Indoor thermal comfort and energy consumption multi-scale prediction method
Through the prediction method of superimposed multi-model fusion, combined with the Brushlet support vector machine, Shearlet neural network and XGBoost algorithm, the problem of difficulty in predicting indoor thermal comfort and energy consumption in the existing technology is solved, and more efficient energy management and better thermal comfort are achieved.
Patent Information
- Application Number
- CN202411638605.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-02-25
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to accurately predict indoor thermal comfort and energy consumption simultaneously, resulting in the problems of waste of energy and insufficient thermal comfort.
Using the prediction method of superimposed multi-model fusion, including Brushlet Support Vector Machine (BSVM), Shearlet Neural Network (SNN) and XGBoost Algorithm (XGBA), multi-scale prediction of thermal comfort and energy consumption is performed by building a meta-learner of the first and second network structures.
It effectively improves the accuracy and efficiency of indoor air conditioners' thermal comfort and energy consumption prediction, provides a theoretical basis for determining the optimal ventilation time and air volume, and achieves better energy-saving effects and thermal comfort.
Smart Images
Figure BDA0005137828610000026 
Figure BDA0005137828610000032 
Figure BDA0005137828610000033
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of indoor environment control, and in particular to a multi-scale prediction method for indoor thermal comfort and energy consumption. Background Art
[0002] With the rapid economic growth and the continuous improvement of living standards, people have higher and higher requirements for the thermal environment of buildings, and the requirements for indoor thermal comfort have also been further improved. The market space of the air-conditioning industry has been expanding, which will inevitably lead to potential energy waste. In addition, due to the different types and spaces of indoor spaces, people have different requirements for thermal comfort. If a unified cooling / heating standard is adopted, it will inevitably cause huge energy waste. In indoor air conditioning, the air-conditioning system usually consumes a lot of energy to maintain indoor thermal comfort. In order to achieve the optimization of thermal comfort and energy saving, the establishment of an effective air-conditioning load thermal parameter model can better reflect the relationship between thermal environment and energy consumption, which is the key prerequisite for achieving thermal comfort and energy saving optimization. In order to control the thermal comfort and energy consumption of indoor air conditioning, it is necessary to grasp the actual situation of building thermal environment and energy use in a timely and accurate manner. The prediction results can provide a theoretical basis for ensuring good thermal comfort and reducing energy consumption.
[0003] Thermal comfort and energy consumption prediction is difficult due to the diversity of indoor types, complexity of structures and randomness of people's use. From the existing research results, the established prediction model of thermal comfort of air-conditioning system has achieved high prediction accuracy. However, energy consumption prediction is not considered at the same time, and there are relatively few studies on indoor air-conditioning energy consumption prediction for thermal comfort, most of which are single prediction models, and have limitations such as limited improvement of accuracy and general generalization ability. Therefore, it is necessary to establish a prediction model that considers thermal comfort and energy consumption in view of the above problems. Summary of the invention
[0004] The purpose of the present invention is to provide a multi-scale prediction method for indoor thermal comfort and energy consumption, construct a prediction model of superimposed multi-model fusion, give full play to the advantages of a single prediction model, effectively grasp the thermal comfort and energy consumption of indoor air conditioning, and provide a suitable theoretical basis for determining the optimal ventilation time and air volume, so that the air-conditioning system can provide better thermal comfort for indoor personnel while achieving better energy-saving effects.
[0005] To achieve the above object, the present invention provides a multi-scale prediction method for indoor thermal comfort and energy consumption, comprising the following steps:
[0006] S1. Construct Brushlet support vector machine BSVM model;
[0007] S2. Design the Shearlet neural network SNN model;
[0008] S3, predict the results based on XGBoost algorithm XGBA;
[0009] Among them, the superimposed multi-model fusion is divided into a first network structure and a second network structure; the first network structure includes three models, namely, Brushlet support vector machine BSVM, Shearlet neural network SNN and XGBoost algorithm XGBA;
[0010] First, the prediction results obtained by the first network structure are input into the second network structure; secondly, the meta-learner in the second network structure uses a linear regression algorithm to integrate and converge the prediction results of the first network structure; finally, the prediction results of the second network structure are used as the output results of the entire model.
[0011] Preferably, in step S1, when constructing the BSVM model, the linear regression function of BSVM is as follows:
[0012]
[0013] The training samples are definition, is the n-dimensional input vector, y i is the corresponding output vector, n is the number of samples, is a nonlinear transformation mapping function, is the weight vector and d is the offset.
[0014] Preferably, the objective optimization function of BSVM is as follows:
[0015]
[0016] The constraints are as follows:
[0017]
[0018] In the formula, ρ represents the regularization term coefficient, ε i represents the relaxation factor.
