Intelligent energy-saving control method and system for green building equipment based on artificial intelligence

Abstract

The present invention relates to the technical field of intelligent building energy-saving control, and discloses an intelligent energy-saving control method and system for green building equipment based on artificial intelligence, including: collecting building environment data and performing preprocessing. Using a digital twin model to simulate the energy consumption of building equipment, training a baseline neural network model, and obtaining an initial control strategy. Using a genetic algorithm to optimize the hyperparameters of the baseline model, a plurality of candidate energy-saving control strategies are generated. The optimized strategy is input into a deep reinforcement learning model for strategy iteration. After each iteration, the energy-saving efficiency and comfort score are calculated, and the optimal strategy is screened through a reward mechanism to improve the accuracy of the energy consumption prediction of building equipment. The hyperparameters of the neural network are optimized using a genetic algorithm to improve the intelligence level of the energy-saving control strategy. Based on deep reinforcement learning, the intelligent agent can adaptively adjust the energy-saving strategy under different environmental conditions to ensure optimal building energy consumption while ensuring user comfort.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent building energy-saving control, and specifically to an intelligent energy-saving control method and system for green building equipment based on artificial intelligence. Background Art

[0002] With the intensifying global energy crisis and the advancement of sustainable development goals, green building technology has become a key development direction in the construction industry. Traditional building equipment accounts for over 40% of global energy consumption, with HVAC (heating, ventilation, and air conditioning), lighting, and elevator systems being the primary contributors. In recent years, the rapid development of technologies such as the Internet of Things (IoT), big data analytics, and artificial intelligence (AI) has made intelligent building energy-saving control possible. Intelligent control technologies primarily include rule-based control, optimization scheduling algorithms, and neural network prediction models, which have significantly improved the energy efficiency of building equipment. Furthermore, the application of digital twin technology has enabled more accurate simulations of building equipment energy consumption, providing data support for further optimizing energy-saving control strategies. Furthermore, the increasing application of intelligent decision-making methods such as reinforcement learning in complex system control has provided new research directions for intelligent energy-saving control of building equipment. However, while the application of these technologies has promoted the development of green buildings, existing energy-saving control methods still have many limitations and are unable to meet the requirements for dynamic optimization of building equipment under diverse environmental conditions.

[0003] Current energy-saving control technologies for building equipment primarily rely on fixed rule control and data-driven predictive control. Fixed rule-based control methods (such as fuzzy logic and PID control) struggle to adapt to dynamic changes in the building's internal environment, limiting energy efficiency. Furthermore, data-driven optimization methods (such as neural networks and genetic algorithms) face practical challenges, such as difficulty adaptively adjusting hyperparameters, lengthy model training times, and a lack of long-term decision-making capabilities in optimization strategies. Furthermore, traditional optimization scheduling algorithms mostly analyze static data and are unable to perceive changes in the building's internal and external environments in real time, resulting in poor adaptability of control strategies. While existing reinforcement learning applications can optimize control decisions to a certain extent, they often suffer from slow training convergence, insufficient policy generalization, and difficulty balancing building energy consumption with user comfort. Furthermore, current intelligent energy-saving control solutions often lack effective multi-objective optimization mechanisms, making it difficult to minimize energy consumption while ensuring a positive user experience. This paper proposes an AI-based intelligent energy-saving control method for green building equipment. By leveraging a digital twin model, genetic algorithms, adaptive neural network optimization, and deep reinforcement learning, this method overcomes these shortcomings, achieving maximum energy efficiency, dynamic policy optimization, and guaranteed user comfort. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are: the existing intelligent energy-saving control methods for building equipment have insufficient adaptability of energy-saving strategies, the optimization models are difficult to adjust dynamically, the control strategies cannot take into account both energy consumption minimization and user comfort, and how to use artificial intelligence to optimize the energy consumption of building equipment and achieve adaptive control.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent energy-saving control method for green building equipment based on artificial intelligence, comprising: collecting building environment data and performing preprocessing.

[0007] The digital twin model is used to simulate the energy consumption of building equipment, train the baseline neural network model, and obtain the initial control strategy.

[0008] Genetic algorithm is used to optimize the hyperparameters of the baseline model and generate multiple candidate energy-saving control strategies.

[0009] The optimized strategy is input into the deep reinforcement learning model for strategy iteration. The energy-saving efficiency and comfort scores are calculated after each iteration, and the optimal strategy is selected through the reward mechanism.

[0010] As a preferred solution of the green building equipment intelligent energy-saving control method based on artificial intelligence described in the present invention, the collecting of building environment data and preprocessing thereof include:

[0011] Deploy sensors in buildings to collect building environment data and equipment operating status data.

