply–demand ratio, and then act on itself under the joint action of the profit margin and generation cost to form a negative feedback. Then, under the joint action of the declared electricity supply–demand ratio and capacity retention ratio, the market clearing price affects the declared quantity of electricity purchased, and then the declared electricity generation. The declared electricity concentration is used to detect whether the market is monopolized. When its value increases, the market tends to monopolize, and the clearing price rises, which will affect the declared quantity of electricity purchased, and then the declared electricity generation. In addition, market clearing rules in the ISO agent are implemented to ensure orderly market clearing: the Dynamic Queuing Algorithm (DQA) is used to collect and sort the quotations of each power plant agent in each round of the electricity market bidding, so as to calculate the market clearing price.

All power plant agents together form a micro bidding model, of which the clearing period is in days. After obtaining market demand information, reinforcement learning algorithms will be used to select the current optimal strategy for each power plant agent to participate in the bidding. Subsequently, each power plant agent calculates the profit and updates its strategy space according to the market winning results.

The commercial user agent mainly includes a multi-objective optimization model of power consumption, which consists of a typical commercial energy management optimization system, as shown in Figure 1. Inputs in the system include electricity and natural gas, enabling the system to deliver electricity, cooling and heating to end users through energy cascade utilization.

Based on the above theoretical framework, a bidding simulation process of the market mechanism can be derived as follows:

Step 1: The simulation starts, and the ISO agent publishes the real-time updated market demand in the SDS. Details are provided in Section 3.2.2.

Step 2: Each power plant agent generates a bidding strategy space; selects a bidding strategy; and submits bidding information, such as the bidding quantity and bidding price. Details are provided in Section 3.2.1.

Step 3: The ISO agent collects the bidding information submitted by each power plant agent, calculates MCP according to the bidding mechanism and feeds it back to the SDS, before selecting the MCP after the overall adjustment of the SDS as the final MCP. See Section 3.1 for details.

Step 4: The ISO agent sends the market clearing information to the commercial user agent, so that it can manage and optimize the electricity demand of different electrical equipment on the user side. See Section 4 for details.

Step 5: The ISO agent sends the market clearing information to the power plant agent for trading, and each power plant agent updates its strategy space to reformulate the bidding strategy for the next round. Details are provided in Section 3.2.1.

Step 6: At this point, the simulation optimization cycle ends.

3. Hybrid Simulation of Electricity Market Bidding

3.1. Macrobidding Modelling of Electricity Markets Based on SDS

On the basis of the causality diagrams in Figure 1, by dividing the whole system into four subsystems, production, transmission, transaction and supervision, the causal relationship between them can be described, so that the whole system can operate normally.

Table 1. Main indicators in SDS of the electricity market.

Indicator	Subsystem	Meaning
declared electricity generation (SRE)	production subsystem	electricity declared by the power plant before bidding
profit margin (LP)	production subsystem	the profit margin of power generation, that is, the relative value of the profit of the power plant (LP/LCOE-1)
generation cost (LCOE)	production subsystem	additional cost per unit of electricity
declared quantity of electricity purchased (DRE)	transmission subsystem	predicted electricity demand published by ISO before the bidding, and its value is equal to the load
clearing price (MCP)	transaction subsystem	market clearing price, that is, the final clearing price after clearing
declared supply–demand ratio (SDR)	transaction subsystem	the ratio of SRE to DRE reflects the commodity supply and demand in the electricity market
declared electricity concentration (HHI)	supervision subsystem	the sum of the squares of the market shares of each power plant reflects whether the market is fully competitive
capacity retention ratio (RP)	production subsystem	reflects the power plant’s control over power consumption, and the value is (1-SRE/actual power generation)

abbreviates the variables of SDS in Figure 1.

In Table 1, different subsystems are represented by different variables. SRE, DRE and LP are selected as state variables to represent the three subsystems of production, transmission and transaction, respectively, while the supervision subsystem is represented by the variable HHI. Additionally, the SDR in SDS in the electricity market bidding model, in practice, is used as an important variable to adjust the market supply and demand. It is worth mentioning that the micro representation of the transaction subsystem is micro bidding modelling based on Multi-Agent Simulation, as shown in Section 3.2.

Additionally, the system flow diagrams shown in Figure 2 are built in AnyLogic software. Stocks (SRE, DRE and MCP), flows (SRER, DRER and MCPR), auxiliary variables (LCOE, HHI, LP and SDR) and constant variables (RP) form four causal feedback cycles, which affect one another and are interlinked.

However, the system flow diagram can only qualitatively show the correlations between various indicators in the system, and cannot make the model run. Therefore, system dynamics equations need to be further determined. Based on [13], through parameter debugging and continuous simulation in AnyLogic, the quantitative relations, Equations (1)–(8), described by differential equations can be obtained. Equations (1)–(3) are state equations, Equations (4)–(6) are rate equations and Equations (7) and (8) are auxiliary equations.

$$d\mathrm{SRE}/dt=\mathrm{SRER}$$

(1)

$$d\mathrm{DRE}/dt=\mathrm{DRER}$$

(2)

$$d\mathrm{MCP}/dt=\mathrm{MCPR}$$

(3)

$$\mathrm{SRER}=a\times (1-\mathrm{SDR})\times \mathrm{SRE}\times (b\times \mathrm{HHI}-1)+c\times \mathrm{LP}+d$$

(4)

$$\mathrm{DRER}=e\times (\mathrm{SDR}-1)\times \mathrm{DRE}\times (f\times \mathrm{HHI}-1)+g\times \mathrm{MCP}+h$$

(5)

$$\mathrm{MCPR}=i\times (1-\mathrm{SDR})\times \mathrm{MCP}+j\times \mathrm{RP}$$

(6)

$$\mathrm{LP}=\mathrm{MCP}/\mathrm{LCOE}-1$$

(7)

$$\mathrm{SDR}=\mathrm{SRE}/\mathrm{DRE}$$

(8)

where SRER, DRER and MCPR are flows, representing the changes in SRE, DRE and MCP per unit of time, respectively. The above equations are the core of the operation of the electricity market, which can reflect the quantitative relationship between various parts of the whole market. By changing parameters and variables, the operation of the whole model can be observed to determine whether there are severe problems in the current electricity market.