[0019] Preferably, the Lagrange multiplier method is used to solve the minimum value of the function:
[0020]
[0021] Among them, β i is the Lagrange multiplier.
[0022] Preferably, the regression model of BSVM is as follows:
[0023]
[0024] Preferably, in step S2, a Shearlet neural network model is designed as follows:
[0025]
[0026] Among them, x 1 ,x 2 ,...,x n is the input variable, is the output variable, is the Shearelet basis function, used as the activation function of the hidden layer, ω ij is the weight between the input layer and the hidden layer, ω jk is the weight between the hidden layer and the output layer, μ i is the expansion factor, σ i is the translation factor, σ(i) is the output value of the i th node in the hidden layer;
[0027] Among them, the Shearlet neural network adopts a compact structure and is divided into three levels: input layer, hidden layer and output layer.
[0028] Preferably, in step S3, the original model of XGBA is as follows:
[0029]
[0030] in, is the predicted value, L is the number of trees, g l is a function related to the tree structure and leaf weight, x i is the i-th input data, and G is the function space.
[0031] Preferably, the objective function before XGBA iteration is as follows:
[0032]
[0033] in, is the training loss function, which is the fit of the model on the training set, y i is a real value, is the predicted value in iteration t, is the regularization term.
[0034] Therefore, the present invention adopts the above-mentioned multi-scale prediction method for indoor thermal comfort and energy consumption, constructs a prediction model of superimposed multi-model fusion, gives full play to the advantages of a single prediction model, effectively grasps the thermal comfort and energy consumption of indoor air conditioning, and provides a suitable theoretical basis for determining the optimal ventilation time and air volume, so that the air-conditioning system can provide better thermal comfort for indoor personnel while achieving better energy-saving effects.
[0035] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 It is a structural diagram of superimposed multi-model fusion of multi-scale prediction methods for indoor thermal comfort and energy consumption;
[0037] Figure 2 It is a schematic diagram of SNN for multi-scale prediction method of indoor thermal comfort and energy consumption;
[0038] Figure 3 is the running time of different prediction models in the embodiments of the present invention. DETAILED DESCRIPTION
[0039] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0040] like Figure 1 As shown, the present invention provides a multi-scale prediction method for indoor thermal comfort and energy consumption, comprising the following steps:
[0041] S1. Construct Brushlet support vector machine BSVM model;
[0042] S2. Design the Shearlet neural network SNN model;
[0043] S3, predict the results based on XGBoost algorithm XGBA;
[0044] Among them, the superimposed multi-model fusion is divided into a first network structure and a second network structure; the first network structure includes three models, namely, Brushlet support vector machine BSVM, Shearlet neural network SNN and XGBoost algorithm XGBA;
[0045] First, the prediction results obtained by the first network structure are input into the second network structure; secondly, the meta-learner in the second network structure uses a linear regression algorithm to integrate and converge the prediction results of the first network structure; finally, the prediction results of the second network structure are used as the output results of the entire model.
[0046] Example
[0047] 1. Overview of Thermal Comfort and Energy Consumption
[0048] Sitting is the main activity type of indoor personnel. PMV (predicted mean vote rate) can be used to accurately reflect the thermal comfort of sitting human body under the radiant cooling air conditioning system. The calculation formula of PMV is as follows:
[0049] PMV=(0.3e -0.04M +0.03)·L (1)
[0050] Among them, M is the human metabolic rate related to the exercise state, and L is the human heat load, that is, the difference between the human body's metabolic heat production rate and heat dissipation rate.
[0051] The human body can effectively regulate body temperature. In order to maintain a certain body temperature and ensure thermal comfort and thermal balance, the human body will undergo complex physiological regulation processes, such as sweat evaporation, blood vessel contraction or expansion, muscle tremors, etc. The thermal balance equation of the human body is as follows:
[0052] MW=D+C 1 +C 2 +E 1 +E 2 +R+S (2)
[0053] Among them, W is the mechanical work done by the human body on the outside world, D is the heat dissipation of the human body, and C 1 is the heat dissipation by human convection, C 2 is the convective heat dissipation due to breathing, E 1 E is the heat dissipation by evaporation from the human body. 2 R is the evaporative heat dissipation caused by breathing, R is the heat dissipation caused by human body radiation, and S is the heat storage rate of the human body.
