Building window-to-wall ratio strategy prediction model construction method and window-to-wall ratio strategy prediction method

Abstract

The invention is applicable to the field of building technology, and discloses a method for constructing a building window-to-wall ratio strategy prediction model and a window-to-wall ratio strategy prediction method. The prediction method comprises providing a building window-to-wall ratio strategy prediction model, acquiring meteorological characteristic data at a time to be predicted; inputting the acquired meteorological characteristic data into the building window-to-wall ratio strategy prediction model to obtain an optimal window-to-wall ratio strategy; the method for constructing the building window-to-wall ratio strategy prediction model comprises performing data preprocessing on the collected building window-to-wall ratio strategy data and meteorological characteristic data; dividing the processed data into training data and data to be predicted; training a multilayer perceptron classifier using the training data, and constructing a multilayer perceptron classifier model according to the trained weight coefficients and biases as the building window-to-wall ratio strategy prediction model. The prediction method has high accuracy and can effectively help the building's external envelope structure realize intelligent automatic control of the window-to-wall ratio.

Description

Technical Field

[0001] The present invention relates to the field of building technology, and in particular to a building window-to-wall ratio strategy prediction model and a window-to-wall ratio strategy prediction method. Background Art

[0002] The window-to-wall ratio is an important factor affecting the energy consumption of the building itself. Reasonable window-to-wall ratio design can reduce the building energy consumption as much as possible while meeting the requirements of ventilation and lighting. At present, traditional building design adopts a static design method to determine the optimal window-to-wall ratio based on the annual energy consumption demand.

[0003] However, there are obvious differences in the requirements for the window-to-wall ratio in different periods of construction. For example, to improve the indoor thermal environment in winter, it is necessary to increase the window-to-wall ratio to increase the heat gain from solar radiation; to reduce the air-conditioning load in summer, it is necessary to reduce the window-to-wall ratio to reduce heat gain while ensuring daylighting. With the rapid development of energy-saving technologies and the improvement of energy-saving requirements, static design methods are increasingly unable to meet the needs of increasingly intelligent buildings. Dynamic, intelligent, and variable enclosure structures that can adapt to the thermal comfort needs of different people are worthy of analysis and research.

[0004] For the adjustment of variable enclosure structures, it is particularly important to be able to more accurately predict the optimal window-to-wall ratio strategy of the building's external enclosure structure. However, in the prior art, the prediction accuracy of the method used to predict the window-to-wall ratio is low, and the accuracy of the prediction needs to be improved. Summary of the invention

[0005] The first object of the present invention is to provide a strategic prediction model for the window-to-wall ratio of a building, which aims to solve the technical problem that the methods for predicting the window-to-wall ratio in the prior art have low prediction accuracy.

[0006] To achieve the above object, the solution provided by the present invention is:

[0007] The method for constructing a strategic prediction model for building window-to-wall ratio includes:

[0008] Preprocess the collected building window-to-wall ratio strategy data and the corresponding historical meteorological characteristic data;

[0009] Divide the processed data into training data and data to be predicted;

[0010] A multilayer perceptron classifier is trained using the training data, and a multilayer perceptron classifier model is constructed according to the trained weight coefficients and biases as a prediction model for the building window-to-wall ratio strategy;

[0011] The training methods for the Multilayer Perceptron classifier include:

[0012] Construct a multilayer perceptron classifier. The input layer of the multilayer perceptron classifier inputs the meteorological feature data, the output layer outputs the window-to-wall ratio strategy, and the number of neurons in the hidden layer is determined according to the following relationship:

[0013] J=2a+1

[0014] In the formula, J represents the number of neurons in the hidden layer, and a represents the number of neurons in the input layer;

[0015] Initialize the learning rate η0, so that η0∈(0,1), and randomly generate the initial value w0 of the weight coefficient and the initial value b0 of the bias of each layer of neurons;

[0016] Starting from the input layer, forward propagation is performed through the network, using the Sigmoid activation function to activate neurons in the hidden layer and the identity function to activate neurons in the output layer;

[0017] Calculate the error and use the training data to update the weight coefficient w of each layer through the back propagation algorithm t and the offset b t .

