A Novel Method of BP Neural Network Based Green Building Design—The Case of Hotel Buildings in Hot Summer and Cold Winter Region of China

Abstract

With the advent of the big data era, architectural design gradually tends to become more quantified and intelligent. This study proposes a novel green design method for energy-saving buildings based on a BP neural network. This study changed the traditional trial–error mode by evaluating energy consumption based on design performance parameters such as building shape, space, and interface. Instead, energy consumption quota values obtained from statistical data, as well as thermal parameters and energy system parameters in energy-saving standards, were taken as input parameters, and then the design scheme of building shape can be obtained through BP neural network technology. Based on data of 61 hotel buildings in a representative city among a hot summer and cold winter climate zone, the BP neural network model is established to control the building design variables, with 41 kgce/m²·a as its energy-saving design target. Through the energy consumption quota, the trained BP network is applied to predict the optimal architectural design parameters, including the building orientation angle, shape coefficient, window–wall ratio, etc., for twelve building typologies in an area range of 5000~60,000 m². With recommended control thresholds of quantifiable architectural design elements obtained, this research can provide effective design decision-making suggestions for architects.

1. Introduction

The concept of green ecological architecture design has won increasing popularity. Moreover, the energy-saving concept is one of the important elements for green ecological architecture. At the same time, optimizing the architectural spatial form and functional layout can reduce the over-dependence of buildings on energy resources. As an important factor affecting building energy consumption, the ontological design of a building, such as its spatial form and functional layout, is the most direct and effective link to achieve building energy saving with the lowest cost. Meanwhile, it can create a green and healthy space environment for users. Therefore, the spatial form and functional layout need to change from extensive design mode to a new green design mode guided by professional, refined, and scientific thinking mode. In the era of big data, the traditional architectural design methods and design standards can only provide result-oriented static design parameters, which can no longer meet the requirements of scientific, refined, and rapid development of information technology for the discipline of architectural design. Data mining (DM) technology should be used in architectural design to build various element databases affecting architectural performance, obtain the hidden laws through data analysis, take the design objectives and requirements as to design input items, and build a quantitative dynamic design method.

DM is to automatically discover useful information in large data repositories and objectively discover hidden knowledge from data sets. Data mining refers to a process of extracting previously unknown potential useful information and knowledge hidden in a large number of incomplete, noisy, fuzzy, and random data. In short, it is the process of mining useful knowledge from a large amount of data stored in the database. Data mining is the combination of database technology, and machine learning technology, so many methods are similar to machine learning methods. The commonly-used data mining methods mainly include association analysis, linear regression, classification analysis, cluster analysis, neural network method. During the process of green building design, it is necessary to extract the useful knowledge contained in the huge data and information quickly and efficiently as the input value of design parameters. With the accumulation of building energy consumption data, data mining can play an increasingly prominent role in energy consumption analysis [1]. We can make use of the advantages of machine learning in performance prediction, carry out DM, prediction, and evaluation through machine learning methods, and introduce intelligent optimization technology to find the optimal solution in group objectives, thereby establishing the influence mechanism, prediction model, an optimization model of performance indicators and design elements, and realizing the synchronization, efficiency and accuracy of multi-performance indicator prediction and multi-objective optimization, and achieving the feedback type decision model of green building.

