A method for optimizing and controlling indoor air conditioning temperature set point of public buildings

Abstract

In view of the problem that the traditional static control strategy of air conditioning system has high energy consumption and unbalanced comfort degree when the environment changes, a three-dimensional model of building is constructed, a joint simulation environment is built by fusing sensor data and energy consumption simulation software, an improved extreme learning machine neural network is used to predict air conditioning energy consumption and personnel thermal comfort, a rotating particle swarm optimization algorithm is used to optimize the network structure and parameters, and the prediction accuracy and generalization ability are improved, then, a Newton-Raphson multi-objective optimization algorithm is combined to rollingly solve the optimal dynamic sequence of air conditioning temperature set value, and the energy consumption and comfort degree are balanced through the Pareto front solution set, so that the air conditioning precise regulation and control are realized, and the thermal comfort is improved.

Description

Technical Field

[0001] This invention belongs to the field of modeling and optimization control technology of building thermal environment, and particularly relates to a method for optimizing and controlling the indoor air conditioning temperature setpoint of public buildings. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the deepening of my country's new urbanization strategy and the continuous expansion of the construction industry, building energy consumption has become a significant bottleneck restricting sustainable urban development. According to statistics from the Ministry of Housing and Urban-Rural Development, the total energy consumption of the construction industry currently accounts for over 35% of total social energy consumption, with the energy intensity per unit area of ​​public buildings being 2.09 times that of residential buildings. Of particular concern is that in public buildings such as large commercial complexes, transportation hubs, and super high-rise office buildings, the energy consumption of air conditioning systems has consistently remained between 40% and 60%, posing a serious challenge to achieving "dual carbon" targets.

[0004] Currently, air conditioning temperature control mostly adopts a static setpoint strategy, which is difficult to adapt to dynamic changes in indoor and outdoor environments (such as fluctuations in population density and time-varying temperature and humidity) and individual differences in people's thermal comfort needs. This leads to the air conditioning system being in a contradictory state of "over-cooling / heating" or "under-cooling / heating" for a long time. This not only causes a lot of energy waste, but also causes health risks and decreased work efficiency due to the imbalance of thermal comfort, further exacerbating the conflict between building energy consumption and user experience. The limitations of traditional control methods make it difficult for building energy management to balance energy-saving goals and people's comfort needs. Therefore, this invention proposes a method for indoor air conditioning temperature setting that integrates data-driven predictive models and model predictive control technology to balance indoor people's thermal comfort and air conditioning system energy consumption. Summary of the Invention

[0005] This invention starts with the prediction of indoor thermal comfort in public buildings and the optimization of air conditioning temperature setpoints:

[0006] To address the aforementioned problems, this invention provides a data-driven model for predicting indoor comfort and air conditioning energy consumption in public buildings, along with a method for controlling air conditioning temperature setpoints based on this predictive model. The invention imports weather files into a constructed building energy consumption simulation environment and sets the capacity and operating schedules of personnel and equipment (lighting, appliances, and air conditioning, etc.). By setting the operating mode of the air conditioning system, data on indoor environment and equipment energy consumption are acquired. Then, the data is normalized, and a predictive model is built using an extreme learning machine neural network model to predict the energy consumption of the indoor air conditioning system and the thermal comfort of personnel. The trained model is embedded into a co-simulation environment, and a multi-objective optimization algorithm is used to continuously optimize the air conditioning temperature setpoint to simultaneously balance indoor thermal comfort and energy consumption.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for optimizing and controlling indoor air conditioning temperature setpoints in public buildings includes the following steps:

[0009] Create a 3D model of the building, and set the number of people, lighting power, electrical equipment power, air conditioning system and personnel and air conditioning equipment parameters for each hot zone in the model;

[0010] Indoor and outdoor environmental data are collected using sensors and electricity meters. The data collected by the sensors is then embedded into the building energy consumption simulation model through the external interface of the building simulation software to simulate the real indoor thermal environment and energy consumption of the equipment.

[0011] Through data acquisition, storage, and visualization functions, the simulation results can be displayed in real time, and training and testing sample sets can be constructed.

[0012] An extreme learning machine neural network model is constructed, and a swarm intelligence optimization algorithm is used to optimize the network structure and parameters to determine the network structure and optimal parameters, thereby predicting the energy consumption of indoor air conditioning systems and the thermal comfort of people.