[0018] Preferably, the method for constructing the building window-to-wall ratio strategy prediction model also includes inputting the data to be predicted into the building window-to-wall ratio strategy prediction model, and measuring the building window-to-wall ratio strategy prediction model using the accuracy index and the confusion matrix. If the accuracy is greater than 80% and the confusion matrix is ​​greater than 70%, it means that the building window-to-wall ratio strategy prediction model has a high accuracy.

[0019] Preferably, the Sigmoid activation function is expressed as:

[0020]

[0021] In the formula, f(a j ) represents the probability distribution of the j-th predicted window-to-wall ratio strategy category output by the hidden layer, a j It represents the jth input of the hidden layer, that is, the product of the meteorological factor data and the weight plus the bias value. A and B are parameters, A∈(0.01, 100), B∈(1, 100).

[0022] Preferably, the meteorological characteristic data include solar irradiance, dry-bulb temperature and relative humidity, the window-to-wall ratio strategy includes three strategy categories, the number of neurons in the input layer of the multilayer perceptron classifier model is 3, the number of neurons in the hidden layer is 7, and the number of neurons in the output layer is 3.

[0023] Preferably, in the back propagation process, the learning rate η is dynamically and adaptively adjusted according to the influence of the learned meteorological characteristic data on the predicted window-to-wall ratio strategy, and η is used tDenotes the learning rate of the tth iteration, η t+1 represents the learning rate of the next iteration, then

[0024]

[0025] In the formula, φ and is a parameter, 0.001<φ<1, ΔE indicates the direction of change of the total error.

[0026] Preferably, the loss function of the multilayer perceptron classifier model adopts a square error loss function, which is expressed as:

[0027]

[0028] In the formula, y j represents the actual window-to-wall ratio strategy of the output layer, Represents the predicted value of the window-to-wall ratio strategy of the output layer.

[0029] Preferably, the weight coefficients w of each layer are updated using the training data t and the offset b t In , when modifying the weight coefficient and the bias, the momentum term β is added to accelerate convergence, then

[0030]

[0031]

[0032] In the formula, w t represents the weight coefficient in the tth iteration, w t+1 represents the weight coefficient of the next iteration, b t represents the bias of the tth iteration, b t+1 Represents the bias for the next iteration, β∈(0,1).

[0033] Preferably, the back propagation algorithm of the output layer and the hidden layer is obtained according to the square error loss function and the chain rule, then:

[0034]

[0035]

[0036] In the formula, w 2 jk represents the weight connecting the kth neuron in the input layer to the jth neuron in the hidden layer; w 3 jk represents the weight of the connection between the kth neuron in the hidden layer and the jth neuron in the output layer, h krepresents the probability distribution of the kth actual window-to-wall ratio strategy category output by the hidden layer; N is the total number of neurons in the output layer, that is, the total number of categories of the window-to-wall ratio strategy, x k It represents the value of the product of the k-th meteorological factor data output by the input layer and the weight plus the bias.

[0037] Preferably, the data preprocessing includes cleaning missing meteorological characteristic data or category labels representing window-to-wall ratio strategies, eliminating outliers in the meteorological characteristic data, and finding the minimum indoor summer cooling load or winter heating load of the building under different window-to-wall ratio strategies for different meteorological characteristic data in each row of data, and obtaining the corresponding optimal window-to-wall ratio strategy as the output target of the multi-layer perceptron classifier.

[0038] A second object of the present invention is to provide a window-to-wall ratio strategy prediction method, comprising:

[0039] Providing a building window-to-wall ratio strategy prediction model, wherein the building window-to-wall ratio strategy prediction model is constructed using the method described above;

[0040] Obtain meteorological characteristic data at the time to be predicted;

[0041] The acquired meteorological characteristic data is input into the building window-to-wall ratio strategy prediction model to obtain the optimal window-to-wall ratio strategy.