An artificial neural network (ANN), especially the BP neural network (BPNN), is a common method in DM, which has been widely used in the fields of energy consumption index evaluation and energy consumption prediction. At present, the application of the intelligent algorithm and parametric design in building performance optimization and design decisions has become an important research direction in the field of architectural design at home and abroad. Kalogirou et al. [2] used an artificial neural network to predict the energy consumption of passive solar buildings and obtained an analysis model for faster prediction of building energy consumption. Olofson et al. [3] established a building energy consumption prediction model by combining an ANN with short-term test data and achieved satisfactory results. Many research results [4,5] also confirm that the ANN model has high accuracy in dynamic real-time prediction. Lei Yarong [6] predicted the energy consumption of residential buildings in Chongqing by analyzing the factors affecting the energy consumption of residential buildings in Chongqing and establishing a BP neural network. Yu Wei [7] analyzed the relationship between influencing factors such as building form and building thermal parameters and building energy consumption and indoor comfort and established a multi-objective BPNN energy consumption prediction model. Taking office buildings as an example, Lin Borong et al. [8] proposed forward calculation and reverse optimization process of building energy conservation optimization at the initial stage of design and analyzed how to reverse design optimization and parameter sensitivity analysis of building shape and space scheme with the overall energy consumption as the goal. Gossard D et al. [9] took French housing as an example and adopted the method of combining an ANN and NSGA-II genetic algorithm to optimize the heat transfer coefficient of each part of the envelope under the dual objective orientation of building annual energy consumption and indoor thermal comfort. Kusiak A et al. [10] proposed an optimization method for office building heating ventilation and air conditioning system. The neural network was applied in establishing a prediction model, and the intensity multi-objective particle swarm optimization algorithm was used to optimize the prediction model so as to keep the thermal comfort acceptable and reduce energy consumption to the minimum. Magnier L et al. [11] used an ANN based on a simulation to characterize the running status of buildings and then combined the ANN with the NSGA-II algorithm to carry out optimization research on residential thermal comfort and building energy consumption. Marcel Macarulla [12] proposes a predictive control strategy based on a neural network in a commercial building energy management system for boilers at the optimum time. Young Tae Chae [13] proposes a short-term building energy usage forecasting model based on an ANN model. Seunghui Lee [14] employs an ANN to predict user-based energy consumption. Qi Dong [15] develops an ANN model to predict the energy consumption and cost of cross-laminated timber office buildings in severe cold regions during the early stage of architectural design. Rafael Pino-Mejías [16] and Le Thi Le [17] use an ANN to predict the heating and cooling energy demands and energy consumptions of buildings. Many researchers [18,19,20,21] have summarized the application of ANNs in building energy consumption, which shows that ANNs can well and flexibly solve a large number of applications related to building energy, including load forecasting and energy consumption forecasting, building equipment system modeling, and energy system management.

The research on the relationship between building form design and the energy consumption is very extensive, and a certain theoretical consensus has been formed on the influencing parameters of building energy consumption form. Liu Jiaping [22] believes that building plane layout, shape coefficient, and building shape are related to energy consumption. Houcem Eddine Mechri [23] believes that parameters such as shape coefficient, building orientation, external shading, surface area coefficient, and building internal heat capacity are related to building energy consumption. Li Ziwei [24,25,26] puts forward that the orientation of windows, window–wall ratio, window width, and window height have a great influence on the solar radiation heat gain and heat loss of buildings.

It can be seen from the above research that the intelligent algorithm has strong applicability in the dynamic prediction of building energy consumption. At present, there are two main research directions in this field: establishing an intelligent prediction model of energy consumption by analyzing the parameters that affect building energy consumption and obtaining design parameters by the intelligent algorithm and reverse optimization under certain design objectives. For architects, building energy consumption, as a measure of energy-saving buildings, is only the result after architectural design, and the fundamental task of architects should be to make building energy consumption as the reason of architectural design, input it into the design process, form an architect-led design mode, form an external architectural form constrained by energy consumption and green performance, and achieve a comprehensive and optimal architectural design scheme. Therefore, the objectives of building energy efficiency design are an important constraint. The key dynamic design parameters are formed through disassembly and then transformed into the design element parameters required by architects through the intelligent algorithm model so as to provide scientific analysis and decision-making technology for the decision-making of the most important external design scheme of building ontology design.

According to the shortcomings and gaps of the current research, this paper finally proposes the optimal design scheme and the green design objectives, which can effectively meet the initial setting based on intelligence technology. In other words, through the BPNN of building energy consumption database, this novel method can establish a design experiment and analysis model, taking the building energy consumption and design requirements as the input parameters of the intelligent algorithm model, to analyze and obtain the optimized key parameters related to building shape design.