[0012] Environmental data includes indoor and outdoor temperature, humidity, light intensity, and carbon dioxide concentration.

[0013] Equipment operation data includes HVAC system energy consumption, elevator start and stop frequency, and intelligent lighting brightness settings.

[0014] Preprocessing includes data cleaning, feature extraction and data normalization.

[0015] As a preferred solution of the green building equipment intelligent energy-saving control method based on artificial intelligence described in the present invention, the simulation of building equipment energy consumption using a digital twin model includes:

[0016] The digital twin model builds a dynamic virtual map of building equipment, integrates environmental data with equipment operating status data, and conducts real-time simulation analysis of energy consumption, including:

[0017] Modeling the impact of environmental factors is expressed as:

[0018] ;

[0019] in, Indicates the comprehensive impact of environmental factors on energy consumption, represents the weight of each environmental factor, Represents the influence function of each environmental variable, Represents a time variable.

[0020] Model the change of equipment energy consumption as follows:

[0021] ;

[0022] in, express The total energy consumption at the time, express The total energy consumption at the time, Indicates the energy consumption impact of the operating status of building equipment, Represents a time variable.

[0023] As a preferred solution of the green building equipment intelligent energy-saving control method based on artificial intelligence described in the present invention, the training baseline neural network model to obtain the initial control strategy includes:

[0024] Based on the preprocessed data and the data obtained by digital twin model simulation, the baseline neural network is trained to learn the characteristics of building equipment energy consumption and predict the energy consumption value, which is expressed as:

[0025] ;

[0026] in, represents the predicted total energy consumption of building equipment, represents the baseline neural network, Indicates the environmental data input, Indicates the device status input, represents the baseline neural network weights.

[0027] According to the loss function, the prediction error of the neural network model is minimized, and the optimal control strategy for the future is predicted through the trained baseline neural network. The strategy goal is to minimize energy consumption while ensuring user comfort, which is expressed as:

[0028] ;

[0029] in, represents the optimal control strategy, including HVAC regulation, lighting optimization, and elevator scheduling. represents the set of control strategies for building equipment, represents the indoor temperature inside the building, represents the comfort constraint weight, represents the optimal temperature set point, Indicates the optimal indoor temperature set based on user comfort and energy saving requirements. represents the comfort constraint weight, Indicates the optimal illumination setting value, It represents the optimal light intensity set based on human visual comfort and lighting energy saving requirements. represents the decision time step.

[0030] As a preferred solution of the green building equipment intelligent energy-saving control method based on artificial intelligence described in the present invention, the optimization of the hyperparameters of the baseline model using a genetic algorithm includes:

[0031] Define optimization goals, including minimizing energy consumption, maximizing comfort, and optimizing control stability:

[0032] ;

[0033] in, represents the fitness function, represents the set of hyperparameters of the genetic algorithm, Represents hyperparameters The corresponding fitness value, the larger the value, the better the hyperparameter combination. Indicates the The predicted energy consumption of each time step, represents the indoor temperature setpoint under the current control decision, Indicates the lighting setting value under the current control decision, represents the number of time steps within the time window considered during the optimization process, represents the volatility of the control strategy and is defined as:

[0034] ;

[0035] in, Representation device In time The control strategy value of is used to measure the stability of the control strategy. Representation device In time The control strategy value of Indicates the total number of smart devices controlled inside the building, Indicates the Smart devices, Represents the balance factor.

[0036] As a preferred solution of the artificial intelligence-based intelligent energy-saving control method for green building equipment described in the present invention, the optimization of the hyperparameters of the baseline model using a genetic algorithm further includes:

[0037] The initial population consists of multiple randomly generated hyperparameter combinations Composition, each individual Including the number of neural network layers , learning rate , regularization parameter , batch size , activation function parameters .

[0038] The population size is set to , the initialization method is expressed as:

[0039] ;

[0040] in, Indicates the Initial individuals, represents a uniform random distribution, represents the minimum value of hyperparameter search, Represents the maximum value of the hyperparameter search.

[0041] According to the fitness function Perform roulette wheel selection to select individuals with high fitness to enter the next generation:

[0042] ;

[0043] in, Represents an individual The probability of being selected, represents the number of individuals in the current genetic algorithm population, Indicates the number of Individuals.

[0044] As a preferred solution of the artificial intelligence-based intelligent energy-saving control method for green building equipment described in the present invention, the generating of multiple candidate energy-saving control strategies includes:

[0045] Energy consumption weighted adaptive crossover is introduced. The crossover weight is allocated according to the energy consumption trend, and low energy consumption individuals give priority to contributing hyperparameters:

[0046] ;

[0047] in, Represents a new individual, Indicates an attached individual.