[0042] The learning process of the building window-to-wall ratio strategy prediction model provided by the present invention consists of two processes: forward propagation of signals and back propagation of errors. It has a simple structure, is easy to implement, and has good fault tolerance, robustness, and extremely strong nonlinear mapping capabilities. It is suitable for the current nonlinear optimal strategy prediction of the window-to-wall ratio, and the prediction result has a high accuracy.

[0043] The window-to-wall ratio strategy prediction method provided by the present invention predicts the building window-to-wall ratio through a building window-to-wall ratio strategy prediction model constructed based on a multi-layer perceptron. The prediction result has high accuracy and can effectively help the building's external envelope structure to achieve intelligent automatic control of the window-to-wall ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0045] Figure 1 is a flowchart of a window-to-wall ratio strategy prediction method provided by an embodiment of the present invention;

[0046] Figure 2It is a flowchart of a method for constructing a building window-to-wall ratio strategy prediction model provided by an embodiment of the present invention;

[0047] Figure 3 is a flowchart of a training method for a multilayer perceptron classifier provided by an embodiment of the present invention;

[0048] Figure 4 is a structural diagram of a multi-layer perceptron classification provided by an embodiment of the present invention;

[0049] Figure 5 This is an evaluation result of the accuracy of the optimized building window-to-wall ratio strategy prediction model evaluated by using a confusion matrix in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0052] It should also be noted that when an element is referred to as being "fixed on" or "disposed on" another element, it may be directly on the other element or there may be an intermediate element at the same time. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element at the same time.

[0053] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0054] like Figures 1 to 4As shown, it is a window-to-wall ratio strategy prediction method of an embodiment of the present invention, which obtains the optimal window-to-wall ratio strategy through prediction, and provides assistance for realizing intelligent window-to-wall ratio automatic control of the building's exterior envelope structure. A reasonable window-to-wall ratio can help reduce the building's energy consumption.

[0055] See also Figure 1-Figure 4 As shown, the window-to-wall ratio strategy prediction method of the embodiment of the present invention includes:

[0056] Step S100: collecting building window-to-wall ratio strategy data and corresponding historical meteorological characteristic data, and constructing a building window-to-wall ratio strategy prediction model based on a multi-layer perceptron classifier (MLPC).

[0057] Step S200: Meteorological characteristic data at the time to be predicted; the meteorological characteristic data can be obtained from meteorological forecast data, or a meteorological information collection module can be set to obtain it.

[0058] Step S300: inputting the acquired meteorological characteristic data into a building window-to-wall ratio strategy prediction model to obtain an optimal window-to-wall ratio strategy.

[0059] It can be understood that meteorological characteristic data include solar radiation illuminance, dry bulb humidity and relative humidity. Among the numerous meteorological characteristic data, solar radiation illuminance, dry bulb humidity and relative humidity have the greatest impact on the building window-to-wall ratio. Therefore, the window-to-wall ratio strategy prediction method of an embodiment of the present invention constructs a building window-to-wall ratio strategy prediction model by collecting solar radiation illuminance, dry bulb humidity and relative humidity, and predicts the building window-to-wall ratio by collecting solar radiation illuminance, dry bulb humidity and relative humidity at the time to be predicted.

[0060] It can be understood that in order to improve the accuracy of the building window-to-wall ratio strategy prediction model, different building window-to-wall ratio strategy prediction models can be constructed for different buildings in different regions. Therefore, when it is necessary to construct a building window-to-wall ratio strategy prediction model for a certain building in a certain area (such as Jinan), the building window-to-wall ratio strategy data of the building and the meteorological characteristic data of the area can be collected.

[0061] It can be understood that the window-to-wall ratio strategy includes changes in the window-to-wall ratio in the south, north, east and west directions. The strategy is expressed in terms of the window-to-wall ratio status and direction. The window-to-wall ratio status includes the most basic fully closed windows (full walls), half-open windows, and fully-open windows. Of course, the window-to-wall ratio status can also include other states, and can be specific to specific window-to-wall ratio values, which helps to achieve intelligent window-to-wall ratio automation control of the building's exterior envelope structure.

