Energy consumption prediction method for broken line type plane office building

Abstract

The invention discloses an energy consumption prediction method for a broken line type plane office building, and the method specifically comprises the following steps: 1, collecting plane office buildings and corresponding plane office building parameter information, and classifying the collected plane office buildings; step 2, carrying out graphical analysis to obtain plane data, and establishing a sample library; 3, drawing a three-dimensional building model, and carrying out annual dynamic energy consumption simulation to obtain energy consumption data; 4, selecting variables of each type of plane office buildings, and then performing variable screening; 5, constructing and training a random forest prediction model and a neural network prediction model; 6, optimizing a training result in the step 5 by adopting a grid search method to obtain an optimal prediction model; and 7, calling the optimal prediction model of each type of plane office building to predict the energy consumption of the broken line type plane office building. According to the invention, scientific prediction of the office building energy consumption can be realized, and the efficiency of predicting the office building energy consumption is improved.

Description

Technical Field

[0001] The field of the present invention is the technical field of building energy consumption prediction, and specifically relates to an energy consumption prediction method for a folded-line planar office building. Background Art

[0002] In recent years, with the growth of the scale of public buildings and the growth of the average energy consumption intensity, the energy consumption of such buildings has exceeded one-third of the building energy consumption in China. In 2021, the public building area in the whole country was about 14.7 billion square meters, and the total energy consumption of public buildings was 386 million tons of standard coal equivalent, accounting for 35% of the total building energy consumption. As a part of public buildings, office buildings play an important role in energy use. It needs to operate continuously throughout the year, whether during normal working hours or non-working hours. Compared with other types of public buildings, the electricity demand of office buildings is continuous, resulting in its relatively high annual average energy consumption.

[0003] The energy consumption difference of office buildings not only depends on whether advanced energy-saving equipment is adopted, but more importantly on the building plane, which determines the general shape, proportion, size, etc. of the building, and has important impacts on construction cost, operation energy consumption and aesthetic effect. Although the research on building plane design has been involved in the field of energy-saving prediction, it mainly focuses on the energy consumption comparison of different planes, and fails to provide specific parameter qualitative and quantitative directions for the low-energy optimization of the building plane according to the prediction results, such as the shape, proportion, size, angle, etc. of the plane. At present, building energy consumption prediction mainly uses software simulation to screen better solutions. This method is limited by the subjective experience of architects and cannot provide scientific optimization suggestions for building space; moreover, performance simulation requires more in-depth building technical knowledge, and many architects need to spend a lot of time understanding technical formulas and models, making it difficult to promote energy conservation in building design quickly and effectively, and reducing the efficiency of office building energy consumption prediction. Summary of the Invention

[0004] The purpose of the present invention is to provide an energy consumption prediction method for a folded-line planar office building. The present invention can realize the scientific prediction of the energy consumption of office buildings and improve the efficiency of office building energy consumption prediction.

[0005] The technical solution of the present invention: An energy consumption prediction method for a folded-line planar office building specifically includes the following steps:

[0006] Step 1: Collect the planar office building and the corresponding planar office building parameter information, and classify the collected planar office buildings;

[0007] Step 2: Based on the planar office building parameter information collected in Step 1, diagram the classified planar office buildings to obtain planar data, and establish a sample library based on the obtained planar data;

[0008] Step 3: Draw a 3D building model based on the plane data, simulate the dynamic energy consumption of the 3D building model throughout the year, obtain the energy consumption data per unit building area of ​​the office building corresponding to the plane data, and then record the energy consumption data in the sample library;

[0009] Step 4: Based on the plan office building parameter information collected in step 1, select variables for each type of plan office building, and then perform variable screening;

[0010] Step 5: Take each type of plan office building and the corresponding variables as input parameters, and take the energy consumption data obtained in step 3 as output parameters to construct a random forest prediction model and a neural network prediction model respectively; train the constructed random forest prediction model and neural network prediction model according to the plan data and energy consumption data in the sample library;

[0011] Step 6: Use the grid search method to optimize the training results of the random forest prediction model and the neural network prediction model in step 5 to obtain the optimal prediction model;

[0012] Step 7: Call the optimal prediction model for each type of flat office building to predict the energy consumption of the broken line flat office building.

[0013] In the aforementioned method for predicting energy consumption of a zigzag plane office building, in step one, the plane office building parameter information includes the width, depth and angle of the plane office building; the classification process is divided into non-zigzag plane, one-fold plane and two-fold plane according to the shapes of different plane office buildings.