Table 2. Parameter settings in the SDS model.

Parameter	Value	Parameter	Value	Parameter	Value
a, e	0.0002	c, g, j	0.001	i	0.01
b, f	4	d, h	0

shows the parameter values in (1)–(8) after the simulation results converge.

3.2. Microbidding Modelling of Electricity Markets Based on Multi-Agent Simulation

This paper sets up four power plant agents, one ISO agent and one commercial user agent. To simulate actual market conditions, different maximum power generation and maximum quotations are set for each power plant agent.

3.2.1. Intelligent Bidding Simulation of Power Plant Agents

(1)

Determination of Quotation Format and Strategy Space

The quotation format of power plants is a segmented step curve [25]. At the same time, since the unit cost increases as power generation increases, the power generation cost function is also set as a piecewise function, which is consistent with the quotation format and is divided into four segments.

Based on the determination of the quotation format, the strategy space of each power plant can be generated, and the piecewise linear quotation strategy can be generated using quadratic polynomials. The subsection cost curve of the power plant is shown in Equation (9), where Q_min and Q_max represent the minimum and maximum supply of the power plant, respectively, and cost represents the power generation cost.

$$cost(q)=\{\begin{array}{l}cos{t}_1,Q_{\min }\le q<Q_1\\ cos{t}_2,Q_1\le q<Q_2\\ cos{t}_3,Q_2\le q<Q_3\\ cos{t}_4,Q_3\le q<Q_{\max }\end{array}$$

(9)

The piecewise quotation curve of each power plant is shown in Equation (10). Price represents the electricity price. The greater the supply, the higher the quotation.

$$price(q)=\{\begin{array}{l}pric{e}_1,q_0\le q<q_1\\ pric{e}_2,q_1\le q<q_2\\ pric{e}_3,q_2\le q<q_3\\ pric{e}_4,q_3\le q<q_4\end{array}$$

(10)

Before each quotation strategy is generated, three parameters need to be determined: P_a, P_b and P_c. P_a represents the quotation of each power plant at the minimum output, q_min; P_b represents the quotation at the median output, q_mid, whose value is the arithmetic mean of q_min and q_max; P_c stands for a quotation at maximum output, q_max.

$$q^{\prime }=q-q_{\min },q_d=q_{mid}-q_{\min }$$

(11)

Substitute Equation (11) into $q=a\times {q^{\prime }}^2+b\times q^{\prime }+c$ and obtain Equation (12):

$$\{\begin{array}{l}a=2(p_a+p_c-2p_b)/q_d{}^2\\ b=(4p_b-p_c-3p_a)/q_d\\ c=p_a\end{array}$$

(12)

To generate all quotations, define the upper and lower boundaries of the segment interval; set the lowest quotation and the highest quotation; and determine the capacity segments.

(2)

Intelligent Learning Algorithm of Power Plant Agent

In order to make the agent autonomous, several variables, parameters and events need to be set.

Table 3. Parameters of each power plant.

Agent	Agent 1	Agent 2	Agent 3	Agent 4
Highest quotation (CNY/kWh)	1.04	1.01	1.01	1.01
Lowest quotation (CNY/kWh)	0.98	0.99	0.08	0.08
Median quotation (CNY/kWh)	1.01	1.00	0.97	0.97
Generation cost (CNY/kWh)	(0.3, 0.4, 0.5)	(0.5, 0.7, 0.9)	(0.5, 0.7, 0.9)	(0.5, 0.7, 0.9)
Power interval (103 kW)	(20, 40, 60, 80)	(70, 90, 110, 130)	(50, 70, 90, 110)	(50, 70, 90, 110)

shows the parameters of each power plant agent.

In addition to setting the parameters of each power plant in advance, it is also necessary to define events to trigger algorithms and variables. The Roth–Erev (RE) algorithm is a reinforcement learning algorithm proposed by Roth and Erev. As an adaptive learning method, in which the agent learns through experience gained by continuously interacting with the environment. The algorithm can be used to simulate the strategy quotation process of generators in incomplete information markets. However, the RE algorithm model has the following two problems: First, if the behavior of a certain strategy leads to a very large negative profit, then its corresponding behavioral tendency is negative, which is likely to make its selection probability negative. This does not conform to the definition of probability; secondly, if the profit is 0, the behavioral tendency of each behavioral strategy will decrease in the same proportion, so that the selection probability corresponding to each behavioral strategy remains unchanged, resulting in the cessation of learning. Therefore, to address the problem of the RE algorithm model, an improved RE algorithm is used to avoid the occurrence of the above two problems. For more information about the RE Reinforcement algorithm and its improvement, see reference [26] for details. In order to allow for each power plant agent to gain experience from the bidding history and improve the intelligence of its bidding strategy, this paper applies the improved RE reinforcement learning algorithm to each power plant agent. After being triggered by a specific event, the learning process begins. Equation (13) is the updating formula of the propensity coefficient in the improved RE algorithm [26].

$$\{\begin{array}{cc}{Q}_t(i)=(1-r)\times Q_{t-1}(i)+(1-e)\times R_{t-1}& \mathrm{if} i=\mathrm{num}\\ Q_t(i)=(1-r)\times Q_{t-1}(i)+e\times Q_{t-1}(i)/M& \mathrm{if} i\ne \mathrm{num}\end{array}$$

(13)

where t stands for learning times; i represents the strategy number selected when t; r is the forgetting factor, which reduces the decisive influence of experience on the current choice; e is the empirical parameter, which represents the accumulation of early successful decision-making experience; R is the reward after each successful decision; M is the number of strategies. Additionally, set its learning steps [22] as follows:

• Generate a random number in (0, 1) as the initial propensity coefficient of each bidding strategy.
• Adjust the cooling coefficient c, where $c_t=\frac{k}{M}{\displaystyle \sum }_i{Q}_t\left(i\right)$.
• The propensity coefficient of each strategy is used to calculate the selection probability $p_t\left(a\right)=\frac{\exp \left(Q_t\left(i\right)/c\right)}{{{\displaystyle \sum }}_{j=1}^M\exp \left(Q_t\left(i\right)/c\right)}$.
• Perform clearing according to electricity market rules.
• Each agent calculates the output and profit according to the market clearing results.
• Modify each propensity coefficient in the strategy space.
• If not end, return to step 2; otherwise, end.