[0062] Without considering the north-facing window status, the window-to-wall ratio changes in the south, east, and west directions are considered respectively, and the window-to-wall ratios in the other two directions are fixed and half-opened, so there are a total of 27 building window-to-wall ratio strategies. The building window-to-wall ratio strategy is represented by the window-to-wall ratio status and direction, where 00 represents the fully closed state, 40 represents the half-open window state, and 80 represents the fully open window state; the east direction is represented by E, the west direction is represented by W, and the south direction is represented by S. The building window-to-wall ratio strategy can be represented by E_40_W_40_S_00, E_40_W_40_S_40, E_40_W_40_S_80, etc.

[0063] The window-to-wall ratio strategy prediction method of the embodiment of the present invention predicts the building window-to-wall ratio through a building window-to-wall ratio strategy prediction model constructed based on a multi-layer perceptron classifier. The prediction result has high accuracy and can effectively help the building's exterior envelope structure achieve intelligent window-to-wall ratio automation control.

[0064] See also Figure 2-Figure 4 As shown in FIG. 1 , the construction method of the building window-to-wall ratio strategy prediction model includes:

[0065] Step S11: preprocessing the collected building window-to-wall ratio strategy data and meteorological characteristic data; the specific implementation method is: after cleaning the missing meteorological characteristic data or the category labels representing the window-to-wall ratio strategy, the outliers in the meteorological characteristic data are eliminated, and the minimum building indoor summer cooling load / winter heating load under different window-to-wall ratio strategies for different meteorological characteristic data in each row of data is found, and the corresponding optimal window-to-wall ratio strategy is obtained as the output target of the multi-layer perceptron classifier.

[0066] Step S12: Divide the processed data into training data and data to be predicted.

[0067] Step S13: Use the training data to train a multilayer perceptron classifier, and construct a multilayer perceptron classifier model according to the trained weight coefficients and bias as a building window-to-wall ratio strategy prediction model.

[0068] That is, the forward propagation algorithm is executed using the weight coefficient and bias finally generated in the training, and the calculated value combining the meteorological factor with the weight coefficient and the bias in the input layer neurons is propagated to the hidden layer. Then, the probability distribution of the window-to-wall ratio strategy category is predicted in the activated neurons of the hidden layer. According to the category output by the output layer neurons, the corresponding working condition is determined to be the optimal window-to-wall ratio strategy, that is, the optimal building variable window-to-wall enclosure structure strategy.

[0069] See also Figure 3 and Figure 4 As shown, the training method of the multilayer perceptron classifier includes:

[0070] Step S131: construct a multilayer perceptron classifier. The input of the input layer of the multilayer perceptron classifier is the meteorological feature data. The nodes of the input layer are determined by the number of meteorological factors of the meteorological feature data. The output of the output layer is the window-to-wall ratio strategy. The output nodes of the output layer are determined by the number of categories of the window-to-wall ratio strategy. According to the Kolmogorov theorem, the number of neurons in the hidden layer is determined according to formula (1).

[0071] J=2a+1 (1)

[0072] In the formula, J represents the number of neurons in the hidden layer, and a represents the number of neurons in the input layer;

[0073] Step S132: Initialize the learning rate η0, so that η0∈(0,1), and randomly generate the initial value w0 of the weight coefficient and the initial value b0 of the bias of each layer of neurons;

[0074] Step S133: Starting from the input layer, forward propagation is performed through the network, the neurons of the hidden layer are activated using the Sigmoid activation function, and the neurons of the output layer are activated using the identity function;

[0075] Preferably, the Sigmoid activation function is expressed as:

[0076]

[0077] In the formula, f(a j ) represents the probability distribution of the j-th predicted window-to-wall ratio strategy category output by the hidden layer, a j represents the jth input of the hidden layer, i.e., the product of the meteorological factor data and the weight plus the bias value, A and B are parameters, A∈(0.01, 100), B∈(1, 100);

[0078] The traditional Sigmoid activation function has the problem of insensitive changes in some areas, which slows down the convergence speed. Therefore, the Sigmoid activation function of this embodiment adds two parameters, A and B, to make all input values ​​within the sensitive change area as much as possible to ensure the rapidity of the training network. According to formula (2), when A and B are less than 0 or greater than 100, it will affect its convergence speed, thereby limiting the parameter range A∈(0.01, 100), which is used to speed up the change rate of the original flat area and change the change range of the training data; B∈(1, 100), which is used to expand the change area and eliminate the insensitive area.