[0014] In the aforementioned method for predicting energy consumption of a zigzag-shaped plane office building, in step 2, the process of establishing the sample library illustrates the classified plane office buildings to determine the plane degrees of freedom; defines a numerical boundary range based on the plane office building parameter information; and then uses the Python platform to traverse the shape samples existing in each type of plane office building according to the defined numerical boundary range, as a sample library for each type of plane office building.

[0015] In the aforementioned method for predicting energy consumption of a zigzag plan office building, in step 3, the annual dynamic energy consumption simulation uses the EnergyPlus engine.

[0016] In the aforementioned method for predicting energy consumption of a zigzag plane office building, in step 4, the process of selecting variables is to introduce trigonometric functions into the collected plane office building parameter information through four arithmetic operations to make parameter combinations, traverse the parameter combinations of four arithmetic operations and trigonometric functions, and calculate the correlation coefficient between the result of each parameter combination and the energy consumption data through the Spearman formula, and use the parameter combination with the highest correlation as the variable for each type of plane office building;

[0017] The specific calculation formula of the Spearman formula is as follows:

[0018]

[0019] In the formula, ρ is the Spearman correlation coefficient, and its value range is [-1, 1]. d i is the rank difference between two variables, and n is the sample size;

[0020] The principles of variable selection include: the number of selected variables does not exceed two, the selected variables include the parameters for controlling the corresponding planar office building, and the selected variables have the highest correlation with energy consumption.

[0021] In the aforementioned energy consumption prediction method for the zigzag planar office building, in step five, the functional relationship of the random forest prediction model is:

[0022]

[0023] In the formula, is the output parameter predicted by the random forest prediction model, RF(x) is the ensemble learning function of decision trees, x is the input parameter of the random forest prediction model, N is the number of decision trees in the random forest model, i represents the number of the decision tree, is the output parameter predicted by the i-th decision tree;

[0024] The prediction function expression of a single decision tree in the random forest prediction model is:

[0025]

[0026] In the formula, Tree i (x) is the prediction function of a single decision tree;

[0027] The training process of the random forest prediction model is based on the variables of each type of planar office building obtained through variable selection, constructing a feature data set and target variable values, dividing the feature data set into a test set and a training set; based on the target variable values, the decision tree conducts independent training by randomly extracting variables from the feature data set.

[0028] In the aforementioned energy consumption prediction method for the zigzag planar office building, the neural network prediction model includes an input layer, an output layer, and a hidden layer. The functional relationship of the neural network prediction model is:

[0029]

[0030] In the formula, is the prediction output of the neural network prediction model, NN(X) is the prediction output function of the neural network prediction model; n last_hiddenis the number of neurons in the last hidden layer of the neural network prediction model, is the weight from the j-th neuron in the last hidden layer of the neural network prediction model to the output layer neuron, is the output of the j-th neuron in the last hidden layer of the neural network prediction model, b out is the bias term of the output layer of the neural network prediction model, and X is the input parameter of the neural network prediction model;

[0031] The output function of a single hidden neuron in the hidden layer is:

[0032] a 1m =ReLU(∑ n w 1mn P n +b 1m );

[0033] In the formula, a 1m represents the output parameter of the m-th neuron in the hidden layer of the neural network prediction model, w 1mn represents the weight matrix between the input layer and the hidden layer of the neural network prediction model, P n represents the input parameter vector, b 1m is the bias vector between the input layer and the hidden layer, n represents the input layer neuron number, and m represents the hidden layer neuron number,

[0034] The output function of the output layer is expressed as:

[0035]

[0036] In the formula, represents the output parameter of the k-th neuron in the output layer, w 2km represents the weight matrix between the hidden layer and the output layer, b 2k represents the bias vector between the hidden layer and the output layer, and k represents the output layer neuron number;

[0037] The specific training process of the neural network prediction model is to batch the plane data and energy consumption data in the sample library into the model through multiple iterations, calculate the loss and gradient, and then update the weights of the neural network prediction model to complete the training; the mean square error MSE is used as the loss function for the loss, and the calculation formula is as follows:

[0038]

[0039] In the formula, is the true value of the i-th sample, is the predicted value of the i-th sample, that is, the output of the model, and N is the number of samples;