The cooling coefficient c, can partially determine the influence of the tendency coefficient on the selection probability. The value of c will affect the convergence of the simulation model. If c adopts a fixed value, the tendency coefficient will increase with the increase in number of simulations, so it will suddenly converge without regularity. Therefore, the dynamic adjustment method is used to adjust the tendency coefficient in each round. The selection probability, $p_t\left(a\right),$ is the probability that each strategy of each power plant agent is selected at time t.

By setting events in the power plant agent and writing a reinforcement learning algorithm, the intelligent quotation function can be realized. First, initialize the bidding strategy, bidding capacity segment and cost of each power plant. Then, set the trigger time when the model starts running. Finally, the strategy update formula is used to update the strategy, which is triggered in the fifth minute, and the recurring update cycle occurs every 24 h.

3.2.2. Market Clearing Rules of ISO Agent

This paper uses unified price clearing (MCP) for market clearing; that is, all electricity market traders trade at the same price in one clearing. In the ISO agent clearing process, the sorting process using the Dynamic Queuing Algorithm (DQA) is as follows:

• Arrange the quotation and electricity quantity of the power plant in ascending order. First, sort the quotation and generate sequence numbers. If the sequence numbers are the same, then sort the electricity quantity.
• Accumulate electricity quantity in sequence.
• Identify the serial number that can meet the demand according to the current electricity demand.
• The quotation corresponding to the selected serial number is regarded as the unified clearing price, and the power plants lower than the quotation are successful in bidding.

By setting events and market clearing rules in ISO agents, orderly market clearing can be realized. The bidding strategies (quotation, capacity segment and serial number) of each power plant are collected and allocated to three groups of array variables. Then, set the trigger time as the first minute and the recurring update cycle as 24 h. Following this, bubble sorting is used to sort the collected information, which is triggered in the second minute, and the recurring update cycle occurs every 24 h.

3.3. Generation of Twenty-Four-Hour Electricity Prices

In order to obtain the dynamic electricity price within a day, it is necessary to generate market demand and electricity prices for different periods. This is achieved by setting the twenty-four-hour demand, as shown in

Table 4. Twenty-four-hour electricity demand.

Time/h	Demand/MWh	Time/h	Demand/MWh	Time/h	Demand/MWh	Time/h	Demand/MWh
1	60	7	80	13	140	19	120
2	60	8	90	14	120	20	140
3	60	9	100	15	120	21	140
4	60	10	120	16	110	22	120
5	60	11	140	17	110	23	60
6	60	12	150	18	110	24	60

. By repeating the simulation 24 times, market clearing prices can be obtained, as shown in

Table 5. Twenty-four-hour electricity price (CNY/kWh).

Time/h	Price	Time/h	Price	Time/h	Price	Time/h	Price
1	0.44	7	0.44	13	1.15	19	0.99
2	0.44	8	0.71	14	0.71	20	1.15
3	0.44	9	0.99	15	0.71	21	1.15
4	0.44	10	0.99	16	0.71	22	0.99
5	0.44	11	1.15	17	0.71	23	0.99
6	0.44	12	1.15	18	0.71	24	0.44

.

4. Multi-Objective Optimization of Energy Management for Commercial Users

The commercial user agent will optimize the operation status of each part of the equipment according to the electricity price information from the electricity market.

4.1. Electrical Equipment Modelling

4.1.1. Mathematical Model of Power Supply Units

(1)

Micro-gas Turbine

The micro-gas turbine is the core equipment of the CCHP (Combined Cooling, Heating and Power) system. The output electric power, $P_{mt,t}$, and residual heat power, $Q_{mt,t}$, of a micro-gas turbine at time t are given by:

$$\{\begin{array}{l}{P}_{mt,t}=F_{mt,t}{H}_{ng}{\eta }_{mt}\\ Q_{mt,t}=F_{mt,t}{H}_{ng}(1-{\eta }_{mt})\end{array}$$

(14)

where $F_{mt,t}$ represents the amount of natural gas consumed by a gas turbine at time t given in m³/t; $H_{ng}$ represents the calorific value of natural gas, and it generally takes the value of 9.78 kWh/m³; ${\eta }_{mt}$ is the electrical efficiency of a gas turbine; $0\le P_{mt,t}\le P_{mt}^{rated}$, where $P_{mt}^{rated}$ denotes the rated output electric power of a micro-turbine.

(2)

Gas Fired Boiler

When the demand for heat is high, such as in winter, a gas boiler is usually used, but less so in summer. The output heat power of a gas-fired boiler at time t is given by:

$$Q_{gb,t}=F_{gb,t}{H}_{ng}{\eta }_{gb}$$

(15)

where $F_{gb,t}$ represents the amount of natural gas consumed by a gas-fired boiler at time t given in m³/t, ${\eta }_{gb}$ represents the efficiency of a gas-fired boiler, $Q_{gb}^{rated}$ represents the rated power of the gas-fired boiler, and $0\le Q_{gb,t}\le Q_{gb}^{rated}$.

(3)

Photovoltaic (PV) Generation Unit

Due to the instability of photovoltaic power generation, it is greatly affected by weather and prone to fluctuations. In addition, the mechanism of photovoltaic output is not within the scope of this paper. Therefore, this paper directly adopts the existing research results of photovoltaic forecasting in the literature [27] to study the optimization strategy of intelligent power consumption operation of commercial users, and does not analyze the working characteristics of photovoltaic power generation devices in detail. For details, please refer to [27].

4.1.2. Mathematical Model of Auxiliary Equipment

(1)

Electric Refrigerator

The mathematical relationship between the input electric power, $P_{ec,t}$, and the output cooling power, $Q_{cec,t}$, of an electric refrigerator at time t is given by:

$$\{\begin{array}{l}{Q}_{cec,t}=P_{ec,t}CO{P}_{ec}\\ 0\le P_{ec,t}\le P_{ec}^{rated}\end{array}$$

(16)

where $CO{P}_{ec}$ represents the refrigeration coefficient of the electric refrigerator, which is also known as the energy efficiency ratio, denotes the ratio of the cooling capacity per unit of time to the electricity, and is used to measure the refrigeration performance of the electric refrigerator. $P_{ec}^{rated}$ is the rated power of the electric refrigerator.