[0054] The mechanical work W, convection heat dissipation D and heat storage rate S of the seated person can be ignored, and the heat load of the human body is as follows:
[0055] L=M-(C 1 +C 2 +E 1 +E 2 +R) (3)
[0056] Since different people have different thermal sensations indoors, the PMV value of each sampling point is also different. Therefore, in the process of calculating the average value of thermal comfort, the positive and negative PMV values of the sampling points may be offset, making the average value equal to or close to 0, thereby confusing the thermal comfort level. In order to avoid this confusion about indoor thermal comfort and accurately reflect the indoor thermal comfort, the total thermal comfort weight index combining OD (occupancy density) and PMV index is used to study the thermal sensation of human body indoors, which is called IPMV (integrated PMV). The calculation formula of IPMV is as follows:
[0057]
[0058] Among them, OD(x,y,z) is the density of people distribution, K is the number of indoor air-conditioned areas, and x, y, z represent the coordinates of a point in space in three directions.
[0059] To predict the percentage of individuals who would feel uncomfortable in an environment, the IPPD index is used to quantitatively predict the number of individuals who would feel uncomfortable as follows:
[0060]
[0061] In order to fully reflect the thermal comfort of the human body, this study adopted the comprehensive thermal comfort coefficient as follows:
[0062]
[0063] Where λ is the comprehensive coefficient of thermal comfort; the relationship between λ and human thermal sensation is shown in Table 1.
[0064] Table 1 Relationship between PMV and human thermal sensation
[0065]
[0066] According to the real-time nominal performance coefficient (COP) of the refrigeration host n ) Calculate the energy consumption of the air conditioning system. The real-time performance coefficient of the refrigeration host is as follows:
[0067] COP n =COP r COP a (7)
[0068] Among them, COP r is the real-time rated coefficient of performance of the refrigeration host, COP a is the real-time rated COP coefficient. The cooling capacity of the refrigeration host is as follows:
[0069] Q l =M·c p ·(T in -T out ) (8)
[0070] Where M is the medium mass flow rate of the refrigeration host, c p is the specific heat capacity of the medium, T in is the inlet temperature of the refrigeration host, T out Preset outlet temperature for the cooling unit.
[0071] The energy consumption of the air conditioning system is as follows:
[0072]
[0073] 2. Overlay Multi-Model Fusion Design
[0074] In order to improve the accuracy and efficiency of thermal comfort and energy consumption prediction, a machine learning algorithm based on stacked multi-model fusion is designed. The stacked multi-model fusion includes two layers, such as Figure 1 shown.
[0075] The superposition multi-model fusion is divided into two layers. The first network structure includes three models, namely Brushlet support vector machine (BSVM), Shearlet neural network (SNN) and XGBoost algorithm (XGBA). The prediction results obtained by the first layer model are input into the second network structure. The meta-learner in the second layer model uses the linear regression algorithm to integrate and converge the prediction results of the first layer model, and the prediction results of the second layer are used as the output results of the entire model.
[0076] S1. Construct Brushlet support vector machine BSVM model.
[0077] A set of training samples consists of definition, is the n-dimensional input vector, y i is the corresponding output vector. n is the number of samples, and the linear regression function of BSVM is as follows:
[0078]
[0079] in, is a nonlinear transformation mapping function, is the weight vector and d is the offset.
[0080] The target optimization function of BSVM is as follows:
[0081]
[0082] The constraints are as follows:
[0083]
[0084] Among them, ρ represents the regularization coefficient, ε i represents the relaxation factor.
[0085] The Lagrange multiplier method is used to solve the minimum value of the function as follows:
[0086]
[0087] Among them, β i is the Lagrange multiplier. By taking partial derivatives of expression (13), we can obtain the equation as follows:
[0088]
[0089]
[0090] By eliminating variables and εi , we can get the solutions of expressions (14)-(17) as follows:
[0091]
[0092] in, is the unit column vector, is the n-th order identity matrix, K(·) is the kernel function that satisfies the Mercer condition.
[0093] The regression model of BSVM is as follows:
[0094]
[0095] In order to improve the approximation ability of the kernel function, the Brushlet function is used as the kernel function, thereby improving the generalization ability of BSVM, as shown below:
[0096]
[0097] S represents the Brushlet wavelet mother function.
[0098] The Brushlet function has the advantages of localization, multi-level and multi-resolution, which can improve the prediction accuracy and iterative convergence speed of BSVM.
[0099] S2. Design a Shearlet neural network (SNN) model.
[0100] Shearlet neural network adopts a compact structure, such as Figure 2 As shown in Figure 2, SNN is divided into three layers: input layer, hidden layer and output layer. 1 ,x 2 ,...,x n is the input variable, is the output variable, is the Shearelet basis function, used as the activation function of the hidden layer, ω ij is the weight between the input layer and the hidden layer, ω jk is the weight between the hidden layer and the output layer, as shown below:
[0101]
[0102] Among them, μ i is the expansion factor, σ i is the translation factor, and σ(i) is the output value of the ith node in the hidden layer.