[0040] The gradient is the partial derivative of the loss function with respect to the parameters of the neural network prediction model, and the calculation formula is as follows:

[0041]

[0042] In the formula, θ represents all trainable parameters of the neural network prediction model, that is, the weight w and the bias b, is the partial derivative of the loss function with respect to the i-th parameter, and L is the loss function,

[0043] The specific calculation process of the gradient is as follows:

[0044] First, through forward propagation, the input data is calculated by the neural network to obtain the predicted value y pred , then calculate the loss function value L; then through backpropagation, starting from the output layer, calculate the partial derivative of the loss function with respect to each parameter layer by layer; finally, use the Adam optimizer to update the model parameters according to the gradient:

[0045]

[0046] In the formula, θ new is the new parameter of the neural network prediction model, θ old is the original parameter of the neural network prediction model, and η is the learning rate.

[0047] In the aforementioned energy consumption prediction method for the zigzag planar office building, in step six, the specific process of the grid search method is to generate all possible parameter combinations, then create, compile, and train a neural network prediction model for each group of parameters, and then make predictions on the training set and the test set;

[0048] The optimization process includes adjusting the number of hidden layers, the number of neurons, and the number of iterations of the neural network prediction model according to the results obtained by prediction on the training set and the test set, and determining the depth of the subtrees and the sample weights of the leaf nodes of the random forest prediction model.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention classifies planar office buildings by collecting various parameter information of planar office buildings, taking into account the characteristics of different types of buildings, laying a solid foundation for subsequent accurate energy consumption prediction, and being able to more accurately reflect the energy consumption characteristics of different types of folded-line planar office buildings. The present invention can predict energy consumption from different perspectives by combining a random forest prediction model and a neural network prediction model, and uses the grid search method to optimize the training results of the model, further optimizing the model, thereby improving the accuracy of prediction, so that the finally obtained optimal prediction model can more accurately predict the energy consumption of folded-line planar office buildings. In addition, the present invention classifies planar office buildings into non-folded planes, single-folded planes and double-folded planes according to their shapes, which can cover various forms of folded-line planar office buildings. Whether it is a folded-line planar office building with a simple or complex shape, its energy consumption can be predicted by this method, which has wide applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flowchart of the method application of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] The present invention will be further described below in conjunction with the drawings and embodiments, but it shall not be used as a basis for limiting the present invention.

[0053] Embodiment: An energy consumption prediction method for a folded-line planar office building, as Figure 1 shown, specifically includes the following steps:

[0054] Step 1: Collect planar office buildings and corresponding parameter information of planar office buildings, and classify the collected planar office buildings;

[0055] In this embodiment, due to the significant demand for summer heat prevention and winter heating in the Yangtze River Delta region, and due to the continuous increase in population and the continuous expansion of urban land area in this region, its ecology has changed significantly, and the number of extreme high and low temperature climates has increased significantly, bringing huge pressure to the energy in this region. The buildings in this region have certain typicality in terms of energy consumption. Therefore, collect the planar office buildings and parameter information of planar office buildings in this region as the actual case data for research; through analyzing the different planar shapes of planar office buildings, conduct a macro classification into three types of planes: non-folded planes, single-folded planes and double-folded planes. A non-folded plane usually refers to a building design without obvious turning or concave-convex changes, presenting a relatively simple and smooth form as a whole; a single-folded plane usually refers to a design in which the walls or construction lines of some parts of the building appear in a folded-line shape; a double-folded plane usually refers to a structural form of the staircase part in building design. A double-folded plane means that there is a turning point during the ascent of the staircase, usually a 180-degree turn.

[0056] Specifically, the parameter information of the planar office building includes the face width, depth, and angle of the planar office building.

[0057] Step 2: Based on the parameter information of the planar office building collected in Step 1, conduct graphical illustration on the classified planar office buildings to obtain planar data, and establish a sample library based on the obtained planar data.

[0058] In this embodiment, according to the parameter information of the zigzag planar office buildings in the Yangtze River Delta region collected in Step 1, conduct graphical illustration on three types of planes respectively to determine the planar degrees of freedom, that is, the parameter combinations and quantities controlling the plane, obtain planar data, and then, according to the numerical boundary range of the case parameters, traverse the possible shape samples of various planes using the Python platform as the sample library of each type of planar office building.