(2)

Absorption Refrigerator

At present, lithium bromide chillers are widely used. The mathematical expression of the heat-driven refrigeration of an absorption chiller is given by:

$$\{\begin{array}{l}{Q}_{cac,t}=Q_{hac,t}CO{P}_{ac}\\ 0\le Q_{hac,t}\le Q_{hac}^{rated}\end{array}$$

(17)

where $Q_{hac,t}$ represents the input heat power of an absorption chiller at time t, $Q_{cac,t}$ is the output cooling power of an absorption chiller at time t, and $CO{P}_{ac}$ is the refrigeration coefficient of an absorption chiller, which represents the ratio of the input heat to the cooling capacity per unit of time. $CO{P}_{ac}$ is used to measure the refrigeration performance of an absorption chiller. Finally, $Q_{hac}^{rated}$ denotes the rated power of an absorption chiller.

(3)

Waste Heat Recovery Unit

The mathematical expression of a waste heat recovery unit is given by:

$$\{\begin{array}{l}{Q}_{cre,t}=Q_{mt,t}{\eta }_{re}\\ 0\le Q_{mt,t}\le Q_{mt}^{rated}\end{array}$$

(18)

where $Q_{mt,t}$ represents the residual heat power generated by a microgas turbine, $Q_{cre,t}$ represents the residual heat power output by a waste heat recovery unit, ${\eta }_{re}$ denotes the recovery of the waste heat recovery unit, and $Q_{mt}^{rated}$ denotes the rated power of a waste heat recovery unit.

(4)

Heat Exchanger

Heat exchangers are primarily used to convert heat to thermal energy required by commercial users. The mathematical expression of a heat exchanger is given by:

$$\{\begin{array}{l}{Q}_{hl,t}=Q_{hx,t}{\eta }_{hx}\\ 0\le Q_{hx,t}\le Q_{hx}^{rated}\end{array}$$

(19)

where $Q_{hx,t}$ represents the input power of a heat exchanger at time t, $Q_{hl,t}$ represents the heat power output by a heat exchanger at time t, ${\eta }_{hx}$ is the operating efficiency of a heat exchanger, and $Q_{hx}^{rated}$ is the rated power of a heat exchanger.

4.1.3. Mathematical Model of Energy Storage Equipment

This paper analyzes the operating characteristics of two components of energy storage equipment, batteries and heat storage tanks.

(1)

Battery

Generally, the battery has three working states: charging, discharging and sleeping, and its mathematical model is given by:

$$E_{ees,t+1}=E_{ees,t}+P_{eesch,t}{\eta }_{eesch}\Delta t-\frac{P_{eesdis,t}\Delta t}{{\eta }_{eesdis}}$$

(20)

$$0\le y_{eesch,t}+y_{eesdis,t}\le 1$$

(21)

$$0\le P_{eesch,t}\le y_{eesch,t}{P}_{eesch}^{\max }$$

(22)

$$0\le P_{eesdis,t}\le y_{eesdis,t}{P}_{eesdis}^{\max }$$

(23)

$$E_{ees}^{\min }\le E_{ees,t}\le E_{ees}^{\max }$$

(24)

Equation (20) defines the charge/discharge characteristics of a battery, where $E_{ees,t}$ represents the amount of charge stored in a battery at time t; $P_{eesch,t}$ and $P_{eesdis,t}$ represent the charge and discharge powers of a battery at time t, respectively; ${\eta }_{eesch}$ and ${\eta }_{eesdis}$ indicate the charge and discharge efficiency, respectively. Equation (21) shows that the battery can be in only one working state at time t; $y_{eesch,t}$ and $y_{eesdis,t}$ are binary variables indicating the charging/discharging states of a battery at time t, respectively; (22) and (23) define battery charging/discharging power constraints, where $P_{eesch}^{\max }$ and $P_{eesdis}^{\max }$ are the upper limits of charging and discharging power of a battery, respectively; (24) defines the constraint of avoiding overcharging and over-discharging of a battery at time t, where $E_{ees}^{\max }$ and $E_{ees}^{\min }$ indicate the upper and lower limits of the stored electric quantity, respectively.

(2)

Heat storage Tank

The mathematical model of a heat storage tank is given by:

$$H_{tes,t+1}=H_{tes,t}+Q_{tesch,t}{\eta }_{tesch}\Delta t-\frac{Q_{tesdis,t}\Delta t}{{\eta }_{tesdis}}$$

(25)

$$0\le y_{tesch,t}+y_{tesdis,t}\le 1$$

(26)

$$0\le Q_{tesch,t}\le y_{tesch,t}{Q}_{tesch}^{\max }$$

(27)

$$0\le Q_{tesdis,t}\le y_{tesdis,t}{Q}_{tesdis}^{\max }$$

(28)

$$H_{tes}^{\min }\le H_{tes,t}\le H_{tes}^{\max }$$

(29)

Equation (25) denotes a description of the operational characteristics of a heat storage tank; $H_{tes,t}$ represents the heat stored in a heat storage tank at time t; $Q_{tesch,t}$ and $Q_{tesdis,t}$ represent the storage and heat release power of a heat storage tank at time t, respectively; ${\eta }_{tesch}$ and ${\eta }_{tesdis}$ represent the heat storage efficiency and heat release efficiency. Equation (26) restricts the heat storage tank to be in only one working state at time t, where $y_{tesch,t}$ and $y_{tesdis,t}$ represent the binary variables, indicating the state of heat storage and heat release of a heat storage tank, respectively; (27) and (28) define the heat storage tank storage and heat release power constraint, where $Q_{tesch}^{\max }$ and $Q_{tesdis}^{\max }$ are the upper limits of the storage and heat release powers of a heat storage tank, respectively; (29) is introduced to avoid the over-storage and over-discharge of a heat storage tank at time t, where $H_{tes}^{\max }$ and $H_{tes}^{\min }$ represent the upper and lower limits of the stored heat, respectively.