2. Energy-Saving Building Design and Research Methodology

2.1. Design Method and Concept of Green Energy-Saving Building Based on DM Technology

The conventional architectural design scheme is formed in the traditional green and energy-saving building design method according to the owner’s needs, site conditions, functional requirements, and relevant architectural design standards. According to the national and local constraints and requirements on green energy conservation, the energy-saving optimization scheme is formulated, and the corresponding technical parameters are determined in combination with the architectural design scheme, and then the green building technology implementation scheme is finally obtained, including the energy-saving rate, level of green building, building shape parameters, thermal parameters, energy resource system parameters, operation, and maintenance, etc.

Traditional green and energy-saving building design methods do not effectively take the statistical data of building energy consumption as an important design constraint target. In the era of big data, the design of green energy-saving buildings should meet the requirements of refinement and quantification. The indispensable need of building energy consumption data is used as the quantitative factor of building design. The local building energy consumption quota obtained through statistics is used as the design input parameter to guide architects to make scientific decisions and generate design schemes. Aiming at the research objective, this paper puts forward a new method of green energy-saving building design based on DM technology. The new design method is divided into two stages, the data analysis stage and the design stage.

In the data analysis stage, the project information database is established through monitoring and collecting the data of similar projects that have been built and operated and after basic data processing. Then the algorithm model of DM is established to provide support for DM. Data types mainly include meteorological parameters, basic building information, building energy consumption, building shape parameters, building thermal parameters, building operation data, and building energy resource system data.

It should be pointed out that the DM model in Figure 1 not only refers to the BPNN model but also includes association analysis, linear regression, classification analysis, clustering analysis, and neural network method. These technologies have many applications in the field of architecture. The BPNN model is used in this study because the BPNN has been the most widely used method compared with other methods. BP neural network has not only pretty good prediction accuracy in the prediction of large samples but also good prediction effect in the prediction of small samples. Its strong nonlinear fitting ability can make it better reflect the mapping relationship among complex nonlinear variables, with a flexible change of input and output variables [19,20,21].

2.2. Process of DM Based Novel Green Energy-Saving Building Design Method

According to the above research basis, taking the core tasks of architectural design and the thinking characteristics of architects as the research benchmark, a new green, and energy-saving building design program based on DM technology is proposed, as shown in Figure 2.

From the perspective of architects’ design thinking, at the beginning of architectural design, the building energy consumption quota can be obtained through the statistical analysis of design demand and building energy consumption data, and then the design parameters can be selected according to relevant design standards and design objectives. The design objectives and design parameters are input into the intelligent algorithm model as input items. The design scheme obtained through DM technology and experimental analysis can be compared and analyzed with relevant design standards so that the design scheme can meet the existing design standards and can improve the shortcomings of design standards to a certain extent.

2.2.1. Design Objective

The design objective is to statistically analyze the energy consumption statistical indicators of various local buildings according to the building energy consumption values in the basic data, obtain the building energy consumption quota through the analysis model, and put forward reasonable design target parameters through organically combining with the local energy-saving standards and typical buildings.

It should be pointed out that the building energy consumption is a very important constraint value in an architect’s design process, not just a single design goal. Through the statistical analysis of building energy consumption data, the local energy consumption quota of the building has been obtained. The energy consumption quota as input parameters, the intelligent algorithm model, has been established to obtain the optimized building shape design scheme.

2.2.2. Design Parameter Screening and Processing

Design parameter screening refers to classifying the building parameters in the basic data sources, which can be divided into five categories: meteorological parameters, building shape parameters, thermal parameters, energy system parameters, and operation parameters. Meteorological parameters can be selected according to the area where the architectural design project is located. The index system of architectural design parameters is shown in

Table 1. Index system of architectural design parameters.