[0059] Step 3: Draw a 3D building model based on the planar data, conduct annual dynamic energy consumption simulation on the 3D building model to obtain the energy consumption data per unit building area of the office building corresponding to the planar data, and then record the energy consumption data into the sample library.

[0060] In this embodiment, draw the 3D building models of each plane in each type of planar office building, call the EnergyPlus engine to conduct annual dynamic energy consumption simulation on them, obtain the energy consumption data per unit building area of the office building corresponding to the planar data, and record them one by one into the sample library.

[0061] The EnergyPlus is a building energy consumption simulation engine, a software for building energy consumption simulation. It can conduct comprehensive energy consumption simulation analysis and economic analysis on the heating, cooling, lighting, ventilation, and other energy consumptions of a building.

[0062] Step 4: Select variables for each type of planar office building based on the parameter information of the planar office building collected in Step 1, and then conduct variable screening.

[0063] In this embodiment, since the parameter information of the planar office building may be somewhat one-sided, trigonometric function operations are introduced into the parameter control of each type of planar office building through four arithmetic operations to make parameter combinations, traverse the parameter combinations of the four arithmetic operations and trigonometric function operations, and calculate the correlation coefficient between the result of each parameter combination and the energy consumption data through the Spearman formula. The parameter combination with the highest correlation degree is used as the variable for each type of planar office building. In view of the non-normal characteristics of the data distribution, the Spearman formula is used here to calculate the correlation coefficient as the core tool for variable screening, aiming to identify the parameter combination that shows the highest correlation degree with the energy consumption data (specific target variable), and use it as the basis for subsequent analysis and model construction.

[0064] The specific calculation formula of the Spearman formula is:

[0065]

[0066] Where ρ is the Spearman correlation coefficient, and its value range is [-1, 1], d i is the rank difference between two variables, and n is the sample size.

[0067] The principles for variable screening include: the number of selected variables does not exceed two, the selected variables include the parameters for controlling the corresponding planar office buildings, and the selected variables have the highest correlation with energy consumption.

[0068] Step 5: Take each type of planar office building and the corresponding variables as input parameters, and take the energy consumption data obtained in Step 3 as output parameters to respectively construct a random forest prediction model and a neural network prediction model; train the constructed random forest prediction model and neural network prediction model according to the planar data and energy consumption data in the sample library;

[0069] In this embodiment, the random forest prediction model and the neural network prediction model are constructed using the TensorFlow platform; TensorFlow is an open-source deep learning framework developed by Google for constructing and training neural networks.

[0070] Random forest is an ensemble learning algorithm that combines multiple decision trees and generates new subtrees by randomly extracting a part of the feature and sample subsets. The random forest model can be applied to various scenarios, such as classification and regression, and has high accuracy in various datasets.

[0071] Specifically, the functional relationship of the random forest prediction model is:

[0072]

[0073] Where is the output parameter predicted by the random forest prediction model, RF(x) is the ensemble learning function of the decision tree, x is the input parameter of the random forest prediction model, N is the number of decision trees in the random forest model, i represents the number of the decision tree, is the output parameter predicted by the i-th decision tree.

[0074] The prediction function expression of a single decision tree in the random forest prediction model is:

[0075]

[0076] Where Tree i (x) is the prediction function of a single decision tree;

[0077] The training process of the random forest prediction model is based on the variables of each type of planar office building obtained through variable screening. A feature dataset (the previously obtained variables are input in dictionary form) and target variable values are constructed, and the feature dataset is divided into a test set and a training set. Based on the target variable values, a random forest regressor is then instantiated. It contains multiple decision trees inside, and each tree is independently trained by randomly sampling samples (Bootstrap sampling) and features from the original training data. Considering the small sample size and features, there is no limit on the depth of the subtrees of each decision tree when constructing the optimal model here, and the sample weights are not considered for the leaf nodes. Subsequently, the random forest prediction model can obtain the final prediction value by calculating the average of the prediction results of each decision tree, and use metrics such as the coefficient of determination, mean absolute percentage error, and mean absolute error to comprehensively evaluate the prediction performance of the random forest prediction model.

[0078] The neural network prediction model is a prediction method based on artificial neural networks (ANN). It mimics the behavioral characteristics of biological neural networks and builds and trains the model through distributed parallel information processing. The basic principle of the neural network prediction model is to learn a large amount of historical data, discover the patterns and regularities in it, and use these patterns and regularities to predict the future. This model usually includes an input layer, hidden layers, and an output layer, where there can be multiple hidden layers. Each layer consists of multiple neurons, and these neurons are connected by weights. During the training process, the model adjusts these weights to minimize the difference between the prediction result and the actual result.