4.1.4. Mathematical Model of Load Equipment

(1)

Electric Vehicles (EVs)

Based on [28,29], assuming that the arrival times $t_s$ of electric vehicles in the commercial building conform to a normal distribution, its probability density function can be described by:

$$f_s(t_s)=\frac{1}{{\sigma }_s\sqrt{2\pi }}\exp \left(-\frac{{(t_s-{\mu }_s)}^2}{2{\sigma }_s{}^2}\right)\begin{array}{cc}& \end{array}1<t_s\le 24$$

(30)

where ${\mu }_s=9$ and ${\sigma }_s=0.41$ indicate that the arrival time of EVs is concentrated at 9:00.

Assuming that the time when EVs leave the commercial office building is concentrated at 17:00 and 22:00, the probability density function is given by:

$$f_e(t_e)=\frac{r}{{\sigma }_{e1}\sqrt{2\pi }}\exp \left(-\frac{{(t_e-{\mu }_{e1})}^2}{2{\sigma }_{e1}{}^2}\right)+\frac{(1-r)}{{\sigma }_{e2}\sqrt{2\pi }}\exp \left(-\frac{{(t_e-{\mu }_{e2})}^2}{2{\sigma }_{e2}{}^2}\right)\begin{array}{cc}& \end{array}1<t_e\le 24$$

(31)

where ${\mu }_{e1}=17$, ${\mu }_{e2}=22$ and ${\sigma }_{e1}={\sigma }_{e2}=0.41$ indicate that the office staff leave the office building at 17:00 and 22:00, and $r=0.7$ indicates the proportion of office staff leaving the office building at 17:00. In addition to the uncertainty of arrival and departure times of EVs, the initial power of EVs is also considered. Assuming that the initial electric quantity $SO{C}_e^0$ of an EV obeys the normal distribution, its probability density function can be described by:

$$f_0(SO{C}_e^0)=\frac{1}{{\sigma }_0\sqrt{2\pi }}\exp \left(-\frac{{(SO{C}_e^0-{\mu }_0)}^2}{2{\sigma }_0{}^2}\right)\begin{array}{cc}& \end{array}0<SO{C}_e^0\le 1$$

(32)

where ${\mu }_0=0.25$ and ${\sigma }_0=0.02$.

The charging/discharging model of EVs based on the V2G technology is given by:

$$SO{C}_{e,t+1}=SO{C}_{e,t}+\frac{P_{e,ch,t}{\eta }_{e,ch}\Delta t}{C_b}-\frac{P_{e,dis,t}\Delta t}{C_b{\eta }_{e,dis}}$$

(33)

where $SO{C}_{e,t}$ represents the state of charge of an EV at time t; ${\eta }_{e,ch}$ and ${\eta }_{e,dis}$ indicate the charging efficiency and discharging efficiency of an EV, respectively; and $C_b$ represents the battery capacity of an EV.

EVs have three states: charging, discharging and not charging. The variables $y_{e,ch,t}$ and $y_{e,dis,t}$ are used to indicate the operating state of an EV. Equation (34) indicates that EVs can be in only one operating state at any given moment.

$$0\le y_{e,ch,t}+y_{e,dis,t}\le 1$$

(34)

The charging and discharging power constraints of EVs are, respectively, defined by:

$$0\le P_{e,ch,t}\le y_{e,ch,t}{P}_{e,ch}^{\max }$$

(35)

$$0\le P_{e,dis,t}\le y_{e,dis,t}{P}_{e,dis}^{\max }$$

(36)

where $P_{e,ch,t}$ and $P_{e,dis,t}$ represent the charging and discharging powers of an EV at time t, respectively, and $P_{e,ch}^{\max }$ and $P_{e,dis}^{\max }$ represent the maximum values of the charging and discharging powers, respectively.

The charging state of an EV is defined by (37), where $SO{C}_e^{up}$ and $SO{C}_e^{down}$ indicate the upper and lower limits of the SOC of the EV, respectively.

$$SO{C}_e^{down}\le SO{C}_{e,t}\le SO{C}_e^{up}$$

(37)

In order to extend the battery life of an EV, the number of charging and discharging times of EVs is limited as follows:

$$y_t\ge y_{e,dis,t}-y_{e,dis,t+1}$$

(38)

$$y_t\ge y_{e,dis,t+1}-y_{e,dis,t}$$

(39)

$${\displaystyle \sum _{t=b_e}^{e_e}{y}_t}\le y_{dis}$$

(40)

where $b_e$ and $e_e$ indicate the start and stop times of the EV during charging, respectively; $y_t$ is the conversion variable of EV during the charging and discharging states; and $y_{dis}$ indicates the number of charging and discharging times, and can be set in advance.

To ensure the safety and reliability of EVs, they need to meet the mathematical constraint of minimum remaining power before each start, which is given by (41), where $SO{C}_e^{set}$ represents the minimum remaining power when an EV leaves.

$$SO{C}_e^{set}\le SO{C}_e(e_e)\le 1$$

(41)

(2)

Lighting Loads (LLs)

Lighting equipment occupies nearly one-third of the electricity load in the commercial office building. Additionally, regarding power-adjustable lighting equipment, changes in lighting power directly affect the lighting level of the office [28,29,30]. Through appropriate strategies, power-adjustable lighting equipment can effectively reduce the peak power of the power grid [31]. For example, Building Automation and Control (BAC) systems applied to lighting systems constitute an effective method [32]. The lighting power $P_{illm,t}$ of the entire commercial office building at time t can be calculated by:

$$P_{illm,t}=Nn{p}_{illm,t}$$

(42)

where N represents the number of rooms in the commercial office building, n represents the amount of lighting equipment in each room, and $P_{illm,t}$ is the power consumption at time t of each lighting equipment.

4.2. Constraints and Objective Function of Energy Management Optimization

4.2.1. Visual Comfort

This section introduces the average illumination [33] $E_{av,t}$ to measure the level of office lighting at time t. The average illuminance calculated by the coefficient method and its adjustable range can be, respectively, described by:

$$E_{av,t}=\frac{n{p}_{illm,t}{\eta }_{s}UK}{S}$$

(43)

$$E_{av}^{\min }\le E_{av,t}\le E_{av}^{\max }$$

(44)

where ${\eta }_s$ represents the luminous efficacy of the lighting equipment; $U$ and $K$ are the utilization factor and the maintenance factor of the lighting equipment, respectively; $S$ represents the area of each office; $E_{av}^{\max }$ and $E_{av}^{\min }$ indicate the upper and lower illuminance limits, respectively.