Category	Name	Performance Index
		Index	Implication	Unit
Influence Factor	building shape parameters	building area x1	—	m2
		building orientation angle x2	The angle between the long side of the building and the east-west axis
		shape coefficient x3	The ratio of the surface area (excluding the ground) of a building in contact with the outside atmosphere to the volume enclosed	m−1
		window–wall ratio (south, north, east, west) x4, x5, x6, x7	The ratio of the total area of external windows (including transparent curtain wall) in a certain orientation to the total area of walls (including window area) in the same orientation	—
	thermal parameters	exterior wall heat transfer coefficient x8	—	W/m2.K
		external window heat transfer coefficient x9	—	W/m2.K
		solar heat gain coefficient of external window x10	—
		roof heat transfer coefficient x11	—	W/m2.K
	operating parameters	personnel density x12	—	m2/p
		average annual occupancy rate x13	—	%
	energy system parameters	integrated partial load performance coefficient of refrigerating machine x14	Obtained through calculation based on the performance coefficient value of unit under partial load, according to the weighted factor of unit running time under various loads.	—
		average efficiency of heating system x15	—	%
Comprehensive Indicator	building energy consumption	annual energy consumption of buildings y	Including energy consumption of heating, air conditioning, lighting and domestic hot water	kgce/m2.a

. In the subsequent algorithm models, some architectural design parameters are used as input parameters and some as output parameters. Different projects adopt different design parameters when establishing an intelligent algorithm model.

In the process of architectural design, the shape parameters related to building energy consumption mainly include the shape coefficient, the window–wall ratio of each façade, and the building orientation, which can directly affect the site layout, architectural form, spatial and functional layout, and the design of an architectural interface. According to the definition of shape coefficient, the larger the shape coefficient is, the larger the external area corresponding to the unit building volume will be, the greater the heat transfer loss of the outer envelope will be, and the more energy consumption will be, and vice versa. It can be seen that the size of the shape coefficient has a significant impact on building energy consumption. The window–wall ratio not only has a great impact on building energy consumption but also is an important factor to be considered in building appearance design so that its value should be considered comprehensively in architectural design [27,28,29]. The amount of heat gained from solar radiation received by buildings is related to the building orientation, which is very important for the cooling load of buildings [28].

Therefore, building shape coefficient, window–wall ratio, and building orientation are taken as parameters that need to be decided and output by other parameters in architectural design. Moreover, other parameters, including building energy consumption, are taken as input parameters.

2.2.3. BPNN Algorithm Model

In this study, a BPNN is used to establish the algorithm model. The BPNN belongs to a feed-forward neural network. The BPNN is the most widely-used neural network at present. The prediction result has reliability and credibility by using a BPNN with a relatively mature algorithm and model. Some research results have shown that BPNNs not only have good accuracy in the prediction of large samples but also have a good prediction effect for small samples [30,31,32].

At present, the existing research in the field of building energy consumption refers to a neural network model established according to the actual physical model, that is, in order to maintain the comfort of building indoor environment, the cooling and heating loads caused by internal and external disturbances are eliminated through the operation of building energy system, resulting in building energy consumption. Therefore, this kind of research can be summarized as follows: with the factor parameters affecting building energy consumption as the input parameters and building energy consumption as the output parameters, it is a neural network model structure based on the logic of building physical process, as shown in Figure 3.

Regardless of the actual physical model, the relationship among data in this paper is only viewed from the perspective of data processing, with building energy consumption as the input parameters and part of building shape parameters as the output parameters. From the perspective of architectural design, the corresponding relationship among building orientation, shape coefficient, and window–wall ratio obtained from these data should not be determined only from architectural design standards. Building energy consumption statistics should also be involved in the decision-making process of building shape parameters. The neural network model based on the BPNN is shown in Figure 4.

Among the input parameters in Figure 4, thermal parameters include external wall heat transfer coefficient x₈, external window heat transfer coefficient x₉, solar heat gain coefficient x₁₀, and roofing heat transfer coefficient x₁₁. Using status includes personal density x₁₂ and average annual occupancy rate x₁₃. Energy system efficiency includes integrated partial load performance coefficient of refrigerating machine x₁₄ and average efficiency of heating system x₁₅. The output parameter window–wall ratio includes the window–wall ratio in the south, north, east, and west directions, namely x₄, x₅, x₆, x₇.