[0079] Specifically, the neural network prediction model includes an input layer, an output layer, and hidden layers. The functional relationship of the neural network prediction model is:

[0080]

[0081] In the formula, is the prediction output of the neural network prediction model, and NN(X) is the prediction output function of the neural network prediction model; n last_hidden is the number of neurons in the last hidden layer of the neural network prediction model, is the weight from the j-th neuron in the last hidden layer of the neural network prediction model to the output layer neuron, is the output of the j-th neuron in the last hidden layer of the neural network prediction model, and b out is the bias term of the output layer of the neural network prediction model, and X is the input parameter of the neural network prediction model;

[0082] Furthermore, the output function of a single hidden neuron in the hidden layer is:

[0083] a 1m = ReLU(∑n w 1mn P n +b 1m );

[0084] In the formula, a 1m represents the output parameter of the m-th neuron in the hidden layer of the neural network prediction model, w 1mn represents the weight matrix between the input layer and the hidden layer of the neural network prediction model, P n represents the input parameter vector, b 1m is the bias vector between the input layer and the hidden layer, n represents the input layer neuron number, m represents the hidden layer neuron number,

[0085] Furthermore, the output function of the output layer is expressed as:

[0086]

[0087] In the formula, represents the output parameter of the k-th neuron in the output layer, w 2km represents the weight matrix between the hidden layer and the output layer, b 2k represents the bias vector between the hidden layer and the output layer, k represents the output layer neuron number.

[0088] In this embodiment, the specific training process of the neural network prediction model is to send the plane data and energy consumption data in the sample library into the model batch by batch through multiple iterations, calculate the loss and gradient, and then update the weights of the neural network prediction model to complete the training.

[0089] The mean squared error MSE is used as the loss function for the loss, and the calculation formula is as follows:

[0090]

[0091] In the formula, is the true value of the i-th sample, is the predicted value of the i-th sample, that is, the output of the model, N is the number of samples;

[0092] The gradient is the partial derivative of the loss function with respect to the parameters of the neural network prediction model, and the calculation formula is as follows:

[0093]

[0094] In the formula, θ is all the trainable parameters of the neural network prediction model, that is, the weight w and the bias vector b, is the partial derivative of the loss function with respect to the i-th parameter, L is the loss function,

[0095] The specific calculation process of the gradient is:

[0096] First, through forward propagation, the input data is calculated by the neural network to obtain the predicted value y pred , and then the loss function value L is calculated; then through backpropagation, starting from the output layer, the partial derivative of the loss function with respect to each parameter is calculated layer by layer; finally, the Adam optimizer is used to update the model parameters according to the gradient:

[0097]

[0098] where θ new is the new parameter of the neural network prediction model, θ old is the original parameter of the neural network prediction model, and η is the learning rate.

[0099] Step Six: Use the grid search method to optimize the training results of the random forest prediction model and the neural network prediction model in Step Five to obtain the optimal prediction model;

[0100] In this example, according to the model training results in Step Five, the grid search method is used to generate all possible parameter combinations, and then for each set of parameters, a neural network prediction model is created, compiled, and trained, and then predictions are made on the training set and the test set to explore the parameter space of the algorithms of the random forest prediction model and the neural network prediction model. By adjusting the number of hidden layers, the number of neurons, the number of iterations of the neural network prediction model, and determining the depth of the subtrees and the sample weights of the leaf nodes of the random forest prediction model, the optimal neural network prediction model and random forest prediction model are found.

[0101] Furthermore, the neural network prediction model and the random forest prediction model are compared and selected through evaluation indicators such as the coefficient of determination, mean square error, and mean absolute error, so that the final neural network prediction model and random forest prediction model not only perform well on the training set but also have good generalization effects in the test set.

[0102] Step Seven: Call the optimal prediction model of each type of planar office building to predict the energy consumption of the linear planar office building.

[0103] In this embodiment, the optimal prediction model of each type of planar office building is called using the PyQt framework and the QtDesigner tool. PyQt is a cross-platform graphical user interface (GUI) application development framework. It is the Python binding of the Qt framework and is written in C++ at the bottom layer, so it has high operating efficiency. Qt Designer is a graphical user interface (GUI) design tool based on the Qt framework. Qt Designer allows users to create GUIs using drag-and-drop operations and can automatically generate corresponding Qt code. Qt Designer provides an intuitive interface that enables users to easily create, edit, and layout GUI elements such as buttons, labels, text boxes, list boxes, and so on.