[0103] S3. Result prediction based on XGBoost algorithm XGBA.
[0104] XGBA can effectively improve the performance of gradient improvement. It is an improved boosting tree algorithm. The original model of XGBA is as follows:
[0105]
[0106] in, is the predicted value, L is the number of trees, g l is a function related to the tree structure and leaf weight, x i is the i-th input data, and G is the function space.
[0107] The objective function of XGBA before iteration is as follows:
[0108]
[0109] in, is the training loss function, which is the fit of the model on the training set, y i is a real value, is the predicted value in iteration t, is the regularization term.
[0110] XGBA initializes a prediction value and adds a new function in each iteration. After several iterations, it gradually generates a new model in the direction of reducing the residual error, so as to continuously reduce the residual error until the error requirement of the system is met and the purpose of improvement is achieved.
[0111] XGBA takes into account both speed and efficiency, adopts a parallel tree boosting method to speed up model learning, can run on different platforms and language environments, enhances nonlinear learning capabilities and scalability, and has significant advantages in efficiency and portability in many prediction problems and practical applications.
[0112] This study combines three algorithms: BSVM, SNN and XGBA, which have strong learning capabilities and are suitable for stacking fusion models. The proposed stacking fusion model has high prediction accuracy for small sample data sets, greatly shortens the training time of the model, and is feasible to apply it to thermal comfort and energy consumption prediction.
[0113] 3. Case Study
[0114] In order to verify the effectiveness of the proposed prediction model, an official building was selected as the research object. The input variables include indoor temperature (x 1 ), indoor humidity (x 2 ), outdoor wind speed (x 3 ), solar radiation (x 4 ), personnel presence rate (x 5 )、heart rate(x6 ), wrist skin temperature (x 7 ), heat transfer coefficient of external window (x 8 ) and roof insulation thickness (x 9 The output variables include the comprehensive thermal comfort coefficient (x 9 ) and energy consumption (P C ).
[0115] A university library was used as the research object. A fan coil fresh air conditioning refrigeration system was used, and the COP of the refrigeration unit was 3.6. The cooling tower was used for cooling. The temperature of the cooling hall was set at 27°C, and the temperature of other cooling rooms was set at 26°C. The maximum power of lighting and equipment was 12W / m 2 and 18W / m 2 In addition to the walls, the heat storage bodies of the library also include books, tables, chairs and bookshelves. The specific heat capacity of books is 1.9 kJ / (kg·K) and the density is 812 kg / m 3 The tables, chairs and bookshelves are all made of wood. The specific heat capacity of wood is 1.83 kJ / (kg·K) and the density is 540 kg / m 3 . 350 sets of thermal comfort and energy consumption data sets under different conditions were collected, the first 325 sets of data were used as training samples, and the last 25 sets of data were used as test samples.
[0116] The evaluation indicators of superimposed multi-model fusion include mean absolute error (MAE), mean square error (MSE), decision coefficient (R 2 ) and mean absolute percentage error (MAPE). The prediction accuracy of superimposed multi-model fusion is proportional to MAE, MSE, and MAPE. The closer the determination coefficient is to 1, the better the prediction performance of superimposed multi-model fusion.
[0117] The stacked multi-model fusion (SMMF) is trained based on the training samples, and the trained stacked multi-model fusion (SMFF) is applied to predict the thermal comfort and energy consumption of indoor air conditioning. In addition, BSVM, SNN and XGBA are used to predict the thermal comfort and energy consumption of the same test samples.
[0118] As shown in Table 2, the thermal comfort prediction results of the test samples based on different models are listed. Compared with other prediction models, the predicted value of λ based on the proposed SMMF is closer to the true value of λ, so the proposed SMMF has a better prediction effect on thermal comfort.
[0119] As shown in Table 3, the table lists the energy consumption prediction results based on different models. Compared with other prediction models, the energy consumption prediction value of SMMF is closer to the actual value, so the proposed SMMF also has advantages in energy consumption prediction.
[0120] The comparison results show that the proposed SMMF can effectively integrate different single prediction models and give full play to the advantages of single prediction models. SMMF has a higher prediction effect than single prediction models. SMMF can effectively grasp the thermal comfort and energy consumption of indoor air conditioning, and provide a suitable theoretical basis for determining the optimal ventilation time and air volume, so that the air conditioning system can provide better thermal comfort for indoor occupants while achieving better energy-saving effects.