The establishment of the BP network model (BPNM) is mainly divided into the following steps:

(1)

Randomly select a training sample, sort the total sample data, and then select a training set and a verification set.

(2)

Normalize the sample set. The input data matrix and target data matrix in the training set are normalized. After the normalization of all variable data, the value of the processed variable is in the range of [−1, 1].

(3)

The logsig transfer function is used to transmit data between the input neurons of the model and the hidden layer neurons. The output function of the i^th neuron of the hidden layer is shown in Function (1).

$$y_i=\log sig(W_i^{T}X+b{1}_i)$$

(1)

where $y_i$ is the output function of the i^th neuron in the hidden layer, $W_i^T$ is the network weight matrix between the input layer and the hidden layer, $X$ is the input parameter matrix, $b{1}_i$ is the network calculation deviation between the input layer and the hidden layer, and i is the number of the hidden layer neurons.

The hidden layer and the output layer are connected by purelin transfer function, with the output function of the j^th neuron in the output layer shown in Function (2).

$$o_j=purelin(V_j^{T}Y+b{2}_j)$$

(2)

where $o_j$ is the output function of the j^th neuron in the output layer, $V_i^T$ is the network weight matrix between the hidden layer and the output layer, $b{2}_j$ is the network calculation deviation between the hidden layer and the output layer, and j is the number of the hidden layer neurons.

(4)

The negative gradient descent principle is used to adjust the weights. The self-learning model is shown in Function (3).

$$\Delta W_i\left(n+1\right)=\eta {\delta }_{i}X+\alpha \Delta W_i\left(n\right)$$

(3)

where $\eta$ is the learning rate, ${\delta }_i$ is the calculation error of node i, and $\alpha$ is the momentum factor.

3. Results and Discussion

3.1. Data Statistics of Building Parameters

3.1.1. Building Energy Consumption

This study tested the energy consumption values of 61 hotels in a city with hot summer and cold winter. Among them, 49 hotels were distributed between 10,000~40,000 m², accounting for about 80.0%. The proportion of four-star hotels was about 70.5%, with the building areas from 6000 to 90,000 m². The energy consumption values of 61 buildings were sorted from the smallest to the biggest, and then the buildings were labeled. The building numbers were used for the statistical data of other design parameters. The distribution of building energy consumption is shown in Figure 5.

The building quota value of the common method in the industry is achieved from a large number of building energy consumption data of the city or region through measurement, by finding the benchmark value, constraint value, and guide value of building energy consumption quota in the region in ascending data distribution [33,34,35]. These indicators can be defined according to the actual situation. For example, the median value is generally taken as the benchmark of the energy consumption quota in the region in China [36].

As can be seen from Figure 5, the building energy consumption of sample hotels in the city varies greatly, with the minimum energy consumption per unit building area 16.1 kgce/m², the maximum 85 kgce/m², the average energy consumption per unit area 45.3 kgce/m², the median energy consumption of sample buildings is 41 kgce/m², the energy consumption of the lower quartile 37 kgce/m², and that of the upper quartile is 55 kgce/m². Under normal circumstances, the lower quartile of sample data can be considered as the guide value of energy consumption quota in this area, the median value as the reference value, and the upper quartile as the constraint value. However, many hotel buildings in this area were built early, without energy-saving standards when they were designed, resulting in high energy consumption. Therefore, in this paper, the average energy consumption of 45 kgce/m² was taken as the constraint value of the energy-saving design of hotel buildings in the city, the median value of 41 kgce/m² as the target value of energy-saving design, and 37 kgce/m² as the guiding value of energy-saving design.

3.1.2. Building Shape Parameters

There are many factors affecting the energy consumption of hotel buildings, including the thermal parameters of the building envelope, window–wall ratio, shape coefficient, building orientation, personnel density in the buildings, occupancy rate, building energy system efficiency, and control strategy. Therefore, these parameters are counted in the data statistics and monitoring of sample buildings in the city.