[0048] Perform dynamic compilation and use exponentially decaying mutation rate to prevent premature convergence, which can be expressed as:

[0049] ;

[0050] in, represents the dynamic mutation rate, represents the initial mutation rate, Indicates the current genetic generation.

[0051] The optimized hyperparameters are used to train multiple candidate neural networks, and the objective function is expressed as:

[0052] ;

[0053] in, It represents the energy-saving control strategy set finally generated by the baseline neural network after genetic algorithm optimization. represents the building energy consumption predicted by the baseline neural network after optimizing the hyperparameters, represents the HVAC temperature setpoint output by the neural network according to the optimization strategy, represents the lighting setting value output by the neural network according to the optimization strategy, Represents the weights for controlling temperature and lighting optimization.

[0054] As a preferred solution of the artificial intelligence-based intelligent energy-saving control method for green building equipment described in the present invention, the step of inputting the optimized strategy into a deep reinforcement learning model and performing strategy iteration includes:

[0055] Build a deep reinforcement learning model, define the state, action, and reward mechanism, and enable the system to learn to optimize the control strategy.

[0056] The status includes indoor temperature and humidity, light intensity, carbon dioxide concentration, HVAC power, lighting power and elevator power, and the actions include HVAC temperature setting, lighting brightness adjustment and elevator scheduling.

[0057] The reward function is calculated based on energy consumption, comfort deviation and control strategy stability to ensure a balance between energy saving goals and user experience.

[0058] As a preferred solution of the artificial intelligence-based intelligent energy-saving control method for green building equipment described in the present invention, wherein: the energy-saving efficiency and comfort score are calculated after each iteration, and the optimal strategy is selected through the reward mechanism, including:

[0059] The long-term benefits of different control strategies are evaluated using a reinforcement learning Q-value function. The Bellman equation is then used to iteratively update the Q-value, enabling continuous strategy optimization. To avoid local optima, a greedy strategy is used to balance exploration and exploitation. During training, the convergence of the Q-value is continuously monitored. Training terminates when the strategy converges, meaning the Q-value change falls below a set threshold.

[0060] The optimal control strategy is selected from the training results and applied to the energy-saving control of building equipment to ensure the lowest energy consumption of the building while maintaining user comfort.

[0061] The intelligent energy-saving control system for green building equipment based on artificial intelligence is characterized by including a preprocessing module for collecting building environment data and performing preprocessing.

[0062] The prediction module uses the digital twin model to simulate the energy consumption of building equipment, train the baseline neural network model, and obtain the initial control strategy.

[0063] The optimization module uses genetic algorithm to optimize the hyperparameters of the baseline model and generate multiple candidate energy-saving control strategies.

[0064] The screening module inputs the optimized strategy into the deep reinforcement learning model, performs strategy iteration, calculates the energy-saving efficiency and comfort score after each iteration, and selects the optimal strategy through the reward mechanism.

[0065] The beneficial effects of this invention include using a digital twin model to simulate building equipment energy consumption. Compared to traditional energy consumption calculation methods based on static data, the digital twin model can dynamically update equipment operating status, simulate the impact of complex building environments, and improve the ability to predict energy consumption trends. If building equipment configurations change or environmental conditions fluctuate, the digital twin model can adjust in real time to ensure the accuracy of energy-saving strategies, overcoming the lack of adaptability of traditional optimization methods in dynamic environments.

[0066] A genetic algorithm is used to optimize the hyperparameters of the baseline neural network. Compared to traditional neural networks with fixed hyperparameters, the genetic algorithm-optimized model can dynamically adjust hyperparameters, improving the neural network's predictive capabilities. Automatic hyperparameter optimization by the genetic algorithm avoids the inefficient manual parameter adjustment required by traditional neural networks, improving computational efficiency. Multiple candidate energy-saving control strategies are generated, providing a diverse decision-making basis for subsequent reinforcement learning optimization.

[0067] The optimized policy is fed into a deep reinforcement learning model for policy iteration. Compared to traditional static rule-based energy-saving control methods, reinforcement learning can continuously optimize the control policy under varying environmental conditions, ensuring that the policy consistently achieves optimal energy-saving results. Traditional energy-saving methods often focus solely on minimizing energy consumption, while reinforcement learning, through a reward mechanism, can simultaneously maintain user comfort while conserving energy, ensuring that intelligent control does not compromise the user experience. Because reinforcement learning can continuously adjust its policy based on environmental feedback, it can adapt to different building types and environmental conditions, improving the versatility of the control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0069] Figure 1 This is an overall flow chart of the artificial intelligence-based green building equipment intelligent energy-saving control method and system provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0070] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0071] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides an intelligent energy-saving control method for green building equipment based on artificial intelligence, comprising:

[0072] S1: Collect building environment data and perform preprocessing.