[0121] Table 2 Prediction results of thermal comfort of test samples based on different models
[0122]
[0123] Table 3 Energy consumption prediction results based on different models
[0124]
[0125]
[0126] Table 4 Evaluation indicators of thermal prediction results of different models
[0127] Prediction Model MAE MSE MAPE / % <![CDATA[R 2 ]]> BSVM 0.154 0.178 0.185 0.926 SNN 0.098 0.089 0.092 0.965 XGBA 0.126 0.135 0.152 0.943 SMMF 0.025 0.038 0.035 0.997
[0128] Table 5 Evaluation indicators of energy consumption prediction results of different models
[0129] Prediction Model MAE MSE MAPE / % <![CDATA[R 2 ]]> BSVM 0.178 0.196 0.198 0.912 SNN 0.095 0.105 0.112 0.975 XGBA 0.139 0.157 0.174 0.953 SMMF 0.034 0.042 0.042 0.998
[0130] From Tables 4 and 5, we can see that the MAE, MSE, and MAPE of the proposed SMMF are smaller than those of BSVM, SNN, and XGBA. In addition, the R 2 It is closer to 1 than BSVM, SNA and XGBA, so the prediction accuracy of SMMF is better than BSVM, SN and XGBA.
[0131] like Figure 3 As shown in the figure, the training time of thermal comfort and energy consumption prediction of different prediction models, the running time of SMMF is much shorter than BSVM, SNN and XGBA, so SMMF has higher prediction efficiency than other single prediction models. Therefore, SMMF can effectively grasp the thermal comfort and energy consumption of indoor air conditioners, thereby obtaining better prediction results, which is of great significance for intelligent temperature control and energy saving and consumption reduction of indoor air conditioners.
[0132] Therefore, the present invention adopts the above-mentioned multi-scale prediction method for indoor thermal comfort and energy consumption, constructs a prediction model of superimposed multi-model fusion, gives full play to the advantages of a single prediction model, effectively grasps the thermal comfort and energy consumption of indoor air conditioning, and provides a suitable theoretical basis for determining the optimal ventilation time and air volume, so that the air-conditioning system can provide better thermal comfort for indoor personnel while achieving better energy-saving effects.
[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A multi-scale prediction method for indoor thermal comfort and energy consumption, characterized by: The following steps are involved: S1. Construct Brushlet support vector machine BSVM model; S2. Design the Shearlet neural network SNN model; S3, predict the results based on XGBoost algorithm XGBA; Among them, the superimposed multi-model fusion is divided into a first network structure and a second network structure; the first network structure includes three models, namely, Brushlet support vector machine BSVM, Shearlet neural network SNN and XGBoost algorithm XGBA; First, the prediction results obtained by the first network structure are input into the second network structure; secondly, the meta-learner in the second network structure uses a linear regression algorithm to integrate and converge the prediction results of the first network structure; finally, the prediction results of the second network structure are used as the output results of the entire model.
2. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 1, characterized in that: In step S1, when constructing the BSVM model, the linear regression function of BSVM is as follows: The training samples are definition, is the n-dimensional input vector, y i is the corresponding output vector, n is the number of samples, is a nonlinear transformation mapping function, is the weight vector and d is the offset.
3. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 2, characterized in that: The target optimization function of BSVM is as follows: The constraints are as follows: In the formula, ρ represents the regularization term coefficient, ε i represents the relaxation factor.
4. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 3, characterized in that: Use the Lagrange multiplier method to find the minimum value of the function: Among them, β i is the Lagrange multiplier.
5. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 4, characterized in that: The regression model of BSVM is as follows:
6. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 1, characterized in that: In step S2, the Shearlet neural network model is designed as follows: where x1,x2,...,x n is the input variable, is the output variable, is the Shearelet basis function, used as the activation function of the hidden layer, ω ij is the weight between the input layer and the hidden layer, ω jk is the weight between the hidden layer and the output layer, μ i is the expansion factor, σ i is the translation factor, σ(i) is the output value of the ith node in the hidden layer; Among them, the Shearlet neural network adopts a compact structure and is divided into three levels: input layer, hidden layer and output layer.
7. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 1, characterized in that: In step S3, the original model of XGBA is as follows: in, is the predicted value, L is the number of trees, g l is a function related to the tree structure and leaf weight, x i is the i-th input data, and G is the function space.
8. A multi-scale prediction method for indoor thermal comfort and energy consumption according to claim 7, characterized in that: The objective function of XGBA before iteration is as follows: in, is the training loss function, which is the fit of the model on the training set, y i is a real value, is the predicted value in iteration t, is the regularization term.