The distribution of shape coefficient and orientation angle of hotel buildings in the city with the hot summer and cold winter area is shown in Figure 6. It can be seen from Figure 5 and Figure 6 that the building shape coefficient has a certain relationship with the building area. The larger the area is, the smaller the shape coefficient will be.

The statistical distribution of window–wall ratio is shown in Figure 7. It can be seen from Figure 7 that the south-facing window–wall ratio is larger than that in other directions, and the window–wall ratio in the south-north direction is larger than that in the east-west direction. In addition, the larger the window–wall ratio is, the greater the energy consumption per unit area will be. It can be seen from Figure 5 and Figure 6 that the building shape coefficient has a certain relationship with the building area. The larger the area is, the smaller the shape coefficient will be.

3.1.3. Building Thermal Parameters

Thermal parameters of a building envelope include heat transfer coefficient of the exterior wall, heat transfer coefficient of an exterior window, roofing heat transfer coefficient, and shading coefficient of exterior window. The distribution of thermal parameters of the envelope of the sample hotel buildings in this city is shown in Figure 8. It can be seen from Figure 8 that the distribution curve of building energy consumption per unit area is consistent with that of thermal parameters of the building envelope. The larger the thermal parameters are, the greater the energy consumption per unit area will be.

3.1.4. Building Occupancy and Energy System Efficiency

According to the investigation and monitoring data, the annual average occupancy rate of the sample hotel buildings in the city was between 55% and 65%, with little fluctuation. As it is a tourist city, the distribution of tourists throughout the year is relatively uniform, and there is no particularly concentrated time period. Therefore, the occupancy rate of sample hotel buildings in the city remains relatively stable throughout the year.

After investigation and monitoring, the average efficiency of the heating system of the sample hotel buildings in the city was distributed between 0.8~0.92, and the integrated partial load performance coefficient of the refrigerating machine was between 4.9 and 5.9. There is a relatively high efficiency of the comprehensive energy system, with little difference among the buildings.

3.2. BP Construction of BPNM

According to the output parameters and input parameters of the network model, a three-layer BPNM was established. There were ten input parameters and six output parameters. After repeated experiments, the number of hidden layer neurons was set as 12. In the BP algorithm, the error accuracy was set as 1 × 10^−4, the learning rate of function selection 0.035, the target error to be achieved by the network 1 × 10⁻⁵, and the maximum number of iterations 50,000.

The performance parameters and energy consumption data of 61 buildings in the city with hot summer and cold winter were obtained through investigation and measurement. Fifty groups of training samples were randomly selected, with the remaining 11 groups of data as test samples. The building numbers of test samples were 13, 14, 20, 27, 28, 32, 33, 44, 50, 55, and 57. When the number of training processes reached 2088, the mean square error was 1 × 10⁻⁵, and then the function converged, and the training ended, as shown in Figure 9. The first 200 times of training were listed in Figure 9. As can be seen from the figure, the mean square error has reached below 1 × 10⁻² when the training has been carried out more than 200 times, and the mean square error has decreased rapidly from the beginning of training to 40 times.

A total of 11 groups of data of test samples were used to verify the model. The errors between output data and measured data were verified by ten parameter values, including input parameters x₁, x₈~x₁₅, y, and six output parameters, including x₂~x₇, as shown in Figure 10, Figure 11, Figure 12, Figure 13, Figure 14 and Figure 15.

As can be seen from Figure 10, Figure 11, Figure 12, Figure 13, Figure 14 and Figure 15, through the testing and verification of the samples, there are small errors between the values of all model output parameters and the values of actual test data, with the maximum error of 8.6% and the minimum error of 1.2%. The test results reveal that the BP network training result is ideal so that the trained model can be used as the energy-saving design of hotel buildings in the city.

3.3. Input Parameter Configurations of Energy Prediction Model

According to the statistical values of tested energy consumption data, the constraint value of hotel building energy efficiency design in the city was 45 kgce/m², with its target value 41 kgce/m², and the target value of higher standard 37 kgce/m². In order to reflect the relationship between building energy consumption and output parameters, the building energy consumption values were set as five values in 37~45 kgce/m².a.