[0073] Deploy sensors in buildings to collect building environment data and equipment operating status data.

[0074] Environmental data includes indoor and outdoor temperature, humidity, light intensity, and carbon dioxide concentration.

[0075] Equipment operation data includes HVAC system energy consumption, elevator start and stop frequency, and intelligent lighting brightness settings.

[0076] Preprocessing includes data cleaning, feature extraction and data normalization.

[0077] It should be noted that the specific preprocessing steps are as follows:

[0078] All collected data is preliminarily screened through the edge computing gateway to remove obviously erroneous or invalid data (such as abnormal jumps and data that has not changed for a long time).

[0079] The data is then time-aligned and noise-reduced. Because different sensors have varying sampling frequencies, time synchronization is required to ensure data comparability within the same time window. For signals with high-frequency fluctuations (such as temperature and humidity sensor data), wavelet transforms are used for smoothing to reduce the impact of noise. Missing data is supplemented using interpolation to ensure data integrity.

[0080] After data cleaning, feature engineering was performed on the data, including feature selection, feature extraction, and dimensionality reduction. The Pearson correlation coefficient was used to analyze the relationship between different environmental variables, removing redundant and highly correlated features. Feature dimensionality reduction was performed based on principal component analysis (PCA), extracting the most representative feature variables and reducing computational complexity.

[0081] The Z-score normalization method is used to unify the numerical ranges of different data to facilitate the subsequent training of the neural network model.

[0082] S2: Use the digital twin model to simulate the energy consumption of building equipment, train the baseline neural network model, and obtain the initial control strategy.

[0083] It should be noted that by constructing a digital twin model, the dynamic changes of building equipment energy consumption under different environmental conditions are simulated, and training is carried out in combination with real data to improve the generalization ability of energy-saving control strategies.

[0084] The digital twin model builds a dynamic virtual map of building equipment, integrates environmental data with equipment operating status data, and conducts real-time simulation analysis of energy consumption, including:

[0085] Modeling the impact of environmental factors is expressed as:

[0086] ;

[0087] in, Indicates the comprehensive impact of environmental factors on energy consumption, represents the weight of each environmental factor, Represents the influence function of each environmental variable, Represents a time variable.

[0088] It should be noted that When 1 is taken, for The indoor temperature at the moment, Take 2, for The outdoor temperature at the time, Take 3 o'clock, for The light intensity at the moment, Take 4, for The carbon dioxide concentration at the moment. Calculated based on regression of historical data.

[0089] An exponential decay function is selected, which is expressed as:

[0090] ;

[0091] in, Represents the index weight of each environmental factor, calculated based on historical data. Indicates the environment variables.

[0092] Model the change of equipment energy consumption as follows:

[0093] ;

[0094] in, express The total energy consumption at the time, express The total energy consumption at the time, Indicates the energy consumption impact of the operating status of building equipment, Represents a time variable.

[0095] Furthermore, by integrating and calculating the changes in energy consumption over time, the building energy consumption can be dynamically simulated, which facilitates the subsequent neural network training.

[0096] It should be noted that Indicates the energy consumption impact of the operating status of building equipment, defined as:

[0097] ;

[0098] in, Indicates the consumption of the equipment, When 1 is taken, for The power consumption of the HVAC system at the moment, Take 2, for The instantaneous energy consumption of the elevator starting and stopping at each moment, Take 3 o'clock, for The power consumption of the smart lighting system at all times. Indicates the energy consumption coefficient of each device, which is determined by technicians based on the rated power and operating time of the equipment.

[0099] Based on the preprocessed data and the data obtained by digital twin model simulation, the baseline neural network is trained to learn the characteristics of building equipment energy consumption and predict the energy consumption value, which is expressed as:

[0100] ;

[0101] in, represents the predicted total energy consumption of building equipment, represents the baseline neural network, Indicates the environmental data input, Indicates the device status input, represents the baseline neural network weights.

[0102] According to the loss function, the prediction error of the neural network model is minimized, and the optimal control strategy for the future is predicted through the trained baseline neural network. The strategy goal is to minimize energy consumption while ensuring user comfort, which is expressed as:

[0103] ;

[0104] in, represents the optimal control strategy, including HVAC regulation, lighting optimization, and elevator scheduling. represents the set of control strategies for building equipment, represents the indoor temperature inside the building, represents the comfort constraint weight, represents the optimal temperature set point, Indicates the optimal indoor temperature set based on user comfort and energy saving requirements. represents the comfort constraint weight, Indicates the optimal illumination setting value, It represents the optimal light intensity set based on human visual comfort and lighting energy saving requirements. represents the decision time step. and It is a value set by technicians based on historical data and experience. In specific applications, it can be dynamically adjusted according to seasonal changes, weather changes, etc.

