Building energy consumption and comfort target optimization method and system based on neural network and genetic algorithm

Abstract

The invention relates to the technical field of building energy saving and comfort level optimization, in particular to a building energy consumption and comfort level target optimization method and system based on a neural network and a genetic algorithm, by adopting the method provided by the invention, the weight and bias of a BP neural network are optimized through a grey wolf algorithm (GWO), and the prediction precision is improved; on the basis of a GWOBP model, NSGA-II is used for multi-objective optimization, a Pareto optimal solution is obtained, the optimal combination of the near-zero-energy-consumption residential building envelope structure (the types and thicknesses of wall bodies and roof thermal insulation materials, the types of windows and the horizontal sunshade overhanging length) and the number of ventilation times is selected, and the number of ventilation times is calculated. The problem of multi-objective optimization of annual primary energy consumption and annual uncomfortable degree hours of the residential building is solved, and an optimal balance point between the energy consumption and the comfort degree is found.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building energy saving and comfort optimization, in particular to a building energy consumption and comfort target optimization method and system based on neural network and genetic algorithm. BACKGROUND

[0002] The existing optimization research of near zero energy buildings focuses on a single target (such as energy consumption or thermal comfort), and lacks collaborative optimization of multi-target conflict relationship. Traditional methods such as genetic algorithm and particle swarm algorithm are prone to fall into local optimum when solving nonlinear and high-dimensional problems, and the prediction model has insufficient accuracy. In addition, the complex nonlinear relationship between building envelope parameters (such as insulation layer thickness, external window type) and energy consumption and comfort is difficult to capture through a simple regression model. SUMMARY

[0003] The purpose of the present application is to provide a building energy consumption and comfort target optimization method and system based on neural network and genetic algorithm to solve the above problems in the prior art.

[0004] The present application is achieved by the following technical solutions: In a first aspect, the present application provides a building energy consumption and comfort target optimization method based on neural network and genetic algorithm, comprising: Obtaining a Latin hypercube sampling to generate a multi-dimensional data set, the data set including envelope parameters, horizontal sunshade overhang length and air change frequency, normalizing the data set, and dividing the normalized data set into training set and test set; Inputting the training set into the GWOBP model for calculation, and evaluating the GWOBP prediction results through the test set, and outputting the GWOBP model after training; Taking the current envelope parameters, horizontal sunshade overhang length and air change frequency as input, and taking the annual primary energy consumption and annual uncomfortable hours as output, dynamically adjusting the weights and biases of the BP neural network in the GWOBP model through the grey wolf optimization algorithm, and obtaining the prediction of the annual primary energy consumption and the annual uncomfortable hours; Based on the GWOBP model and combining the NSGA-II algorithm, a multi-objective optimization model is established, taking the annual primary energy consumption and the annual uncomfortable hours as optimization targets, and generating a Pareto frontier solution set; Selecting the optimal parameter combination from the Pareto frontier solution set, and outputting the envelope optimization scheme.

[0005] Preferably, the Latin hypercube sampling generates a multi-dimensional data set comprising:

[0006] wherein a multi-dimensional vector for the jth sample point, a minimum value of the ith variable, a maximum value of the ith variable, (j) a permutation function of the ith variable, input is the sample index j, output is the sub-interval number (1 to n), a uniform random number of the ith variable in the jth sample, ~U(0,1), n is the sample number, d is the variable dimension, i is the variable index, and j is the sample index.

[0007] Preferably, the grey wolf optimization algorithm comprises: defining a behavior model of grey wolves tracking prey; establishing a position relationship model of grey wolf individuals tracking prey; setting an attack condition, when the attack condition is reached, outputting the result of the current model.

[0008] Preferably, the behavior model of grey wolves tracking prey comprises:

[0009] wherein t is the current generation number, and represent the positions of the grey wolves of the tth generation and the t+1th generation, represents the position of the prey in the tth iteration, is the distance between the grey wolf and the prey, and is a coefficient vector:

[0010] wherein is a convergence factor, which is a control parameter that linearly decreases from 2 to 0 as the iteration number increases, and is a random number between [0, 1] taken by the modulus of

[0011] Preferably, the position relationship model of grey wolf individuals tracking prey comprises:

[0012] wherein, and respectively represent the distances between the alpha, beta and delta wolves and the prey; , , and respectively represent the current positions of the alpha, beta and delta wolves; , , is a random vector, is the current position of the grey wolf. , , updating the positions of alpha wolf, beta wolf and delta wolf, , , is a convergence factor.