The visual comfort index G at time t is represented by indoor lighting, which varies within an acceptable range that is set by the user, and it is defined by [30]:

$$C_{v,t}=1-{\left(\frac{E_{av,t}-E_{av}^{set}}{E_{av}^{set}}\right)}^2$$

(45)

where $E_{av}^{set}$ denotes the standard illuminance, and its unit is lux.

4.2.2. Objective Function

(1)

Economic Objective Function

The energy cost of commercial users includes the cost of interaction with the power grid, $C_{grid}$, and the cost of natural gas, $C_{ng}$. The cost of natural gas represents the cost of natural gas purchased for microgas turbines and gas boilers. The calculation formula of the two costs is given by:

$$C_{grid}={\displaystyle \sum _{t=1}^T{c}_{com,t}^p{P}_{com,t}^p}\Delta t-{\displaystyle \sum _{t=1}^T{c}_{com,t}^s{P}_{com,t}^s\Delta t}$$

(46)

$$C_{ng}={\displaystyle \sum _{t=1}^T{c}_{com,t}^{ng}(F_{mt,t}+F_{gb,t})}$$

(47)

where $c_{com,t}^p$ and $c_{com,t}^s$ are the purchase and sale prices of the electricity market, respectively; $P_{com,t}^p$ and $P_{com,t}^s$ indicate the purchasing and sales power of commercial users, respectively; and $c_{com,t}^{ng}$ is the price of the natural gas purchased.

(2)

Comfort Objective Function

Assuming that the entire lighting power of commercial users are synchronously adjusted by the commercial power system, the comfort index $C_{visual}$ of the commercial integrated energy management can be defined by:

$$C_{visual}={\displaystyle \sum _{t=1}^T{C}_{v,t}}$$

(48)

Commercial users can determine the weight coefficient between economy and comfort according to actual needs, and improve the overall economics while ensuring indoor visual comfort remains within an acceptable range [30].

(3)

Total Objective Function

This paper uses the linear weighted-sum method to solve the proposed multi-objective optimization problem by introducing the weight coefficient $\alpha$ and transforming the multi-objective function established using the optimal economic goal and optimal comfort index into a single-objective function, which is given by:

$$\min F=\alpha (C_{grid}+C_{ng})-(1-\alpha )C_{visual}$$

(49)

where $\alpha$ is 0.00014.

4.2.3. Constraints

The power constraints that commercial users need to meet include the electric power balance, heating power balance, and cooling power balance, which are expressed by:

$$P_{mt,t}+P_{com,t}^p+P_{PV,t}+P_{ec,t}+P_{eesdis,t}+{\displaystyle \sum _{e=1}^{N_e}{P}_{e,dis,t}}=P_{com,t}^s+P_{eesch,t}+{\displaystyle \sum _{e=1}^{N_e}{P}_{e,ch,t}}+P_{illm,t}+P_{l,t}$$

(50)

$$Q_{cre,t}+Q_{gb,t}-Q_{cac,t}=Q_{hl,t}/{\eta }_{hx}$$

(51)

$$Q_{cec,t}+Q_{cac,t}=Q_{cl,t}$$

(52)

where $P_{PV,t}$ and $P_{l,t}$ represent the PV and power consumption of uncontrollable equipment of commercial users, respectively; N_e represents the number of EVs owned by commercial users; $Q_{hl,t}$ and $Q_{cl,t}$ represent the heating power demand and cooling power demand, respectively, of commercial users.

In addition to the mentioned energy balance constraints, commercial users need to meet the purchase and sale state constraints and the maximum power constraints for purchase and sale, which are defined by:

$$0\le z_{com,t}^p+z_{com,t}^s\le 1$$

(53)

$$0\le P_{com,t}^b\le z_{com,t}^p{P}_{com,p}^{\max }$$

(54)

$$0\le P_{com,t}^s\le z_{com,t}^s{P}_{com,s}^{\max }$$

(55)

where $z_{com,t}^p$ and $z_{com,t}^s$ are variables indicating the state of purchase and sale of commercial users at time t, and $P_{com,p}^{\max }$ and $P_{com,s}^{\max }$ represent the upper limits of electricity purchase and power sale of commercial users, respectively.

5. Multi-Objective Optimization Implementation Based on Hybrid Simulation

5.1. Optimization Model Verification of Power Consumption for Commercial Users

5.1.1. Optimization Solution

The decision variables in the multi-objective optimization model of the power consumption management of commercial users in this paper include both linear continuous variables and linear integer variables; that is, the optimization problem not only bears conditional constraints, but also integer constraints. As the traditional algorithms based on intelligent optimization GA and PSO have long optimization times, and the optimization results are greatly affected by the penalty value, the linear weighting-sum method with a fast calculation speed can obtain the optimization results in time. Therefore, on the commercial user side, the integrated energy management model can be transformed into a solvable mixed-integer linear programming (MILP) problem, including linear continuous and integer variables. Consequently, the integrated energy management model established in this paper is solved by using the CPLEX solver on the YALMIP platform in the MATLAB software environment. The YALMIP is a modelling platform suitable for solving large-scale optimization problems.

5.1.2. Validation of Optimization Model

A general commercial office building was investigated as the case study. One day is divided into 48 periods, each 30 min long. The building has 30 floors; each floor is 3 m high and covers an area of 1500 m². The equipment parameters of the CCHP unit are shown in

Table 6. Cooling and heating electrical equipment parameters.