[0105] It should be noted that traditional energy consumption prediction methods have limited accuracy. Optimization methods based on static energy consumption models cannot accurately simulate the dynamic changes in the building environment, resulting in strategy failure. Existing technologies often optimize for specific buildings, but when the physical characteristics of a building change, the optimization strategy is difficult to generalize.

[0106] Furthermore, our invention combines environmental data with equipment status data to create a digital twin model that can simulate building energy consumption in real time, making optimization strategies more precise. Compared to traditional optimization methods, this technology can adapt to different building environments, improve model portability, and avoid overfitting strategies to specific buildings.

[0107] In step S2, a digital twin model is used to simulate building equipment energy consumption. Compared to traditional energy consumption calculations based on static data, digital twin models can dynamically update equipment operating status, simulate the impact of complex building environments, and improve the ability to predict energy consumption trends. If building equipment configurations change or environmental conditions fluctuate, the digital twin model can adjust in real time to ensure the accuracy of energy-saving strategies, overcoming the lack of adaptability of traditional optimization methods in dynamic environments.

[0108] S3: Genetic algorithm is used to optimize the hyperparameters of the baseline model and generate multiple candidate energy-saving control strategies.

[0109] It should be noted that the genetic algorithm is used to optimize the hyperparameters of the neural network, including the learning rate, number of layers, and regularization parameters, to improve the prediction accuracy and stability of the neural network in energy-saving control.

[0110] Define optimization goals, including minimizing energy consumption, maximizing comfort, and optimizing control stability:

[0111] ;

[0112] in, represents the fitness function, represents the set of hyperparameters of the genetic algorithm, Represents hyperparameters The corresponding fitness value, the larger the value, the better the hyperparameter combination. Indicates the The predicted energy consumption of each time step, represents the indoor temperature setpoint under the current control decision, Indicates the lighting setting value under the current control decision, represents the number of time steps within the time window considered during the optimization process, represents the volatility of the control strategy and is defined as:

[0113] ;

[0114] in, Representation device In time The control strategy value of is used to measure the stability of the control strategy. Representation device In time The control strategy value of Indicates the total number of smart devices controlled inside the building, Indicates the Smart devices, Represents the balance factor.

[0115] It should be noted that It is obtained by technical personnel based on experimental calculations and can be adjusted in specific implementations.

[0116] Furthermore, the volatility of the control strategy is introduced , can prevent frequent adjustments to device parameters from causing unnecessary energy consumption or user discomfort. Taking the inverse, we encourage low-energy consumption solutions and introduce comfort and stability constraints to make the optimization results more in line with actual needs.

[0117] The initial population consists of multiple randomly generated hyperparameter combinations Composition, each individual Including the number of neural network layers , learning rate , regularization parameter , batch size , activation function parameters .

[0118] The population size is set to , use uniform distribution to initialize hyperparameters to ensure the diversity of the initial population and avoid premature convergence. The initialization method is expressed as:

[0119] ;

[0120] in, Indicates the Initial individuals, represents a uniform random distribution, represents the minimum value of hyperparameter search, Represents the maximum value of the hyperparameter search.

[0121] According to the fitness function Perform roulette wheel selection to select individuals with high fitness to enter the next generation:

[0122] ;

[0123] in, Represents an individual The probability of being selected, represents the number of individuals in the current genetic algorithm population, Indicates the number of Individuals.

[0124] Energy consumption weighted adaptive crossover is introduced. The crossover weight is allocated according to the energy consumption trend, and low energy consumption individuals give priority to contributing hyperparameters:

[0125] ;

[0126] in, Represents a new individual, Indicates an attached individual.

[0127] Perform dynamic compilation and use exponentially decaying mutation rate to prevent premature convergence, which can be expressed as:

[0128] ;

[0129] in, represents the dynamic mutation rate, represents the initial mutation rate, Indicates the current genetic generation.

[0130] The optimized hyperparameters are used to train multiple candidate neural networks, and the objective function is expressed as:

[0131] ;

[0132] in, It represents the energy-saving control strategy set finally generated by the baseline neural network after genetic algorithm optimization. represents the building energy consumption predicted by the baseline neural network after optimizing the hyperparameters, represents the HVAC temperature setpoint output by the neural network according to the optimization strategy, represents the lighting setting value output by the neural network according to the optimization strategy, Represents the weights for controlling temperature and lighting optimization.

[0133] It should be noted that neural network hyperparameters are difficult to adjust manually. Traditional methods rely on manual tuning, which is inefficient and unstable. Neural network models with fixed hyperparameters are difficult to adapt to different building environments, resulting in reduced prediction accuracy.