[0013] Preferably, the setting attack conditions comprises: when the value of the linearly decreases from 2 to 0, the corresponding the value of also varies within the interval [-a, a], when the wolf pack launches an attack on the prey.

[0014] Preferably, the NSGA-II algorithm comprises: dominance relation judgment, non-dominated sorting, and crowdedness calculation; the non-dominated sorting comprises: dividing the population P into multiple non-dominated front layers, , , initialization, for each solution x P, calculating the solution set dominated by x calculating the number of solutions dominating x collecting all solutions in the first layer =0, to form , and then iterating the layering, for each solution x in , traversing , and reducing its domination count by 1. If =0, y is added to , until all solutions are layered.

[0015] Preferably, the crowdedness calculation comprises: calculating the crowded distance of solutions within the same front layer to measure the distribution density of solutions; sorting by objective function, for each objective function , arranging the solutions in in ascending order; in boundary processing, setting the crowded distance of solutions at both ends to infinity (I= ), and calculating the crowded distance of solutions within, for i solutions, the crowded distance is:

[0016] wherein, is the crowded distance of individual i, and These are the maximum and minimum values ​​of the objective function m in the current layer, where M is the number of objective functions. and Let m be the function values ​​of individual i for each of the two individuals adjacent to the objective function m.

[0017] Preferably, the selection of the optimal parameter combination from the Pareto front solution set includes: The ideal point method is used to select the optimal parameter combination from the Pareto front solution set, including:

[0018] in( , ) represents the coordinates of the corresponding ideal point, ( , ( ) represents the coordinates before Pareto; after calculating the distances between each point in the optimal solution set and the ideal point, for the point with the smallest distance, the optimal solution is... for: .

[0019] Secondly, the present invention also provides a building energy consumption and comfort target optimization system based on neural networks and genetic algorithms, comprising: Data acquisition module: Obtains building performance data related to near-zero energy residential buildings through building energy consumption simulation software and Latin hypercube sampling; Prediction model building module: Trains energy consumption and comfort prediction models using the GWOBP algorithm; Multi-objective optimization module: Generates Pareto optimal solution sets based on NSGA-II; Decision output module: Combines the ideal point method to output the optimal parameter combination and performance indicators.

[0020] The technical solution of the present invention has at least the following advantages and beneficial effects: The method provided in this invention optimizes the weights and biases of the BP neural network using the Grey Wolf (GWO) algorithm to improve prediction accuracy. Based on the GWOBP model, NSGA-II is used for multi-objective optimization to obtain the Pareto optimal solution. The optimal combination of the building envelope (type and thickness of wall and roof insulation materials, window type, and horizontal shading overhang length) and air exchange rate of the near-zero energy residential building is selected to solve the multi-objective optimization problem of the annual energy consumption and the number of uncomfortable hours of the residential building throughout the year, and to find the best balance between energy consumption and comfort. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Fig. 1 This is a schematic diagram of the process of the present invention; Fig. 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] The independently described modules or sub-modules may or may not be physically separated; they may be implemented in software or hardware, and some modules or sub-modules may be implemented in software, with the processor calling the software to implement the function of these modules or sub-modules, while other modules or sub-modules may be implemented in hardware, such as through hardware circuits. Furthermore, some or all of the modules can be selected to achieve the purpose of this application's solution according to actual needs.

[0025] Please refer to Figs. 1-2 This invention provides a method for optimizing building energy consumption and comfort objectives based on neural networks and genetic algorithms, comprising: S101: Obtain a multi-dimensional dataset generated by Latin hypercube sampling. The dataset includes building envelope parameters, horizontal shading overhang length and air exchange rate. Normalize the dataset and divide the normalized dataset into training and test sets. In this invention, a multi-dimensional dataset can be generated by Latin hypercube sampling using energy consumption simulation software.

[0026] Data normalization is a crucial preprocessing step when building and training neural network models. The formula for normalizing input data is as follows:

[0027] in, and These are the maximum and minimum values ​​in the data. Normalized value For input values.

[0028] S102: Input the training set into the GWOBP model for computation, evaluate the GWOBP prediction results through the test set, and output the completed GWOBP model. Specifically, this includes neuron settings: The input layer of a neural network receives input data, and the number of neurons is determined by the number of features in the input data. In the code settings, the number of neurons in the input layer is calculated using `inputnum=size(P,1)`, where P is the input data matrix. This input layer has 7 neurons, corresponding to 7 features: type of external wall insulation material, thickness of external wall insulation material, type of roof insulation material, thickness of roof insulation material, type of external window, length of horizontal shading overhang, and number of air changes. The number of neurons in the hidden layer is calculated using `hiddennum=2`. The inputnum+1 operation yields 15 neurons. The number of neurons in the output layer is calculated using outputnum=size(T,1), where T is the target data matrix. This output layer has 2 neurons, corresponding to the two output features: annual energy consumption and annual uncomfortable hours.