[0013] The established prediction model is embedded into the co-simulation module to achieve dynamic coupling calculation with the building energy consumption simulation model, and the optimal dynamic sequence of air conditioning temperature setpoints within the prediction interval is solved based on the multi-objective optimization algorithm.

[0014] For the obtained optimal dynamic sequence, the first element is selected as the air conditioning temperature setpoint for the next moment to avoid deviations in control performance caused by environmental changes in future moments.

[0015] As an alternative implementation method, AutoCAD, BIM, SketchUp, etc. are used to draw a 3D model of the building, and energy consumption simulation software such as TRNSYS and EnergyPlus are used to set the internal parameters of the building. The 3D model of the building and its envelope parameters are constructed based on the actual building construction drawings. The capacity of the air conditioning equipment (including COP, cooling capacity, heating capacity, etc.) is determined with reference to the actual deployed air conditioning system or HVAC construction drawings. The power of indoor electrical equipment and lighting is set according to the power of the actual deployed equipment or electrical construction drawings.

[0016] As an alternative implementation method, the number of people indoors and their hourly occupancy rate can be detected by passive infrared sensors, the operating status of electrical equipment and air conditioning systems can be detected by electricity meters, and the indoor and outdoor temperature and humidity can be detected by temperature and humidity sensors. The detected data can be transmitted to the building energy consumption simulation model through an external interface to set the work schedules of personnel and various equipment.

[0017] As an alternative implementation method, the datasets for training and testing the model need to be collected at different air conditioning temperatures at different times; after data collection, the data should be stored in an Excel spreadsheet or database.

[0018] As an alternative implementation, the neural network model is an extreme learning machine model. This model incorporates a particle swarm optimization algorithm based on the rotational motion characteristics of particles to determine the model's structure and optimal parameters. The model mainly includes an input layer, hidden layers, an output layer, and a network structure solving layer.

[0019] As a further limitation, the implementation process of the extreme learning machine prediction model is as follows:

[0020] Step 101: Select the outdoor temperature output from the building energy consumption simulation model. Indoor temperature Indoor relative humidity Indoor radiant temperature Indoor air conditioning energy consumption Indoor occupant comfort Air conditioning temperature setting As input data for the prediction model, vectors This indicates the energy consumption of the indoor air conditioning system output from the building energy consumption simulation model. thermal comfort As the output data of the prediction model, using vectors The dataset described is... ;

[0021] Step 102: Split the dataset into training and testing datasets. In order to balance feature dimensions, improve numerical stability, optimize activation function performance, and significantly improve the model's prediction accuracy, convergence speed and generalization ability, the data is preprocessed by standardization or normalization.

[0022] As an alternative implementation, the dataset is preprocessed using max-min normalization:

[0023]

[0024] In the formula, This is the original data; The input data is normalized; This is a column vector composed of the minimum values ​​of each feature of the original data; The input and output data are column vectors composed of the maximum values ​​of each feature of the original data; the range of the normalized input and output data is within... Within the range; Represents the Hadamard product;

[0025] Step 103: Train the extreme learning machine model using different activation functions on the normalized dataset and optimize it with a specific objective function;

[0026] As an alternative implementation, the specific objective function is a least squares method incorporating an L2 regularization term, as shown in the following equation:

[0027]

[0028] In the formula, This represents the number of hidden layers in the Extreme Learning Machine prediction model. Let be the weight matrix connecting the input layer and the hidden layer, where: ; Let be the bias matrix of the hidden layer, where: ; Let be the weight matrix connecting the hidden layer and the output layer, where: ; The input data for the model; The output data of the model; The output data is normalized; is the activation function of the neural network;

[0029] As an alternative implementation, the optimal activation function for the hidden layer can be sigmoid, tansig, tanh, relu, leaky_relu, elu, etc.

[0030] Step 104: Initialize the population, set the hyperparameters of the rotating particle swarm optimization algorithm, and define the motion state equations of the particles to iteratively update each particle;

[0031] As an alternative implementation, the number of hidden layers, the weight matrix connecting the input layer and the hidden layers, and the bias matrix of the hidden layers are used as particles in the particle swarm optimization algorithm, defined by vectors. Representation; definition of population size Maximum number of iterations The charge carried by each particle magnetic field strength and particle mass Initialize each population ;

[0032] As an alternative implementation method, the population initialization method is uniform random initialization:

[0033]

[0034] In the formula, Let be the component of the k-th particle in the q-th dimension of the search space; These are generated random numbers, uniformly distributed within the interval [0,1]. and These are the upper and lower bounds of the search space, respectively.