[0134] Furthermore, genetic algorithms can search for optimal hyperparameter combinations in a relatively short time, improving training efficiency. Compared to fixed hyperparameter models, genetically optimized neural networks can adapt to different building environments, improving the effectiveness of energy-saving strategies.

[0135] In step S3, a genetic algorithm is used to optimize the hyperparameters of the baseline neural network. Compared to traditional neural networks with fixed hyperparameters, the genetic algorithm-optimized model can dynamically adjust hyperparameters, improving the neural network's predictive capabilities. Automatic hyperparameter optimization with a genetic algorithm avoids the inefficient manual adjustment required in traditional neural networks, improving computational efficiency. Multiple candidate energy-saving control strategies are generated, providing a diverse decision-making basis for subsequent reinforcement learning optimization.

[0136] S4: Input the optimized strategy into the deep reinforcement learning model and perform strategy iteration. After each iteration, calculate the energy-saving efficiency and comfort score, and select the optimal strategy through the reward mechanism.

[0137] It should be noted that deep reinforcement learning is used to optimize the control strategy of building equipment, and through Q-value function updates and reward mechanisms, energy consumption is minimized while ensuring user comfort.

[0138] Build a deep reinforcement learning model, define the state, action, and reward mechanism, and enable the system to learn to optimize the control strategy.

[0139] Define the state: at the moment , the state of the agent It is composed of the environment status and the equipment status:

[0140] ;

[0141] in, Indicates the current indoor temperature. Indicates the current humidity. Indicates the current light intensity. represents the carbon dioxide concentration, Indicates the current power of the HVAC system. Indicates the current power of the lighting system. Indicates the current power of the elevator system.

[0142] Define actions: The control strategies that a reinforcement learning agent can choose are expressed as:

[0143] ;

[0144] in, express HVAC temperature settings at time nodes, express Lighting settings for time nodes, express Elevator dispatching status at a time node.

[0145] The reward function is calculated based on energy consumption, comfort deviation, and control strategy stability to ensure a balance between energy saving goals and user experience, and is expressed as:

[0146] ;

[0147] in, Indicates immediate reward, It represents weight, which is obtained by technicians based on experiments and experience and can be modified in actual applications. represents the control strategy volatility, which is expressed as:

[0148] ;

[0149] in, Representation device In time The control strategies adopted, Representation device In time Control strategies adopted.

[0150] It should be noted that the first term of the reward function encourages low-energy solutions. The second and third terms ensure that temperature and light intensity are close to their optimal values. The fourth term penalizes frequent adjustments to control parameters, improving policy stability.

[0151] The Q-value function of reinforcement learning is used to evaluate the long-term benefits of different control strategies:

[0152] ;

[0153] in, Indicates that the status Take action long-term benefits. Represents the discount factor, which controls the long-term reward weight and has a value range of 0 to 1.

[0154] The Q value is iteratively updated through the Bellman equation to continuously optimize the strategy.

[0155] To avoid local optimality, a greedy strategy is used to balance exploration and utilization. During the training process, the convergence of the Q value is continuously monitored. When the strategy converges, that is, the Q value change is lower than the set threshold, the training is terminated.

[0156] The optimal control strategy is selected from the training results and applied to the energy-saving control of building equipment to ensure the lowest energy consumption of the building while maintaining user comfort. The optimal control strategy is expressed as:

[0157] ;

[0158] in, Represents the optimal control strategy, including the optimal control strategy of the HVAC system,

[0159] It should be noted that traditional optimization methods often aim to minimize energy consumption while ignoring user comfort. Control methods based on short-term optimization are prone to falling into local optimality, while reinforcement learning can evaluate long-term benefits.

[0160] Furthermore, compared to fixed rule-based control, reinforcement learning can continuously optimize strategies based on historical data and real-time feedback, improving adaptability. Through a reward mechanism, it can ensure energy savings while maintaining user comfort.

[0161] In step S4, the optimized policy is input into a deep reinforcement learning model for policy iteration. Compared to traditional energy-saving control methods based on static rules, reinforcement learning can continuously optimize the control policy under different environmental conditions, ensuring that the policy always achieves the optimal energy-saving effect. Traditional energy-saving methods often focus solely on minimizing energy consumption, while reinforcement learning, through a reward mechanism, can maintain user comfort while saving energy, ensuring that intelligent control does not affect the user experience. Because reinforcement learning can continuously adjust its policy based on environmental feedback, this method can adapt to different building types and environmental conditions, improving the versatility of the control system.

[0162] The above embodiments also include an artificial intelligence-based green building equipment intelligent energy-saving control system, specifically:

[0163] The preprocessing module collects building environment data and performs preprocessing.

[0164] The prediction module uses the digital twin model to simulate the energy consumption of building equipment, train the baseline neural network model, and obtain the initial control strategy.