[0029] Set the activation function: Each hidden layer neuron receives a weighted sum of inputs from the input layer or the previous hidden layer, and then processes it through an activation function. The activation function used in the hidden layer is the logistic sigmoid function, whose formula is:

[0030] in, For activation function, This is the current input value.

[0031] Input layer The seven building envelope design parameters are transferred to the hidden layer respectively, resulting in... Node output: =

[0032] in, These are the weights from input layer neuron i to hidden layer neuron j. Hidden layer neurons The threshold of the input layer The output of .

[0033] The result Passing output to the output layer And annual primary energy consumption and annual uncomfortable hours: =

[0034] In the formula, The threshold value for the k-th node of the output layer.

[0035] Prediction accuracy: To measure the accuracy of the GWOBP prediction model, the coefficient of determination R² and RMSE can be used to evaluate its performance, as shown in the following formulas:

[0036]

[0037] in, It is the first One training value, It is the first One predicted value, It is the average of the training values. It represents the total number of training values.

[0038] S103: The current building envelope parameters, horizontal shading overhang length and ventilation frequency are used as inputs, and the annual primary energy consumption and annual discomfort hours are used as outputs. The weights and biases of the BP neural network in the GWOBP model are dynamically adjusted through the Grey Wolf Optimization Algorithm to obtain the prediction of annual primary energy consumption and annual discomfort hours. S104: A multi-objective optimization model is established based on the GWOBP model and combined with the NSGA-II algorithm. The Pareto front solution set is generated with the annual primary energy consumption and the annual number of uncomfortable hours as optimization objectives. S105: Select the optimal parameter combination from the Pareto front solution set and output the optimized enclosure structure scheme.

[0039] The method provided in this invention optimizes the weights and biases of the BP neural network using the Grey Wolf (GWO) algorithm to improve prediction accuracy. Based on the GWOBP model, NSGA-II is used for multi-objective optimization to obtain the Pareto optimal solution. The optimal combination of the building envelope (type and thickness of wall and roof insulation materials, window type, and horizontal shading overhang length) and air exchange rate of the near-zero energy residential building is selected to solve the multi-objective optimization problem of the annual energy consumption and the number of uncomfortable hours of the residential building throughout the year, and to find the best balance between energy consumption and comfort.

[0040] In one exemplary embodiment of the present invention, the Latin hypercube sampling for generating a multidimensional dataset includes:

[0041] in Let j be the multidimensional vector of the j-th sample point. Let i be the minimum value of the i-th variable. Let be the maximum value of the i-th variable. (j) is the permutation function of the i-th variable. The input is the sample index j, and the output is the sub-interval number (1 to n). The i-th variable is a uniformly random number in the j-th sample. ~U(0,1), where n is the number of samples, d is the variable dimension, i is the variable index, and j is the sample index.

[0042] In one exemplary embodiment of the present invention, the gray wolf optimization algorithm includes: In GWO, the social hierarchy of gray wolves defines the top three best wolves (optimal solutions) as α, β, and δ, which guide other wolves in their search towards the target. The remaining wolves (candidate solutions) are defined as ω, and they update their positions around α, β, and δ.

[0043] Define a behavioral model of gray wolves tracking prey; establish a positional relationship model of individual gray wolves tracking prey; set attack conditions, and output the result of the current model when the attack conditions are met.

[0044] Specifically, the defined behavioral model of gray wolves tracking prey includes: In the hunting process, the behavior of gray wolves tracking prey is defined as follows:

[0045] Where t is the current generation number. and This represents the positions of the gray wolves in generations t and t+1. This represents the position of the prey in the t-th iteration. It is the distance between the gray wolf and its prey. and It is a coefficient vector:

[0046] in It is the convergence factor, a control parameter that decreases linearly from 2 to 0 as the number of iterations increases. and The modulo of is taken as a random number between [0,1].

[0047] Gray wolves can locate prey and surround them. Once the gray wolves have located the prey, the beta and delta wolves, led by the alpha wolf, guide the pack to surround the prey. In the decision space of the optimization problem, the optimal solution (the location of the prey) is unknown.