Based on the thermal parameter limit of the public building envelope in regions with hot summer and cold winter according to the Design Standard for Energy Efficiency of Public Buildings GB 50189-2015 [28], the thermal parameter values of the envelope and the performance parameter values of the energy system were set. The values of heat transfer coefficient of exterior window x₉ and solar heat gain coefficient x₁₀ were related to the window–wall area ratio. If the predicted window–wall ratio does not meet the requirements of the design standard, the group of values will be discarded, and the value will be adjusted again until the window–wall ratio meets the requirements. According to the survey data, the average personnel density of hotel buildings was 13 m²/person, and the average annual occupancy rate x₁₃ was 65%. The model input parameter data is shown in

Table 2. Model input parameter data list.

Building Area (m2) x1	Heat Transfer Coefficient of Exterior Wall (W/m2.K) x8	Heat Transfer Coefficient of Exterior Wall (W/m2.K) x9	Solar Heat Gain Coefficient of External Window x10	Roofing Heat Transfer Coefficient (W/m2.K) x11	Personnel Density (m2/p) x12	Average Annual Occupancy Rate (%) x13	Integrated Partial Load Performance Coefficient of Refrigerating Machine x14	Average Efficiency of Heating System (Primary Energy) x15	Annual Energy Consumption per Unit Area (kgce/m2.a) Y
5000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 45,000 50,000 55,000 60,000	0.6	2.0	0.26	0.4	13	65	5.9	0.9	45 43 41 39 37

.

3.4. Impact Validation Based on Energy Prediction Model

3.4.1. Model Output Parameters

By inputting the data in Table 2 into the BPNM, the output data of building orientation angle, shape coefficient, and window–wall ratio in each direction can be obtained. According to different building areas and energy consumption data, the design parameters of buildings under different building area requirements and energy-saving targets can be obtained. Since the thermal parameters of the building envelope and the performance parameters of the energy system are the minimum requirements specified in the national standards, the architectural design parameters output by the model also meet the index values under the minimum requirements. When other parameters change, the architectural design parameters need to make corresponding design changes conducive to energy efficiency. Therefore, in this paper, only the values of different building areas and energy-saving objectives are discussed. The building design parameter values corresponding to different building areas and energy-saving targets are shown in Figure 16, Figure 17, Figure 18, Figure 19, Figure 20 and Figure 21.

3.4.2. Building Orientation Angle

It can be seen from Figure 16 that for buildings with an area of 5000~60,000 m², the building orientation angle is mainly concentrated in the range of 100~200°. The angle range indicates that the building orientation should be south by west. From the perspective of solar heat gain, the main direction of the building is southwest. Compared with buildings in the southeast direction, it is more beneficial to reduce the amount of solar radiation entering the building and delay the time of forming cooling load in the building due to the solar heat gain.

Under the same building energy consumption index, the orientation angle can increase with the increase in building area. From the result, it can be observed that the larger the building area is, the closer the orientation angle will be to 180°, and the closer the building orientation will be to the north-south direction. In the same building area, the orientation angle can increase with the decrease in energy consumption index value. The reason for this result is that the main direction of the building is due south, with the least amount of solar radiation entering the building.

3.4.3. Building Shape Coefficient

It can be seen from Figure 17 that the building shape coefficient increases with the increase in building area, which is determined by the definition of shape coefficient. In the same building area, the building energy consumption index increases, but the shape coefficient decreases because the heat transfer through the building envelope increases with the increase in the external area of the building in contact with the air. Therefore, in the architectural design, the external area of a building with the same area should be reduced as much as possible.