[0104] In addition, the present invention also develops an interactive visualization interface for predicting the energy consumption of folded-line planar office buildings. Through drag-and-drop operations and signal-slot connections, an intuitive and easy-to-use visualization platform is built. Users can input relevant parameters of the building and obtain corresponding energy consumption prediction results and planar office building parameter information such as the corresponding frontage, depth, and angle of the planar office building.

[0105] The results in this example show that the data of the non-folded plane and the one-fold plane are more excellent when trained by the random forest algorithm compared to the neural network, and the data of the two-fold plane are more excellent when trained by the neural network algorithm compared to the random forest. The data is divided into a training set and a test set in a ratio of 8:2. Whether data augmentation is performed or not, the final effect on the test set is relatively excellent.

[0106] By combining the random forest prediction model and the neural network prediction model, the present invention can predict energy consumption from different perspectives. The grid search method is used to optimize the training results of the model, further optimizing the model, thereby improving the accuracy of prediction and enabling the finally obtained optimal prediction model to more accurately predict the energy consumption of folded-line planar office buildings. This method classifies planar office buildings by collecting various parameter information of planar office buildings, considering the characteristics of different types of buildings, laying a solid foundation for accurately predicting energy consumption later and being able to more precisely reflect the energy consumption characteristics of different types of folded-line planar office buildings. In addition, the present invention divides planar office buildings into non-folded planes, one-fold planes, and two-fold planes according to their shapes, which can cover various forms of folded-line planar office buildings. Whether it is a simple or complex-shaped folded-line planar office building, the energy consumption can be predicted by this method, showing wide applicability.

[0107] In summary, the present invention can achieve scientific prediction of the energy consumption of office buildings and improve the efficiency of energy consumption prediction for office buildings.

Claims

1. A method for predicting energy consumption of a zigzag plan office building, characterized in that: The specific steps include: Step 1: Collect plane office buildings and corresponding plane office building parameter information, and classify the collected plane office buildings; Step 2: Based on the plane office building parameter information collected in step 1, the classified plane office buildings are illustrated to obtain plane data, and a sample library is established based on the obtained plane data; Step 3: Draw a 3D building model based on the plane data, simulate the dynamic energy consumption of the 3D building model throughout the year, obtain the energy consumption data per unit building area of ​​the office building corresponding to the plane data, and then record the energy consumption data in the sample library; Step 4: Based on the plan office building parameter information collected in step 1, select variables for each type of plan office building, and then perform variable screening; Step 5: Take each type of plan office building and the corresponding variables as input parameters, and take the energy consumption data obtained in step 3 as output parameters to construct a random forest prediction model and a neural network prediction model respectively; train the constructed random forest prediction model and neural network prediction model according to the plan data and energy consumption data in the sample library; Step 6: Use the grid search method to optimize the training results of the random forest prediction model and the neural network prediction model in step 5 to obtain the optimal prediction model; Step 7: Call the optimal prediction model for each type of flat office building to predict the energy consumption of the broken line flat office building.

2. The energy consumption prediction method for a zigzag plan office building according to claim 1 is characterized by: In step 1, the plane office building parameter information includes the width, depth and angle of the plane office building; the classification process is divided into non-fold plane, one-fold plane and two-fold plane according to the shapes of different plane office buildings.

3. The energy consumption prediction method for a zigzag plan office building according to claim 2 is characterized by: In step 2, the process of establishing the sample library illustrates the classified plane office buildings and determines the plane degrees of freedom; defines the numerical boundary range based on the plane office building parameter information; and then uses the Python platform to traverse the shape samples of each type of plane office building according to the defined numerical boundary range, as the sample library of each type of plane office building.

4. The method for predicting energy consumption of a zigzag plan office building according to claim 1, characterized in that: In step 3, the annual dynamic energy consumption simulation adopts the EnergyPlus engine.