[0048] The establishment of the positional relationship model for an individual gray wolf tracking its prey includes: To simulate the hunting behavior of gray wolves, it is assumed that α, β, and δ wolves have a better understanding of the prey's potential location. The three optimal solutions obtained so far are stored, and their positions are used to determine the prey's location. To simulate the gray wolf's hunting behavior, other gray wolf individuals (including ω) are forced to update their positions based on the position of the optimal gray wolf individual, gradually approaching the prey. The mathematical model for gray wolf individuals tracking the prey's location is described below:

[0049] in, and α, β, and δ represent the distances between the wolf and its prey, respectively; , ,and These represent the current positions of α, β, and δ, respectively. , , It is a random vector. This is the current position of the Grey Wolves. , , Updated as the positions of α wolf, β wolf, and δ wolf change. , , is the convergence factor.

[0050] The attack conditions to be set include: When the prey stops moving, the gray wolf attacks to complete the hunt. This simulates approaching the prey. The value of is gradually decreased, therefore The fluctuation range also decreases accordingly. In other words, during the iteration process, when When the value of decreases linearly from 2 to 0, its corresponding The value also varies within the interval [-a, a], when Wolves will attack their prey.

[0051] In one exemplary embodiment of the present invention, the NSGA-II algorithm includes: After determining the dominance relationship and sorting the non-dominance relationships, the crowding degree is calculated. Among them, the dominance relationship judgment includes the condition that solution x dominates solution y (denoted as x < y) for all objective functions:

[0052] There exists at least one objective function:

[0053] In the formula, for…, M is...[A1].

[0054] The non-dominated sorting includes: dividing the population P into multiple non-dominated front layers. , ,... , Initialize the solution x for each solution x. P, calculate the solution set dominated by x. Calculate the number of solutions that dominate x. Collect all in the first layer The solutions that equal 0 constitute Then, iterative layering is performed. For each solution x in the equation, iterate through... Count its dominance Subtract 1. If =0, then add y This continues until all solutions are layered.

[0055] The congestion calculation includes: For the same frontier layer The congestion distance within the solution is calculated to measure the distribution density of the solution; Sort by objective function, for each objective function ,Will The solution in the middle Sort in ascending order; In boundary treatment, the crowding distance between the two endpoints is set to infinity (I= ), calculate its internal solutions, and for the i solutions, the crowding distance is:

[0056] in, For individual i, the crowding distance and These are the maximum and minimum values ​​of the objective function m in the current layer, where M is the number of objective functions. and Let m be the function values ​​of individual i for each of the two individuals adjacent to the objective function m.

[0057] In one exemplary embodiment of the present invention, selecting the optimal parameter combination from the Pareto front solution set includes: The ideal point method is used to select the optimal parameter combination from the Pareto front solution set, including:

[0058] in( , ) represents the coordinates of the corresponding ideal point, ( , ( ) represents the coordinates before Pareto; after calculating the distances between each point in the optimal solution set and the ideal point, for the point with the smallest distance, the optimal solution is... for: .

[0059] Furthermore, this invention also provides a building energy consumption and comfort target optimization system based on neural networks and genetic algorithms, comprising: Data acquisition module: Obtains building performance data related to near-zero energy residential buildings through building energy consumption simulation software and Latin hypercube sampling; Prediction model building module: Trains energy consumption and comfort prediction models using the GWOBP algorithm; Multi-objective optimization module: Generates Pareto optimal solution sets based on NSGA-II; Decision output module: Combines the ideal point method to output the optimal parameter combination and performance indicators.

[0060] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0061] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer software product, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing building energy consumption and comfort objectives based on neural networks and genetic algorithms, characterized in that, include: A multi-dimensional dataset is generated by Latin hypercube sampling. The dataset includes building envelope parameters, horizontal shading overhang length, and air exchange rate. The dataset is normalized, and the normalized dataset is divided into training and test sets. The training set is input into the GWOBP model for computation, and the GWOBP prediction results are evaluated using the test set. The completed GWOBP model is then output. Using the current building envelope parameters, horizontal shading overhang length, and air exchange rate as inputs, and the annual primary energy consumption and annual discomfort hours as outputs, the weights and biases of the BP neural network in the GWOBP model are dynamically adjusted through the Grey Wolf optimization algorithm to obtain predictions of annual primary energy consumption and annual discomfort hours. A multi-objective optimization model was established based on the GWOBP model and combined with the NSGA-II algorithm. The Pareto front solution set was generated with the annual primary energy consumption and the annual number of uncomfortable hours as optimization objectives. The optimal parameter combination is selected from the Pareto front solution set, and the optimized enclosure structure scheme is output.