[0165] The optimization module uses genetic algorithm to optimize the hyperparameters of the baseline model and generate multiple candidate energy-saving control strategies.

[0166] The screening module inputs the optimized strategy into the deep reinforcement learning model, performs strategy iteration, calculates the energy-saving efficiency and comfort score after each iteration, and selects the optimal strategy through the reward mechanism.

[0167] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of the power monitoring system. The I / O interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent energy-saving control method for green building equipment based on artificial intelligence.

[0168] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processors (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0169] Example 2 is an embodiment of the present invention, which provides an intelligent energy-saving control method and system for green building equipment based on artificial intelligence. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0170] This example aims to demonstrate the effectiveness of an AI-based intelligent energy-saving control method and system for green building equipment in optimizing building energy consumption and improving energy efficiency. The experiment selected a smart office building as the research object. The building includes an HVAC (heating, ventilation, and air conditioning) system, an intelligent lighting system, and an elevator dispatch system. During the experiment, sensors were used to collect building environmental data, including indoor and outdoor temperature and humidity, light intensity, and carbon dioxide concentration. The data was then preprocessed to eliminate outliers and perform normalization.

[0171] Subsequently, a digital twin model was used to simulate building equipment energy consumption and train a baseline neural network model to generate a preliminary energy-saving control strategy. The neural network inputs were environmental data and equipment status data, and its output was energy consumption predictions. To optimize the model's hyperparameters, a genetic algorithm was used to search for hyperparameters, including the number of neural network layers, learning rate, and regularization parameters, to ensure the generation of the optimal energy-saving control strategy. Ultimately, the optimized strategy was input into a deep reinforcement learning model, which continuously iterated the energy-saving strategy through reinforcement learning. After each iteration, energy-saving efficiency and comfort scores were calculated, and a reward mechanism was used to select the optimal strategy.

[0172] The experiment set up a comparative test, using the most common rule-based energy-saving approach currently used in the industry as a reference group. Rule-based control methods set equipment operating strategies based on fixed environmental parameters, such as maintaining a constant HVAC temperature of 22°C and maintaining lighting brightness at 80% during the day. The method presented in this paper, on the other hand, optimizes control strategies based on dynamic data, enabling real-time adjustments to energy-related parameters such as HVAC temperature, lighting brightness, and elevator scheduling to adapt to varying environmental requirements.

[0173] The experiment lasted 30 days, with building energy consumption data recorded 24 hours a day and occupant comfort feedback collected. The recorded environmental parameters, equipment operating status, and optimized control strategies are shown in Table 1.

[0174] Table 1 Experimental data of our invention

[0175]

[0176] In order to verify the advantages of the method of the present invention, the energy-saving effect of the rule-based control method is compared, as shown in Table 2. The comfort calculation method is as follows:

[0177] Temperature score: An optimal temperature range is set based on the human body's comfort zone (usually 20°C to 26°C). If the room temperature is within this range, the highest score is given. As the temperature deviates from this range, the score decreases, with greater deviations resulting in lower scores.

[0178] Humidity score: The ideal relative humidity for humans is typically between 40% and 60%. Humidity that is too high or too low affects comfort, so the score is adjusted based on the degree of humidity deviation.

[0179] Lighting Score: Consider your daily office or living environments and set an optimal lighting range (e.g., 300-500 lux is recommended for office areas). If the actual lighting intensity falls within this range, the score will be the highest. Lighting that is too bright or too dim will lower the score.

[0180] Carbon Dioxide Concentration Score: Excessively high CO2 concentrations can degrade air quality, impacting concentration and comfort. Generally, a CO2 concentration range of 400-800 ppm is ideal, while levels exceeding 1000 ppm can significantly impact comfort. Therefore, the score is dynamically adjusted based on concentration changes.

[0181] Equipment Noise Score: Operating noise from building equipment, such as HVAC systems and elevators, also impacts comfort. If noise levels are maintained within reasonable limits (e.g., typically no more than 50 dB in office areas), the score is high. If equipment frequently starts and stops, generating high noise levels, the score is lowered accordingly.

[0182] The scores of each of the above factors are weighted according to the custom weights to obtain the final overall comfort score. The weighting can be adjusted according to the actual needs of the experimental environment. In this experiment, the weighting was designed to be 35% for indoor temperature, 20% for humidity, 20% for CO2 concentration, 15% for light intensity, and 10% for equipment noise.

[0183] Finally, all scores are weighted and summed to obtain a comprehensive comfort score, which is normalized to the range of 0 to 10 to intuitively compare the comfort performance under different control strategies.