[0035] As an alternative implementation, the particle motion equation is mainly determined by the velocity of the particle during rotational motion and its velocity during basic motion, and is calculated by the following formula:

[0036]

[0037] In the formula, This represents the current iteration number; The velocity of the particle during its rotational motion; The velocity of a particle during its fundamental motion; It is the integration constant; The electric charge carried by each particle; The magnetic field strength; For particle mass; The inertial weights for particles in rotational motion; The inertial weights for particles in their basic motion; , It is a learning factor; , , It is a random number between [0, 1]. It is the individual optimal value at the current iteration number; It is the globally optimal value at the current iteration number;

[0038] Step 105: During the algorithm iteration process, a population adaptive adjustment strategy and an inertia weight dynamic optimization strategy are adopted to suppress the excessive accumulation of local optima while ensuring the full exploration of potential solutions.

[0039] As an alternative implementation, the population adaptive adjustment strategy is a particle chaotic collision strategy, and the specific mathematical model is as follows:

[0040]

[0041] In the formula, These are new particles generated by a particle collision strategy. The particles are randomly selected from the current iteration number; The result is a random number, between [0,1], and follows a uniform distribution.

[0042] As an alternative implementation method, the inertia weight dynamic optimization strategy is a particle chaotic collision strategy, and the specific mathematical model is as follows:

[0043]

[0044] Step 106: After all particles have participated in the update, if the maximum number of iterations in the optimization process has been reached, then the optimization is terminated.

[0045] Step 107: After the prediction model training is completed, the model output data is inversely normalized to test the model accuracy, as shown in the following formula:

[0046]

[0047] Step 108: Embed the tested model into the co-simulation module to predict air conditioning energy consumption and thermal comfort during the simulation process;

[0048] As an alternative implementation, the multi-objective optimization algorithm for solving the optimal dynamic sequence of air conditioning temperature setpoints within the prediction period is the Newton-Raphson multi-objective optimization algorithm. The model trained via a neural network is applied to the co-simulation module, and building energy consumption simulation software is used to calculate the indoor average thermal radiation temperature, thermal comfort, and air conditioning system energy consumption in real time. Based on the calculation results of the energy consumption simulation model and the prediction model, the Newton-Raphson multi-objective optimization algorithm is used to determine the optimal dynamic sequence of air conditioning temperature setpoints within the prediction interval.

[0049] As a further constraint, when the prediction step size is... Within a given step, the collected data is fed into the trained neural network for prediction, and the temperature setpoint within the prediction step is solved using a rolling optimization method. The objective function for optimization is shown in the following equation:

[0050]

[0051] As an alternative implementation method, the specific process of the optimal dynamic sequence optimization method for the air conditioning temperature setpoint within the prediction interval is as follows:

[0052] Step 201: Based on the data collected at the current moment and the established prediction model, update the model output within the prediction step size so that the objective function of the rolling optimization reaches the Pareto front.

[0053]

[0054] In the formula, The input data for the model within the prediction step; To predict the model's output data within the predicted step size; Other feasible solutions in the solution set; For Pareto front solutions;

[0055] Step 202: Determine the air conditioning temperature setpoint using a multi-objective optimization algorithm based on the Newton-Raphson search rule, according to the defined objective function and constraints, and define the population size. Maximum number of iterations Use Latin hypercube initialization to randomly generate N p Individual populations:

[0056]

[0057] In the formula, Let be the j-th component of the i-th element in the search space; These are generated random numbers, uniformly distributed within the interval [0,1]. and These are the upper and lower bounds of the search space, respectively. For a random permutation of elements in the j-th dimension;

[0058] Step 203: After initializing the parameters and population, substitute each element into the objective function, calculate the fitness value of each element, perform non-dominated sorting, and select the non-dominated solution for the current iteration number; then, use the ideal point method to calculate the distance between the fitness value of each element for the current iteration number and the current optimal fitness value, as shown in the following formula:

[0059]

[0060] In the formula, This represents the Euclidean distance between the fitness value of each element and the optimal fitness value at the current iteration number; Let i be the fitness value of each element for the i-th optimization objective at the current iteration number; Let be the optimal fitness value of the i-th optimization objective for each element at the current iteration number; Let be the worst fitness value of the i-th optimization objective for each element at the current iteration number;

[0061] Step 204: Select the farthest distance in the current iteration number The corresponding element is the worst position. Its j-th dimension component is Select the nearest distance in the current iteration number. The corresponding element is the worst position. Its j-th dimension component is ;Calculate using the Newton-Raphson search rule:

[0062]

[0063] In the formula, These are guiding parameters used to guide the population in the correct direction at the current iteration number; , A random number between (0,1); , Let be the j-th dimension component of an element randomly selected from the population, and ; The result is the Newton-Raphson solution;

[0064] Step 205: After calculating the value of each element in each dimension using the Newton-Raphson search rule, calculate the updated element position using the following formula:

[0065]

[0066] In the formula, This represents the j-th dimension component of the i-th element after the update. A random number between [0,1]; , , for Update parameters; This is an adaptive coefficient used to ensure population diversity;

[0067] Step 206: When the algorithm reaches the maximum number of iterations, terminate the algorithm and output the Pareto surface; then, use the ideal point method to calculate the distance between the solution sets on the Pareto surface and select the optimal solution. .

[0068] As an alternative implementation method, to minimize the impact of future environmental changes on the air conditioning control effect, the first element of the obtained optimal solution is selected. As the air conditioner temperature setting value.

[0069] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0070] This invention constructs a prediction model based on an extreme learning machine neural network, combining a rotating particle swarm optimization algorithm to optimize network parameters with a Newton-Raphson multi-objective optimization algorithm to solve for dynamic temperature setpoints, thereby achieving synergistic optimization of air conditioning system energy consumption and indoor thermal comfort. This invention utilizes a building energy consumption simulation model and sensor data to dynamically couple and establish a co-simulation model. Simultaneously, this invention uses a neural network model for real-time prediction and a multi-objective optimization algorithm to dynamically adjust the air conditioning temperature setpoint. This not only solves the problem of "over-cooling / over-heating" caused by dynamic environmental changes in traditional static control strategies, significantly reducing energy consumption, but also effectively improves adaptability to personalized comfort needs by dynamically balancing thermal comfort indices and energy efficiency through the Pareto front solution set. Furthermore, the integration of the rotating particle swarm optimization algorithm and multi-objective optimization mechanism enhances the model's generalization ability and computational efficiency, providing high-precision, real-time response technical support for the intelligent and energy-saving control of building air conditioning systems. Attached Figure Description

[0071] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0072] Figure 1 This is a schematic diagram of building co-simulation according to at least one embodiment of the present invention;

[0073] Figure 2 This is a neural network architecture diagram of at least one embodiment of the present invention;

[0074] Figure 3 A flowchart illustrating the modeling process for a prediction model according to at least one embodiment of the present invention;

[0075] Figure 4 This is a flowchart illustrating the process of solving the air conditioning temperature setpoint according to at least one embodiment of the present invention. Detailed Implementation

[0076] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0077] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0078] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0079] A method for optimizing and controlling the indoor air conditioning temperature setpoint in public buildings includes the following steps:

[0080] S1: Draw a 3D model of the building, and set the number of people in each room, the power of the lighting fixtures, the power of the electrical equipment, the air conditioning system of each room, and the parameters of people and air conditioning equipment for the drawn model;

[0081] S2: Use sensors and electricity meters to collect indoor and outdoor environmental data, and embed the data collected by the sensors into the energy consumption simulation software through the external interface of the building simulation software to simulate the real indoor thermal environment and energy consumption of the equipment.

[0082] S3: Through data acquisition, storage, and visualization functions, it enables real-time display of simulation results and constructs training and testing sample sets;

[0083] S4: Construct a neural network prediction model and use a swarm intelligence optimization algorithm to optimize the network structure and parameters in order to determine the network structure and optimal parameters, thereby enabling the prediction of energy consumption and thermal comfort of indoor air conditioning systems;

[0084] S5: Embed the established prediction model into the co-simulation module to achieve dynamic coupling calculation with the building energy consumption simulation model, and solve the optimal dynamic sequence of air conditioning temperature setpoints within the prediction period based on the multi-objective optimization algorithm.

[0085] S6: For the obtained optimal dynamic sequence, select the first element as the air conditioning temperature setpoint for the next moment.

[0086] Specifically, the 3D building model mentioned in step S1 is drawn using AutoCAD, BIM, or SketchUp, and the internal parameters of the building are set using TRNSYS or EnergyPlus energy consumption simulation software. The 3D building model and its envelope parameters are constructed based on the actual building construction drawings. The capacity of the air conditioning equipment (including COP, cooling capacity, heating capacity, etc.) is determined according to the actual deployed air conditioning system or HVAC construction drawings. The power of indoor electrical equipment and lighting is set according to the actual deployed equipment power or electrical construction drawings.