2. The method for optimizing building energy consumption and comfort targets based on neural networks and genetic algorithms according to claim 1, characterized in that, The Latin hypercube sampling generates a multi-dimensional dataset including: in Let j be the multidimensional vector of the j-th sample point. Let i be the minimum value of the i-th variable. Let be the maximum value of the i-th variable. (j) is the permutation function of the i-th variable. The input is the sample index j, and the output is the sub-interval number (1 to n). The i-th variable is a uniformly random number in the j-th sample. ~U(0,1), where n is the number of samples, d is the variable dimension, i is the variable index, and j is the sample index.

3. The method for optimizing building energy consumption and comfort objectives based on neural networks and genetic algorithms according to claim 1, characterized in that, The gray wolf optimization algorithm includes: Define a behavioral model of gray wolves tracking prey; Establish a positional relationship model for individual gray wolves tracking their prey; Set attack conditions, and when the attack conditions are met, output the result of the current model.

4. The method for optimizing building energy consumption and comfort targets based on neural networks and genetic algorithms according to claim 3, characterized in that, The defined behavioral model of gray wolves tracking prey includes: Where t is the current generation number. and This represents the positions of the gray wolves in generations t and t+1. This represents the position of the prey in the t-th iteration. It is the distance between the gray wolf and its prey. and It is a coefficient vector: in It is the convergence factor, a control parameter that decreases linearly from 2 to 0 as the number of iterations increases. and The modulo of is taken as a random number between [0,1].

5. The method for optimizing building energy consumption and comfort targets based on neural networks and genetic algorithms according to claim 4, characterized in that, The establishment of the positional relationship model for an individual gray wolf tracking its prey includes: in, and α, β, and δ represent the distances between the wolf and its prey, respectively; , ,and These represent the current positions of α, β, and δ, respectively. , , It is a random vector. This is the current position of the Grey Wolves. , , Updated as the positions of α wolf, β wolf, and δ wolf change. , , is the convergence factor.

6. The method for optimizing building energy consumption and comfort objectives based on neural networks and genetic algorithms according to claim 5, characterized in that, The attack conditions to be set include: when When the value of decreases linearly from 2 to 0, its corresponding The value also varies within the interval [-a, a], when The wolf pack launched an attack on their prey.

7. The method for optimizing building energy consumption and comfort targets based on neural networks and genetic algorithms according to claim 6, characterized in that, The NSGA-II algorithm includes: After determining the dominance relationship and sorting the non-dominance relationships, the crowding degree is calculated. The non-dominated sorting includes: dividing the population P into multiple non-dominated front layers. , ,... , Initialize the solution x for each solution x. P, calculate the solution set dominated by x. Calculate the number of solutions that dominate x. Collect all in the first layer The solutions that equal 0 constitute Then, iterative layering is performed. For each solution x in the equation, iterate through... Count its dominance Subtract 1. If =0, then add y This continues until all solutions are layered.

8. The method for optimizing building energy consumption and comfort targets based on neural networks and genetic algorithms according to claim 7, characterized in that, The congestion calculation includes: For the same frontier layer The congestion distance within the solution is calculated to measure the distribution density of the solution; Sort by objective function, for each objective function ,Will The solution in the middle Sort in ascending order; In boundary treatment, the crowding distance between the two endpoints is set to infinity (I= ), calculate its internal solutions, and for the i solutions, the crowding distance is: in, For individual i, the crowding distance and These are the maximum and minimum values ​​of the objective function m in the current layer, where M is the number of objective functions. and Let m be the function values ​​of individual i for each of the two individuals adjacent to the objective function m.

9. The method for optimizing building energy consumption and comfort targets based on neural networks and genetic algorithms according to claim 7, characterized in that, The selection of the optimal parameter combination from the Pareto front solution set includes: The ideal point method is used to select the optimal parameter combination from the Pareto front solution set, including: in( , ) represents the coordinates of the corresponding ideal point, ( , ( ) represents the coordinates before Pareto; after calculating the distances between each point in the optimal solution set and the ideal point, for the point with the smallest distance, the optimal solution is... for: .

10. A building energy consumption and comfort target optimization system based on neural networks and genetic algorithms, characterized in that, include: Data acquisition module: Obtains building performance data related to near-zero energy residential buildings through building energy consumption simulation software and Latin hypercube sampling; Prediction model building module: Trains energy consumption and comfort prediction models using the GWOBP algorithm; Multi-objective optimization module: Generates Pareto optimal solution sets based on NSGA-II; Decision output module: Combines the ideal point method to output the optimal parameter combination and performance indicators.

Images