3.4.4. Window–Wall Ratio

The window–wall ratio is a design parameter that architectural design engineers have a lot of room to play. From the perspective of energy conservation, the smaller the window–wall ratio is, the more beneficial it will be to building energy conservation, which is fully proved in Figure 18, Figure 19, Figure 20 and Figure 21. Meanwhile, it can also be seen from the figures that under the same building energy consumption index, the larger the building area is, the smaller the window–wall ratio should be selected. Under the same building energy index values, the window–wall ratio in the south direction is smaller than that in the other three directions. According to the Design Standard for Energy Efficiency of Public Buildings GB 50189-2015 [27], different exterior window heat transfer coefficients and solar heat gain coefficients are adopted for different window–wall ratios. Similarly, if the window–wall ratio of a building is relatively large, it will be necessary to select a smaller exterior window heat transfer coefficient and solar heat gain coefficient.

3.4.5. Suggestions on Building Shape Design Parameters under Energy-Saving Design Objective

On the basis of the above analysis, in order to achieve the city’s building energy conservation target of 41 kgce/m².a, with the thermal parameters of the building envelope and the energy system performance indicators selected according to the Design Standard for Energy Efficiency of Public Buildings GB 50189-2015, as well as the parameters, such as building personnel density and occupancy rate, obtained from investigation and testing, the range of design parameters for different building areas, namely building shape coefficient, orientation angle and window–wall ratio, can be predicted by the BPNN model. Based on the recommended design values of building shape parameters, the design scheme can be optimized so that the building energy consumption can reach the target value.

Table 3. Suggested values of shape parameters under energy-saving target value.

Building Area m2		5000	10,000	15,000	20,000	25,000	30,000	35,000	40,000	45,000	50,000	55,000	60,000
Shape coefficient		≤0.18	≤0.15	≤0.14	≤0.13	≤0.12	≤0.11	≤0.10	≤0.09	≤0.08	≤0.07	≤0.06	≤0.05
Orientation angle		100~180	120~180	130~180	140~180	150~180	160~180	170~180	170~190	170~200	180~200	180~200	180~210
Win dow–wall ratio	South	0.46	0.38	0.35	0.30	0.25	0.2	0.15	0.1	0.08	0.07	0.06	0.05
	North	0.62	0.58	0.55	0.43	0.40	0.33	0.22	0.20	0.16	0.11	0.08	0.07
	East	0.52	0.45	0.40	0.35	0.3	0.25	0.2	0.15	0.12	0.10	0.07	0.05
	West	0.52	0.47	0.42	0.35	0.3	0.25	0.2	0.15	0.12	0.10	0.07	0.05

shows the recommended values of shape parameters under the energy-saving target value.

4. Conclusions

This study put forward and set up an energy-saving building design method based on DM technology. The objective of the method is to change the design method of the trial–error mode of green building design by using the building energy consumption value and the building thermal parameters and energy system parameters stipulated in the existing energy-saving standards to deduce the building shape design parameters. Specifically, through the statistical analysis of design requirements and building energy consumption data, architects play a leading role in obtaining the design objective of building energy conservation, establishing an intelligent algorithm model, putting the design objective and the selected input and output design parameters into the algorithm model, and then obtaining the required design scheme according to the relevant design standards.

Based on the established design method of BPNN technology, through the specific case of the city in hot-summer and cold-winter area, this study demonstrated the design process according to statistical data, established a BPNN model with building energy consumption, thermal parameters, energy system performance parameters and service performance parameters as input, and building shape parameters as output. Through training and testing of a large number of data, the ideal prediction model was obtained. Aiming at the building energy-saving target of 41 kgce/m².a, the trained BP network was applied to predict the architectural design parameters of 12 buildings with an area of 5000~60,000 m². Afterward, the design parameters, such as orientation angle, shape coefficient, and window–wall ratio, were successfully obtained, and the design suggestions were given to provide quantitative design strategies for architects.

Based on the statistical data of building energy consumption, the paper can obtain the local building energy consumption quota and take the energy consumption quota data as the input data. Through the design method of BPNN technology, the quantitative design conditioned by building energy-saving objectives can be realized. In the future energy-saving building design, architects can apply the design method in this paper to form the database of various building types in each climate area, establish the DM model and design templates, and then set up a convenient and accurate look-up table, thus providing a methodology for the intelligent and detailed design of buildings and a powerful means for architectural design decision-making.