[0079] Step S134: Calculate the error and use the training data to update the weight coefficients w of each layer through the back propagation algorithm t and the offset b t .

[0080] Preferably, in the back propagation process, the learning rate η is dynamically and adaptively adjusted according to the influence of solar radiation, dry bulb temperature and relative humidity on the prediction strategy of window-to-wall ratio. t , so as to ensure the speed and accuracy of learning, using η t Denotes the learning rate of the tth iteration, η t+1 Represents the learning rate of the next iteration, then:

[0081]

[0082] In the formula, φ and is a parameter, 0.001<φ<1, ΔE indicates the direction of change of the total error.

[0083] Furthermore, the loss function of the multilayer perceptron classifier model adopts the square error loss function, which is expressed as:

[0084]

[0085] In the formula, y j represents the actual window-to-wall ratio strategy of the output layer, Represents the predicted value of the window-to-wall ratio strategy at the output layer;

[0086] Furthermore, according to the square error loss function and the chain rule, the back propagation algorithm of the output layer and the hidden layer is obtained:

[0087]

[0088]

[0089] Formula (5) is the back propagation formula of the output layer, and formula (6) is the back propagation formula of the hidden layer. In the formula, w 2 jk represents the weight connecting the kth neuron in the input layer to the jth neuron in the hidden layer; w 3 jk represents the weight of the connection between the kth neuron in the hidden layer and the jth neuron in the output layer, h k represents the probability distribution of the kth actual window-to-wall ratio strategy category output by the hidden layer; N is the total number of neurons in the output layer, that is, the total number of categories of the window-to-wall ratio strategy, x k It represents the value of the product of the k-th meteorological factor data output by the input layer and the weight plus the bias.

[0090] Furthermore, the weight coefficients w of each layer are updated using the training data t and the offset b t In the process of correcting the weight coefficient and the bias, the momentum term β is added to accelerate the convergence. The formula is as follows:

[0091]

[0092]

[0093] In the formula, w t is the weight coefficient in the tth iteration, w t+1 is the weight coefficient for the next iteration, b t is the bias of the tth iteration, b t+1 is the bias for the next iteration; η t+1 Represents the learning rate for the next iteration, β∈(0,1).

[0094] Step S14: Input the data to be predicted into the building window-to-wall ratio strategy prediction model, output the prediction target strategy, measure the building window-to-wall ratio strategy prediction model using the accuracy index and the confusion matrix, and evaluate the prediction accuracy of the building window-to-wall ratio strategy prediction model. The larger the Accuracy and the larger the confusion matrix, the more accurate the prediction of the model, and vice versa.

[0095] The embodiment of the present invention constructs a building window-to-wall ratio strategy prediction model based on a multi-layer perceptron, and the prediction result has a high accuracy, which can effectively help the building exterior envelope structure to realize intelligent automatic control of the window-to-wall ratio, which is specifically embodied in:

[0096] First, the learning process of the building window-to-wall ratio strategy prediction model constructed based on multi-layer perceptron consists of two processes: forward propagation of signals and back propagation of errors. It has a simple structure, is easy to implement, and has good fault tolerance, robustness, and extremely strong nonlinear mapping capabilities. It is suitable for the current nonlinear optimal strategy prediction of window-to-wall ratio.

[0097] Second, the building window-to-wall ratio strategy prediction model built based on a multi-layer perceptron adopts a dynamic and adaptive change of the learning rate, so that it can automatically adjust according to the influence of the learned meteorological factors on the predicted window-to-wall ratio strategy, thereby ensuring the learning speed and accuracy.

[0098] Third, the building window-to-wall ratio strategy prediction model built based on multi-layer perceptron selects the Sigmoid function as the activation function of the neurons in the hidden layer, and introduces the two parameters A and B to ensure the rapidity of training the network.