[0087] Specifically, the environmental data mentioned in step S2 is measured by various sensors and electricity meters. The main environmental data to be detected are: the number of people indoors and their hourly occupancy rate can be detected by passive infrared sensors; the operating status of electrical equipment and air conditioning systems can be detected by electricity meters; and the indoor and outdoor temperature and humidity can be detected by temperature and humidity sensors. The detected data is transmitted to the building energy consumption simulation model through an external interface to set the work schedules for personnel and various equipment.

[0088] Specifically, the data acquisition, storage, and visualization functions described in step S3 involve collecting datasets for training and testing models by setting different air conditioning temperatures at different times. After data acquisition, the data is stored in an Excel spreadsheet or database.

[0089] Specifically, the neural network prediction model described in step S4 is an extreme learning machine model incorporating the rotating particle swarm optimization algorithm. This model introduces the rotating particle swarm optimization algorithm to determine the model's structure and optimal parameters. The model mainly includes an input layer, hidden layers, an output layer, and a network structure solving layer.

[0090] Specifically, the implementation process of the neural network prediction model in step S4 is as follows:

[0091] Step 101: Select the outdoor temperature output from the building energy consumption simulation model. Indoor temperature Indoor relative humidity Indoor radiant temperature With air conditioner temperature setting As input data for the prediction model, vectors This indicates the energy consumption of the indoor room air conditioning system output from the building energy consumption simulation model. thermal comfort As the output data of the prediction model, using vectors The dataset described is... ;

[0092] Step 102: Split the dataset into training and testing datasets. In order to balance feature dimensions, improve numerical stability, optimize activation function performance, and significantly improve the model's prediction accuracy, convergence speed and generalization ability, the data is preprocessed by standardization or normalization.

[0093] As an alternative implementation, the dataset is preprocessed using max-min normalization:

[0094]

[0095] In the formula, This is the original data; The input data is normalized; This is a column vector composed of the minimum values ​​of each feature of the original data; The input and output data are column vectors composed of the maximum values ​​of each feature of the original data; the range of the normalized input and output data is within... Within the range; Represents the Hadamard product;

[0096] Step 103: Train the Extreme Learning Machine model using different activation functions on the normalized dataset, and optimize it with a specific objective function;

[0097] As an alternative implementation, the specific objective function is a least squares method incorporating an L2 regularization term, as shown in the following equation:

[0098]

[0099] In the formula, This represents the number of hidden layers in the Extreme Learning Machine prediction model. Let the weight matrix be the connection between the input and the hidden layer, where: ; Let be the bias matrix of the hidden layer, where: ; Let be the weight matrix connecting the hidden layer and the output layer, where: ; The input data for the model; The output data of the model; The output data is normalized; is the activation function of the neural network;

[0100] As an alternative implementation, the optimal activation function for the hidden layer can be sigmoid, tansig, tanh, relu, leaky_relu, elu, etc.

[0101] Step 104: Initialize the population, set the hyperparameters of the rotating particle swarm optimization algorithm, and define the motion state equations of the particles to iteratively update each particle.

[0102] As an alternative implementation, the number of hidden layers, the weight matrices of the input layer and hidden layers, and the bias matrices of the hidden layers are used as particles in the particle swarm optimization algorithm, defined by vectors. Representation; definition of particle population size Maximum number of iterations The charge carried by each particle magnetic field strength and particle mass Initialize each population ;

[0103] As an alternative implementation method, the population initialization method is uniform random initialization:

[0104]

[0105] In the formula, Let be the component of the k-th particle in the q-th dimension of the search space; These are generated random numbers, uniformly distributed within the interval [0,1]. and These are the upper and lower bounds of the search space, respectively.

[0106] As an alternative implementation, the equation of motion of the particle is mainly determined by the velocity of the particle during rotational motion and the velocity during its basic motion, and is calculated by the following formula:

[0107]

[0108] In the formula, This represents the current iteration number; The velocity of the particle during its rotational motion; The velocity of a particle during its fundamental motion; It is the integration constant; The electric charge carried by each particle; The magnetic field strength; For particle mass; The inertial weights for particles in rotational motion; The inertial weights for particles in their basic motion; , It is a learning factor; , , It is a random number between [0, 1]. It is the individual optimal value at the current iteration number; It is the globally optimal value at the current iteration number;

[0109] Step 105: During the algorithm iteration process, a population adaptive adjustment strategy and an inertia weight dynamic optimization strategy are adopted to suppress the excessive accumulation of local optima while ensuring the full exploration of potential solution sets.