${\eta }_{mt}$	0.3	${\eta }_{eesch}$	0.95	${\eta }_{hesch}$	0.9	${\eta }_{re}$	0.75
${\eta }_{gb}$	0.73	${\eta }_{eesdis}$	0.95	${\eta }_{hesdis}$	0.9	${\eta }_{hx}$	0.9
$P_{ec}^{rated}$ (kW)	300	$E_{ees}^{\min }$ (kWh)	40	$P_{mt}^{rated}$ (kW)	500	$Q_{gb}^{rated}$ (kW)	500
$Q_{hac}^{rated}$ (kW)	500	$E_{ees}^{\max }$ (kWh)	480	$P_{eesch}^{\max }$ (kW)	40	$P_{hesdis}^{\max }$ (kW)	100
$Q_{mt}^{rated}$ (kW)	800	$CO{P}_{ec}$	4	$P_{eesdis}^{\max }$ (kW)	40	$E_{hes}^{\min }$ (kWh)	100
$Q_{hx}^{rated}$ (kW)	300	$CO{P}_{ac}$	0.7	$P_{hesch}^{\max }$ (kW)	100	$E_{hes}^{\max }$ (kWh)	450

. The parameters of single EV and single lighting equipment are shown in

Table 7. Parameters of EV charging and discharging.

$N_e$	200	$C_b$ (kWh)	24	$P_{ch}^{\max }$/$P_{dis}^{\max }$ (kW)	3.3	$SO{C}_e^{set}$	0.9
${\eta }_{ch}$	0.95	${\eta }_{dis}$	0.95	$SO{C}_e^{down}$	0.2	$SO{C}_e^{up}$	1

and

Table 8. Lighting equipment parameters.

N	500	${\eta }_s$	180	$U/K$	0.8	$E_{av}^{set}$	500
$S$	100	$n$	10	$E_{av}^{\min }$	480	$E_{av}^{\max }$	520

, respectively. Additionally, a total of 200 EVs are designed in this paper, which is represented by N_e.

The fitting results of the arrival and departure times of EVs subject to normal distribution are shown in Figure 3a,b, respectively. The unit price of natural gas is CNY 3.24/m³, and the unit calorific value is 9.78 kWh/m³. Commercial users participate in the demand response of the electricity market through the twenty-four-hour electricity price information provided by the electricity market, and the on-grid electricity price of their remaining electricity is summarized as CNY 0.7/kWh.

(1)

Natural Gas and Electricity Purchased/Sold Results

Gas consumption and electricity purchased/sold before and after power consumption optimization are shown in Figure 4, where the electricity price unit is defined as CNY/kWh, which is the same as in other optimization figures. Before optimizing the commercial energy office building, natural gas is generated only by the consumption of heat generated by gas boilers, while the electricity is sourced from the electricity market. From Figure 4a, it can be seen that while local PV is being absorbed, the commercial office building purchases a large amount of electricity during the peak price period, which greatly increases electricity costs and leads to energy inefficiency.

After the optimization of the commercial energy office building, natural gas is mainly consumed by gas turbines during peak price periods, which replaces purchasing electricity from the market. From Figure 4b, it can be seen that the electricity purchased by commercial users is transferred from peak electricity prices to the valley electricity prices evenly. The electricity is sold back to the electricity market from 08:00 to 12:00, which significantly improves the electricity economy and energy utilization of commercial users. Moreover, it improves the stability of the power system with a smoother optimized purchasing curve.

The cost represents the most direct concern of commercial users. The cost analysis of the commercial office building before and after power consumption optimization is shown in

Table 9. Cost of electricity consumption for each equipment in the commercial office building (unit: yuan).

Optimization	Electrical Load	Electric Refrigerator	EVs	LLs	Electricity Purchase	Electricity Sales	Natural Gas	Total Cost
before	7472.8	3753.8	4051	6875	7888	-	1235.4	9123.4
after	7472.8	2615.6	2700.9	6831	3706.7	582.7	3674.1	6798

. Before optimization, commercial users do not install local PV generation equipment, and the total electricity cost would reach CNY 12312. After installing local PV, the total electricity cost would be notably reduced to CNY 9123.4.

From Table 9, it can be seen that after power consumption optimization, the total cost of power consumption for each equipment is optimized except for the fixed power load, which remains unchanged. In addition, the reverse sale during peak electricity prices greatly reduces the total cost of power consumption for commercial users. It should be noted that the regulation of electric power is limited in order to ensure the visual comfort of indoor office staff.

(2)

Electric Power Optimization Results

The optimization results before and after energy management in the commercial office building are shown in Figure 5. Before optimization, the commercial building does not participate in the demand response; the CCHP units do not work; and the electricity demand is mainly met by the electricity market, supplemented by PV equipment. When the EV reaches the commercial building, it begins to charge, and commercial users use electricity centrally from 10:00 to 12:00, which remains within peak electricity prices. This not only greatly increases the electricity costs of commercial users, but also causes peak loading of the power grid, which threatens the stability of the power system. After optimization, the charging periods of EVs are obviously shifted to the valley periods (13:00–18:00), and discharges occur during the peak electricity price.

The arrival time of EVs in the commercial building is concentrated around 9:00 (peak price periods). From Figure 4b and Figure 5b, it can be seen that the commercial power management system will select the EVs with discharge capability to participate in the electricity market demand response and sell the extra electricity to the electricity market for profit.

The departure time of EVs is concentrated at approximately 17:00 and 22:00. It can be seen that some EVs with a discharging capability can discharge at approximately 16:00 and 21:00 to reduce the total electricity cost of commercial users, as shown in Figure 4b and Figure 5b. In addition, the combined use of micro-gas turbines and accumulators allows commercial users to sell electricity during peak price periods (08:00–11:00) after power consumption optimization in the commercial office building.

(3)

Visual Comfort Optimization Results

The results of the indoor lighting index and visual comfort for commercial office buildings after power management optimization are shown in Figure 6. Considering that lighting at night is not required by indoor office staff, the indoor lighting index between 00:00 and 6:00 is considered to be 0.

The simulation results show that the optimized indexes are limited to the set range, which ensures the visual comfort of indoor office staff. Under the effect of the objective function, when the electricity price is low, the indoor lighting index remains near the standard illumination (500 lux), which achieves better indoor visual comfort. On the contrary, the indoor lighting index decreases slightly when the electricity price is high, which improves the economy of commercial users during peak price periods without affecting the indoor office environment.

Through the comparison of power consumption and cost before and after the commercial power consumption optimization, we can see that the proposed commercial power consumption optimization strategy can effectively balance the operation economy of commercial users with the visual comfort of indoor office staff. At the same time, it can achieve the cascade and efficient utilization of different types of energy for commercial users.