5. The method for predicting energy consumption of a zigzag plan office building according to claim 2, characterized in that: In step 4, the process of selecting variables is to introduce trigonometric functions into the collected plane office building parameter information through four arithmetic operations to make parameter combinations, traverse the parameter combinations of four arithmetic operations and trigonometric functions, and calculate the correlation coefficient between the result of each parameter combination and the energy consumption data through the Spearman formula, and use the parameter combination with the highest correlation as the variable of each type of plane office building; The specific calculation formula of the Spearman formula is: Where ρ is the Spearman correlation coefficient, ranging from [-1,1], d i is the rank difference between two variables, n is the sample size; The principles of variable screening include: the number of selected variables does not exceed two, the selected variables include parameters that control the corresponding plane office building, and the selected variables have the highest correlation with energy consumption.

6. The method for predicting energy consumption of a zigzag plan office building according to claim 5, characterized in that: In step 5, the functional relationship of the random forest prediction model is: In the formula, is the output parameter predicted by the random forest prediction model, RF(x) is the ensemble learning function of the decision tree, x is the input parameter of the random forest prediction model, N is the number of decision trees in the random forest model, and i represents the number of the decision tree. is the output parameter predicted by the i-th decision tree; The prediction function expression of a single decision tree of the random forest prediction model is: In the formula, Tree i (x) is the prediction function of a single decision tree; The training process of the random forest prediction model is to construct a feature data set and a target variable value based on the variables of each type of flat office building obtained by variable screening, and divide the feature data set into a test set and a training set; Based on the target variable value, the decision trees are trained independently by randomly sampling variables from the feature dataset.

7. The method for predicting energy consumption of a zigzag plan office building according to claim 5, characterized in that: The neural network prediction model includes an input layer, an output layer and a hidden layer, and the functional relationship of the neural network prediction model is: In the formula, is the prediction output of the neural network prediction model, NN(X) is the prediction output function of the neural network prediction model; n last_hidden is the number of neurons in the last hidden layer of the neural network prediction model, is the weight from the jth neuron in the last hidden layer of the neural network prediction model to the output layer neuron, is the output of the jth neuron in the last hidden layer of the neural network prediction model, b out is the bias term of the output layer of the neural network prediction model, and X is the input parameter of the neural network prediction model; The output function of a single hidden neuron in the hidden layer is: am 1m NReLU(Σ n w 1mn P.S n +b 1m )4 Where a1 m Represents the output parameter of the mth neuron in the hidden layer of the neural network prediction model, w 1mn Represents the weight matrix between the input layer and the hidden layer of the neural network prediction model, P n represents the input parameter vector, b 1m is the deviation vector between the input layer and the hidden layer, n is the input layer neuron number, m is the hidden layer neuron number, The output function of the output layer is expressed as: In the formula, represents the output parameter of the kth neuron in the output layer, w 2km represents the weight matrix between the hidden layer and the output layer, b 2k represents the bias vector between the hidden layer and the output layer, and k represents the neuron number of the output layer; The specific training process of the neural network prediction model is to feed the plane data and energy consumption data in the sample library into the model in batches through multiple iterations, calculate the loss and gradient, and then update the weight of the neural network prediction model to complete the training; The loss uses mean square error MSE as the loss function, and the calculation formula is as follows: In the formula, is the true value of the i-th sample, is the predicted value of the i-th sample, that is, the output of the model, and N is the number of samples; The gradient is the partial derivative of the loss function with respect to the parameters of the neural network prediction model, and is calculated as follows: Where θ is all the trainable parameters of the neural network prediction model, namely the weight w and the bias b, is the partial derivative of the loss function with respect to the i-th parameter, L is the loss function, The specific calculation process of the gradient is: First, through forward propagation, the input data is calculated through the neural network to obtain the predicted value y pred , then calculate the loss function value L; then through back propagation, starting from the output layer, calculate the partial derivative of the loss function for each parameter layer by layer; finally, use the Adam optimizer to update the model parameters according to the gradient: In the formula, θ new is the new parameter of the neural network prediction model, θ old is the original parameter of the neural network prediction model, and η is the learning rate.

8. The method for predicting energy consumption of a zigzag plan office building according to claim 7, characterized in that: In step 6, the specific process of the grid search method is to generate all possible parameter combinations, then create, compile and train a neural network prediction model for each set of parameters, and then make predictions on the training set and the test set; The optimization process includes adjusting the number of hidden layers, the number of neurons and the number of iterations of the neural network prediction model according to the results predicted on the training set and the test set, and determining the depth of the subtree and the sample weight of the leaf node of the random forest prediction model.

Images