[0110] As an alternative implementation, the population adaptive adjustment strategy is a particle chaotic collision strategy, and the specific mathematical model is as follows:

[0111]

[0112] In the formula, These are new particles generated by a particle collision strategy. The particles are randomly selected from the current iteration number; The result is a random number, between [0,1], and follows a uniform distribution.

[0113] As an alternative implementation method, the inertia weight dynamic optimization strategy is a particle chaotic collision strategy, and the specific mathematical model is as follows:

[0114]

[0115] Step 106: After all particles have participated in the update, if the maximum number of iterations in the optimization process is reached or the objective function meets the requirements, then the optimization is terminated.

[0116] Step 107: After the prediction model training is completed, the model output data is inversely normalized to test the model accuracy, as shown in the following formula:

[0117]

[0118] Step 108: Embed the tested model into the co-simulation module to predict air conditioning energy consumption and thermal comfort during the simulation process;

[0119] Specifically, the multi-objective optimization algorithm for solving the optimal dynamic sequence of air conditioning temperature setpoints within the prediction period, as described in step S5, is the Newton-Raphson multi-objective optimization algorithm. The model trained via a neural network is applied to the co-simulation module, and building energy consumption simulation software is used to calculate the indoor average thermal radiation temperature, thermal comfort, and air conditioning system energy consumption in real time. Based on the calculation results and the prediction model, the Newton-Raphson multi-objective optimization algorithm is used to determine the optimal dynamic sequence of air conditioning temperature setpoints within the prediction interval.

[0120] Specifically, the detailed process of the optimal dynamic sequence optimization method for the air conditioning temperature setpoint within the prediction interval described in step S5 is as follows:

[0121] Step 201: Based on the data collected at the current moment and the established prediction model, update the model output within the prediction step size so that the objective function of the rolling optimization reaches the Pareto front.

[0122]

[0123] In the formula, The input data for the model within the prediction step; To predict the model's output data within the predicted step size; Other feasible solutions in the solution set; For Pareto front solutions;

[0124] Step 202: Determine the air conditioning temperature setpoint using a multi-objective optimization algorithm based on the Newton-Raphson search rule, according to the defined objective function and constraints, and define the population size. Maximum number of iterations Randomly generated using Latin hypercube initialization Individual populations:

[0125]

[0126] In the formula, Let i be the component of the i-th element in the j-th dimension of the search space; These are generated random numbers, uniformly distributed within the interval [0,1]. and These are the upper and lower bounds of the search space, respectively. For a random permutation of the elements at the j-th position;

[0127] Step 203: After initializing the parameters and population, substitute each element into the objective function, calculate the fitness value of each element, perform non-dominated sorting, and select the non-dominated solution for the current iteration number; then, use the ideal point method to calculate the distance between the fitness value of each element and the current optimal fitness value for the current iteration number, as shown in the following formula:

[0128]

[0129] In the formula, This represents the Euclidean distance between the fitness value of each element and the optimal fitness value at the current iteration number; Let be the fitness value of each element for the i-th optimization objective at the current iteration number; Let be the fitness value of each element for the i-th optimization objective at the current iteration number; Let be the worst fitness value of the element for the i-th optimization objective at the current iteration number;

[0130] Step 204: Select the farthest distance in the current iteration number The corresponding element is the worst position. Its j-th dimension component is Select the nearest distance in the current iteration number. The corresponding element is the worst position. Its j-th dimension component is ;Calculate using the Newton-Raphson search rule:

[0131]

[0132] In the formula, These are guiding parameters used to guide the population in the correct direction at the current iteration number; , A random number between (0,1); , Let be the j-th dimension component of an element randomly selected from the population, and ; The result is the Newton-Raphson solution;

[0133] Step 205: After calculating the value of each element in each dimension using the Newton-Raphson search rule, calculate the updated element position using the following formula:

[0134]

[0135] In the formula, This represents the j-th dimension component of the i-th element after the update. A random number between (0,1); , , for Update parameters; This is an adaptive coefficient used to ensure population diversity;

[0136] Step 206: When the algorithm reaches the maximum number of iterations, terminate the algorithm and output the Pareto surface; then, use the ideal point method to calculate the distance between the solution sets on the Pareto surface and select the optimal solution. .