5.2. Multi-Objective Optimization Model Verification Based on Hybrid Bidding Simulation

5.2.1. Mathematical Model of Power Supply Units Simulation-Based Optimization Solution

The hybrid simulation model based on electricity market bidding and the multi-objective optimization model based on commercial energy management are integrated by AnyLogic software based on Java. This paper processes the mixed-integer programming model using the CPLEX package. Due to the differences between the power consumption state in summer and winter and the working state of the CCHP system, commercial energy management optimization is discussed below, considering the bidding of the electricity market in summer and winter separately.

5.2.2. Power Consumption Optimization Considering Market Bidding in Summer

The electricity demand in summer is aimed at cooling, which can be converted directly from electricity or heat. As a special kind of electrical equipment, the charging and discharging behaviors of EVs are random. The lighting demand is evenly distributed during the day, while fixed electricity demand will rise during the peak period of power consumption. The EV starts charging as soon as it reaches the building before optimization. The demand for cooling is highest at noon and in the afternoon. The commercial electricity demand is shown in Figure 7. Among them, the cooling energy generated by electric refrigerators is used for cooling load. In addition, PV generation can supplement electrical energy.

Before interacting with the electricity market, electricity prices are divided into daily peak and valley electricity prices. The peak electricity price occurs during the daytime peak period, when electricity costs are higher. After interacting with the electricity market, electricity prices change several times daily. For instance, at approximately 11:00 a.m. and 8:00 p.m., electricity prices remain in a high range, while at other times, electricity prices will fall.

Under the same background, the electricity purchase and natural gas consumption conform to Gaussian distribution and mixed Gaussian distribution. During the day, there is no optimization of power consumption because the electricity price does not fluctuate. In summer, only electricity and lower quantities of gas are used. After interacting with the electricity market, the electricity price in the peak price periods increases, which forces the electricity purchase during peak price periods to decrease. As a result, there is tendency to buy natural gas for electricity generation. Figure 8 shows the comparison between the purchase and sale of natural gas and electricity before and after the commercial office building is associated with the electricity market in summer. The left figures are before optimization, and the right ones are after optimization.

After optimization, the charging time of EVs is delayed to the off-peak periods. In addition, when the price is high, the electric storage equipment will discharge to supplement the power; EVs will also release electricity for the peak power consumption at night. At the peak of electricity price, reducing electricity purchase and increasing gas purchase can fully mobilize the CCHP system and improve energy utilization efficiency. Before optimization, the daily purchased electricity is about 10,550 kWh, and after optimization, the purchased electricity is about 3980 kWh; During summer (three months), the electricity purchased before optimization is about 970,600 kWh, while the electricity purchased after optimization is about 366,160 kWh.

5.2.3. Power Consumption Optimization Considering Market Bidding in Winter

The demand in winter is for electricity and heat, in which heat can be directly produced by natural gas or converted from waste heat of natural gas power generation. Similar to summer, the lighting demand is uniform during the day, and the fixed electricity demand rises during the peak period. Before optimization, EVs will start charging as soon as they arrive at the building. Due to the temperature, the heat demand is strong in the morning and evening. The commercial electricity demand is shown in Figure 9.

In winter, before interacting with the electricity market, the distribution of the electricity purchase and natural gas purchase are the same as those in summer. There will be small fluctuations in the initial gas consumption stage. Since the electricity price does not fluctuate during the day, power consumption is not optimized. In winter, heat consumption increases so that there is more gas purchase. After interacting with the electricity market, the electricity price in the peak periods is higher than before. The optimal result is to reduce the purchase of electricity and buy more natural gas in the peak periods, part of which is used for electricity generation. The comparison of natural gas and electricity purchased/sold before and after the interaction between the commercial buildings and the electricity market in winter is shown in Figure 10. Before optimization, the daily purchased electric energy is about 6260 kWh, and after optimization, the purchased electric energy is about 4050 kWh; During winter (3 months), the purchased electricity is about 575,920 kWh before optimization, and about 372,600 kWh after optimization.

6. Conclusions and Discussion

In this paper, a novel energy management optimization theoretical framework for commercial users is proposed based on the hybrid simulation of electricity market bidding. This framework effectively integrates electricity market bidding simulation and power consumption optimization for commercial users. It realizes the real-time direct interaction of electricity price and electricity quantity between power suppliers and consumers by using the simulation-based optimization method.

A hybrid simulation model of electricity market bidding is established based on Multi-Agent Simulation (MAS) with Reinforcement Learning and System Dynamics Simulation (SDS). The hybrid simulation model can self-regulate through the mutual feedback of its elements, so that the whole model does not need external intervention. The hybrid simulation model can solve the problem where the market clearing price cannot fully reflect the market competition under a single market regulation mechanism.

Considering the uncertainty of Electric Vehicles (EVs) traveling and Lighting Loads (LLs), a multi-objective optimization model of energy management for commercial users is established. The model compensates for the lack of single-objective optimization of commercial power consumption. The calculation results show that the proposed optimization strategy can realize the optimization of energy management and improves the utilization rate of energy.

A multi-objective optimization model of power consumption for commercial users is established based on the hybrid simulation of electricity market bidding. By running the multi-objective optimization model based on the hybrid simulation, the energy management optimization of commercial users based on overall bidding is realized, laying a foundation for the development of the smart grid and Energy Internet.

The novel energy management optimization theoretical framework based on the hybrid simulation of the electricity market bidding demonstrates good generalization and can be promoted to commercial buildings among commercial users. The main reasons are as follows: (1) The charging and discharging scheduling of energy storage will change the market clearing result and the operation plan of the system. When energy storage directly participates in the market, it will bear an important impact on market competition, prices and the benefits of market members [34]. (2) In view of the dispatching reality of a large number of electric vehicles participating in energy storage, combined with the current working modes of Chinese state-owned enterprises and Internet companies, two peak dispatching time intervals are designed. (3) The management optimization framework in the commercial user agent optimizes the users’ electricity cost on the basis of the time-of-use electricity price signal in the electricity market, and ensures the visual comfort level of office staff within the commercial building is maintained. In summary, the novel energy management optimization theoretical framework based on the hybrid simulation of the electricity market bidding proposed in this paper demonstrates good application prospects and positive practical significance in the operation of commercial users.