[0099] Fourth, the building window-to-wall ratio strategy prediction model constructed based on multi-layer perceptron introduces the momentum term β, and uses the last adjustment of the weight coefficient and bias to modify the current weight coefficient and bias, smoothing the learning path, thereby accelerating the convergence speed and preventing oscillation. When the amount of data is large or solving the multi-classification problem of the window-to-wall ratio strategy, the performance of the building window-to-wall ratio strategy prediction model can be improved.

[0100] The embodiment of the present invention takes a building in Jinan as an example, collects the window-to-wall ratio strategy of the building and the historical meteorological characteristic data of the Jinan area, and constructs a building window-to-wall ratio strategy prediction model based on a multi-layer perceptron classifier, and finally evaluates the optimized building window-to-wall ratio strategy prediction model.

[0101] In this embodiment, the meteorological characteristic data include three meteorological factor data: solar radiation illumination, dry bulb humidity and relative humidity. The embodiment of the present invention only considers the change of the window-to-wall ratio in the south direction, fixes the window-to-wall ratio in the east and west directions and shows a half-open window state. Therefore, the target strategy is the following three strategy categories, with labels 1, 2, and 3 as output targets respectively. Label 1 is represented by E_40_W_40_S_00, label 2 is represented by E_40_W_40_S_40, and label 3 is represented by E_40_W_40_S_80.

[0102] Therefore, if Figure 4 As shown, the number of neurons in the input layer x of the constructed multilayer perceptron classifier model is 3, the number of neurons in the hidden layer h is 7, and the number of neurons in the output layer o is 3. Of course, if the number of window-to-wall ratio strategies is 27, the number of neurons in the output layer is 27.

[0103] When evaluating the prediction model of building window-to-wall ratio strategy, the predicted data is input into the prediction model, and the predicted optimal building window-to-wall ratio strategy is output. Finally, the actual historical optimal window-to-wall ratio strategy and the predicted optimal window-to-wall ratio strategy are compared, and the accuracy of the optimized prediction model of building window-to-wall ratio strategy is evaluated through accuracy and confusion matrix. The evaluation results are shown in Table 1. The accuracy is 85.65%, that is, the probability that the optimized model predicts the correct optimal window-to-wall ratio strategy is greater than 80%; at the same time, the area in the confusion matrix located in strategy category 3 is the actual strategy E_40_W_40_S_80 and the predicted output strategy E_40_W_40_S_80, which accounts for 73%, indicating that the actual strategy category of the test data is mostly 3, and the probability of accurately predicting strategy category 3 is 73%, indicating that the prediction model of building window-to-wall ratio strategy has high accuracy and can accurately predict the optimal window-to-wall ratio strategy.

[0104] Table 1 Evaluation results of the building window-to-wall ratio strategy prediction model

[0105] Evaluation Criteria Accuracy Accuracy 85.65％ Confusion Matrix 73％

[0106] It can be understood that the building window-to-wall ratio strategy prediction model of the embodiment of the present invention can also be applied to other window-to-wall ratio strategy prediction methods.

[0107] The embodiments of the present invention may be implemented by hardware, software, or a combination of software and hardware. The hardware portion may be implemented using dedicated logic; the software portion may be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. A person of ordinary skill in the art will appreciate that the above-mentioned apparatus and method may be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier.

[0108] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for constructing a building window-to-wall ratio strategy prediction model, characterized in that: include: Preprocess the collected building window-to-wall ratio strategy data and the corresponding historical meteorological characteristic data; Divide the processed data into training data and data to be predicted; A multilayer perceptron classifier is trained using the training data, and a multilayer perceptron classifier model is constructed according to the trained weight coefficients and biases as a prediction model for the building window-to-wall ratio strategy; The training methods for the Multilayer Perceptron classifier include: Construct a multilayer perceptron classifier. The input layer of the multilayer perceptron classifier inputs the meteorological feature data, the output layer outputs the window-to-wall ratio strategy, and the number of neurons in the hidden layer is determined according to the following relationship: J=2a+1 In the formula, J represents the number of neurons in the hidden layer, and a represents the number of neurons in the input layer; Initialize the learning rate η0, so that η0∈(0,1), and randomly generate the initial value w0 of the weight coefficient and the initial value b0 of the bias of each layer of neurons; Starting from the input layer, forward propagation is performed through the network, using the Sigmoid activation function to activate neurons in the hidden layer and the identity function to activate neurons in the output layer; Calculate the error and use the training data to update the weight coefficient w of each layer through the back propagation algorithm t and the offset b t .

2. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 1, characterized in that: It also includes inputting the data to be predicted into a building window-to-wall ratio strategy prediction model, and measuring the building window-to-wall ratio strategy prediction model using an accuracy index and a confusion matrix. If the accuracy is greater than 80% and the confusion matrix is ​​greater than 70%, it indicates that the building window-to-wall ratio strategy prediction model has high accuracy.

3. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 1, characterized in that: The Sigmoid activation function is expressed as: In the formula, f(a j ) represents the probability distribution of the j-th predicted window-to-wall ratio strategy category output by the hidden layer, a j It represents the jth input of the hidden layer, that is, the product of the meteorological factor data and the weight plus the bias value. A and B are parameters, A∈(0.01, 100), B∈(1, 100).

4. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 1, characterized in that: The meteorological characteristic data include solar irradiance, dry-bulb temperature and relative humidity. The window-to-wall ratio strategy includes three strategy categories. The number of neurons in the input layer of the multilayer perceptron classifier model is 3, the number of neurons in the hidden layer is 7, and the number of neurons in the output layer is 3.

5. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 1, characterized in that: In the back propagation process, the learning rate η is dynamically and adaptively adjusted according to the influence of the learned meteorological characteristic data on the prediction window-to-wall ratio strategy. t Denotes the learning rate of the tth iteration, η t+1 represents the learning rate of the next iteration, then In the formula, φ and is a parameter, 0.001<φ<1, ΔE indicates the direction of change of the total error.

6. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 4, characterized in that: The loss function of the multilayer perceptron classifier model adopts the square error loss function, which is expressed as: In the formula, y j represents the actual window-to-wall ratio strategy of the output layer, Represents the predicted value of the window-to-wall ratio strategy of the output layer.

7. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 5, characterized in that: Use training data to update the weight coefficients w of each layer t and the offset b t In , when modifying the weight coefficient and the bias, the momentum term β is added to accelerate convergence, then In the formula, w t represents the weight coefficient in the tth iteration, w t+1 represents the weight coefficient of the next iteration, b t represents the bias of the tth iteration, b t+1 Represents the bias for the next iteration, β∈(0,1).

8. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 5, characterized in that: According to the square error loss function and the chain rule, the back propagation algorithm of the output layer and the hidden layer is obtained: In the formula, w 2 jk represents the weight connecting the kth neuron in the input layer to the jth neuron in the hidden layer; w 3 jk represents the weight of the connection between the kth neuron in the hidden layer and the jth neuron in the output layer, h k represents the probability distribution of the kth actual window-to-wall ratio strategy category output by the hidden layer; N is the total number of neurons in the output layer, that is, the total number of categories of the window-to-wall ratio strategy, x k It represents the value of the product of the k-th meteorological factor data output by the input layer and the weight plus the bias.

9. The method for constructing a building window-to-wall ratio strategy prediction model according to claim 1, characterized in that: The data preprocessing includes cleaning missing meteorological characteristic data or category labels representing window-to-wall ratio strategies, eliminating outliers in the meteorological characteristic data, and finding the minimum indoor summer cooling load or winter heating load of the building under different window-to-wall ratio strategies for different meteorological characteristic data in each row of data, and obtaining the corresponding optimal window-to-wall ratio strategy as the output target of the multi-layer perceptron classifier.

10. A window-to-wall ratio strategy prediction method, characterized in that: include: Providing a building window-to-wall ratio strategy prediction model, wherein the building window-to-wall ratio strategy prediction model is constructed using the method described in any one of claims 1 to 9; Obtain meteorological characteristic data at the time to be predicted; The acquired meteorological characteristic data is input into the building window-to-wall ratio strategy prediction model to obtain the optimal window-to-wall ratio strategy.

Images