Claims

1. A method for optimizing and controlling the indoor air conditioning temperature setpoint in public buildings, characterized in that, include: Building energy consumption simulation models are built using building energy consumption simulation software. These models output key parameters such as the number of people indoors, indoor thermal comfort, indoor and outdoor environment, and air conditioning system energy consumption. The data visualization function of building energy consumption simulation software is used to visualize the collected data (data collection charts, optimization result diagrams, etc.) and obtain the training sample set and test sample set for training the prediction model. A neural network prediction model was constructed, and optimization algorithms were used to optimize the network structure and parameters to predict the thermal comfort of indoor occupants and the energy consumption of the air conditioning system. With a prediction step size of Within a given step, the collected data is fed into the trained neural network for prediction, and the temperature setpoint within the prediction step is solved using a rolling optimization method. The objective function for rolling optimization includes: ； By using the co-simulation function of building energy consumption simulation software, prediction models and optimization algorithms are embedded into the building energy consumption simulation model to achieve data acquisition, data preprocessing, and optimization and control of air conditioning temperature setpoints. The method for optimizing and controlling the air conditioning temperature setpoint is the Newton-Raphson multi-objective optimization algorithm. It uses the Newton-Raphson search rule to solve the defined objective function. The specific process includes: Step 101: Based on the data collected at the current moment and the established prediction model, update the model output within the prediction step size so that the objective function of the rolling optimization reaches the Pareto front. ； in, The input data for the model within the prediction step; To predict the model's output data within the predicted step size; Other feasible solutions in the solution set; For Pareto front solutions; Step 102: Determine the air conditioning temperature setpoint using a multi-objective optimization algorithm based on the Newton-Raphson search rule, according to the defined objective function and constraints, and define the population size. Maximum number of iterations Randomly generated using Latin hypercube initialization Individual populations: ; in, Let i be the component of the i-th element in the j-th dimension of the search space; These are generated random numbers, uniformly distributed within the interval [0,1]. and These are the upper and lower bounds of the search space, respectively. For a random permutation of the elements at the j-th position; Step 103: After initializing the parameters and population, substitute each element into the objective function, calculate the fitness value of each element, perform non-dominated sorting, and select the non-dominated solution at the current iteration number; then, use the ideal point method to calculate the distance between the fitness value of each element at the current iteration number and the current optimal fitness value, as shown in the following formula: ； in, This represents the Euclidean distance between the fitness value of each element and the optimal fitness value at the current iteration number; For each element in the current iteration number, the i-th i The fitness value corresponding to each optimization objective; Let be the fitness value of each element for the i-th optimization objective at the current iteration number; Let be the worst fitness value of the element for the i-th optimization objective at the current iteration number; Step 104: Select the farthest distance in the current iteration number The corresponding element is the worst position. Its j-th dimension component is Select the nearest distance in the current iteration number. The corresponding element is the worst position. Its j-th dimension component is ;Calculate using the Newton-Raphson search rule: ; in, These are guiding parameters used to guide the population in the correct direction at the current iteration number; , A random number between (0,1); , Let be the j-th dimension component of an element randomly selected from the population, and ; The result is the Newton-Raphson solution; Step 105: After calculating the value of each element in each dimension using the Newton-Raphson search rule, calculate the updated element position using the following formula: ； in, This represents the j-th dimension component of the i-th element after the update. A random number between (0,1); , , for Update parameters; This is an adaptive coefficient used to ensure population diversity; Step 106: When the algorithm reaches the maximum number of iterations, terminate the algorithm and output the Pareto surface; then, use the ideal point method to calculate the distance between the solution sets on the Pareto surface and select the optimal solution. ; Based on the established prediction model, the prediction interval and objective function are defined, and the optimal air conditioning temperature setpoint is solved using a multi-objective optimization algorithm. After the solution process is completed, the results are input into the external interface of the building energy consumption simulation model for joint calculation until the simulation process ends.

2. The method for optimizing and controlling the indoor air conditioning temperature setpoint in public buildings as described in claim 1, characterized in that, The building energy consumption simulation model is characterized by including: Use AutoCAD, BIM, or SketchUp to create a 3D building model. Set the number of people in each hot zone, the power of lighting fixtures, the power of electrical equipment, and the air conditioning system for each room in the model. Use sensors and electricity meters to collect indoor and outdoor environmental data. Embed the data collected by the sensors into the building energy consumption simulation model through the external interface of the building simulation model to set the work schedules of personnel and equipment. Through data acquisition and visualization functions, realize the real-time display of simulation results and build training and test sample sets.

3. The method for optimizing and controlling the indoor air conditioning temperature setpoint in public buildings as described in claim 1, characterized in that, include: The neural network prediction model is an Extreme Learning Machine (ELM) neural network; this neural network integrates the ELM network structure and the rotating particle swarm optimization algorithm, and includes an input layer, hidden layers, an output layer, and a network structure solving layer.

4. The method for optimizing and controlling the indoor air conditioning temperature setpoint in public buildings as described in claim 1, characterized in that, The implementation process of the neural network prediction model includes: Step 201: Select the outdoor temperature output from the building energy consumption simulation model. Indoor temperature Indoor relative humidity Indoor radiant temperature Indoor air conditioning energy consumption Indoor occupant comfort Air conditioning temperature setting As input data for neural network prediction models, vectors This indicates the energy consumption of the indoor room air conditioning system output from the building energy consumption simulation model. With human thermal comfort As output data of a neural network prediction model, using vectors express; Step 202: Perform max-min normalization on the dataset: ； in, This is the original data; The input data is normalized; It is a matrix composed of the minimum values ​​of each feature of the original data; This is a matrix composed of the maximum values ​​of each feature of the original data; the range of the normalized input and output data is within... Within the range; Represents the Hadamard product; Step 203: Input the data into the Extreme Learning Machine neural network for training. The objective function is shown in the following equation: ； in, The number of hidden layers in the Extreme Learning Machine neural network; Let the weight matrix be the connection between the input and the hidden layer, where: ; Let be the bias matrix of the hidden layer, where: ; Let be the weight matrix connecting the hidden layer and the output layer, where: ; The input data for the model; The output data of the model; The activation function for the neural network can be selected from sigmoid, tansig, tanh, relu, leaky_relu, elu, etc. Step 204: The number of hidden layers, the weights connecting the input layer to the hidden layers, and the biases of the hidden layers are used as particles in the particle swarm optimization algorithm. Indicate; define population size Maximum number of iterations The charge carried by each particle magnetic field strength and particle mass Initialize each population As shown in the following formula: ； in, Let r be the component of the k-th particle in the q-th dimension of the search space; r is a generated random number, uniformly distributed in the interval [0,1]. and These are the upper and lower bounds of the search space, respectively. ； in, This represents the number of iterations. The velocity of the particle during its rotational motion; The velocity of a particle during its fundamental motion; It is the integration constant; The electric charge carried by each particle; The magnetic field strength; For particle mass; The inertial weights for particles in rotational motion; The inertial weights for particles in their basic motion; , It is a learning factor; , , It is a random number between [0, 1]. It is the individual optimal value at the current iteration number; It is the globally optimal value at the current iteration number; Step 205: To avoid the particle swarm optimization algorithm getting stuck in local optima, a particle chaotic collision strategy and an adaptive adjustment strategy for inertia weights are used to prevent the population from getting stuck in local optima, as shown in the following formula: ； in, These are new particles generated by a particle collision strategy. The particles are randomly selected from the current iteration number; The value is a random number, between [0,1], and follows a uniform distribution. If the fitness value calculated by substituting the updated particle into the objective function is less than the fitness value of the particle before the update, then the new particle is selected; otherwise, the original particle is retained and proceeds to the next iteration. Step 206: After all particles have participated in the update, if the maximum number of iterations in the optimization process has been reached, then the optimization is terminated. Step 207: After the prediction model training is complete, the data output by the model is denormalized, and the model accuracy is tested, as shown in the following formula: ； Step 208: Embed the tested model into the co-simulation module to predict air conditioning energy consumption and thermal comfort during the simulation process.

5. The method for optimizing and controlling the indoor air conditioning setpoint of a public building as described in claim 1, characterized in that, The setpoint optimization method is the Newton-Raphson multi-objective optimization algorithm, which specifically includes: The trained Extreme Learning Machine neural network was applied to a building energy consumption simulation model; Data is acquired through building energy consumption simulation models, and extreme learning machine neural networks are used to predict air conditioning energy consumption and human comfort. The Newton-Raphson multi-objective optimization algorithm was used for optimization. Based on the optimization results, update the air conditioning temperature setting for the next moment.

6. The method for optimizing and controlling the indoor air conditioning temperature setpoint in public buildings as described in claim 1, characterized in that, To ensure that the air conditioning control effect is as unaffected as possible by future environmental changes, the first element of the obtained optimal solution is taken as the air conditioning temperature setpoint.

Images