[0184] Table 2 Comparative experimental data of existing technologies

[0185]

[0186] Analysis of experimental data demonstrates that the proposed method outperforms existing technologies in both energy efficiency and comfort. First, in terms of HVAC energy consumption, the implementation of AI-based intelligent optimization control reduced energy consumption from 15.2 kWh to 11.8 kWh, achieving a 22.4% energy saving. This is primarily due to the optimized control strategy's ability to adaptively adjust temperature set points based on environmental changes, avoiding the energy waste associated with fixed rule-based control.

[0187] In terms of lighting energy consumption, the optimization strategy dynamically adjusts lighting brightness based on light intensity, reducing energy consumption from 4.5kWh to 3.2kWh, a 28.9% energy saving. Furthermore, the elevator scheduling optimization strategy reduced the number of unnecessary elevator starts and stops from 22 to 15, reducing energy consumption and extending equipment life.

[0188] The improvement in comfort scores (from 7.5 to 8.9) demonstrates that this optimization strategy ensures energy savings while avoiding a degradation in user experience. Traditional energy-saving methods often focus on minimizing energy consumption while neglecting indoor comfort. This method utilizes a reinforcement learning model to balance energy consumption and comfort, resulting in a control strategy that better meets user needs.

[0189] Overall, the proposed method not only offers significant advantages in optimizing energy consumption but also significantly improves user experience. This method is adaptable to diverse building environments and offers greater versatility. Compared to traditional rule-based control methods, this method dynamically optimizes energy-saving strategies, ensuring minimal building energy consumption while maintaining user comfort, making it widely applicable in practical applications.

[0190] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent energy-saving control method for green building equipment based on artificial intelligence, characterized in that: include: Collect building environment data and perform pre-processing; Use the digital twin model to simulate the energy consumption of building equipment, train the baseline neural network model, and obtain the initial control strategy; Genetic algorithms are used to optimize the hyperparameters of the baseline model and generate multiple candidate energy-saving control strategies; The optimized strategy is input into the deep reinforcement learning model for strategy iteration. After each iteration, the energy-saving efficiency and comfort scores are calculated, and the optimal strategy is selected through the reward mechanism. The simulation of building equipment energy consumption using digital twin models includes: The digital twin model builds a dynamic virtual map of building equipment, integrates environmental data with equipment operating status data, and conducts real-time simulation analysis of energy consumption, including: Modeling the impact of environmental factors is expressed as: ； in, Indicates the comprehensive impact of environmental factors on energy consumption, represents the weight of each environmental factor, Represents the influence function of each environmental variable, represents the time variable; Model the change of equipment energy consumption as follows: ； in, express The total energy consumption at the time, express The total energy consumption at the time, Indicates the energy consumption impact of the operating status of building equipment, represents the time variable; The training baseline neural network model to obtain the initial control strategy includes: Based on the preprocessed data and the data obtained by digital twin model simulation, the baseline neural network is trained to learn the characteristics of building equipment energy consumption and predict the energy consumption value, which is expressed as: ； in, represents the predicted total energy consumption of building equipment, represents the baseline neural network, Indicates the environmental data input, Indicates the device status input, represents the baseline neural network weights; According to the loss function, the prediction error of the neural network model is minimized, and the optimal control strategy for the future is predicted through the trained baseline neural network. The strategy goal is to minimize energy consumption while ensuring user comfort, which is expressed as: ； in, represents the optimal control strategy, including HVAC regulation, lighting optimization, and elevator scheduling; represents the set of control strategies for building equipment, represents the indoor temperature inside the building, represents the comfort constraint weight, represents the optimal temperature set point, Indicates the optimal indoor temperature set based on user comfort and energy saving requirements. represents the comfort constraint weight, Indicates the optimal illumination setting value, It represents the optimal light intensity set based on human visual comfort and lighting energy saving requirements. represents the decision time step; The genetic algorithm is used to optimize the hyperparameters of the baseline model, including: Define optimization goals, including minimizing energy consumption, maximizing comfort, and optimizing control stability: ； in, represents the fitness function, represents the set of hyperparameters of the genetic algorithm, Represents hyperparameters The corresponding fitness value, the larger the value, the better the hyperparameter combination; Indicates the The predicted energy consumption of each time step, represents the indoor temperature setpoint under the current control decision, Indicates the lighting setting value under the current control decision, represents the number of time steps within the time window considered during the optimization process, represents the volatility of the control strategy and is defined as: ； in, Representation device In time The control strategy value of is used to measure the stability of the control strategy; Representation device In time The control strategy value of Indicates the total number of smart devices controlled inside the building, Indicates the Smart devices, represents the balance factor; The initial population consists of multiple randomly generated hyperparameter combinations Composition, each individual Including the number of neural network layers , learning rate , regularization parameter , batch size , activation function parameters ; The population size is set to , the initialization method is expressed as: ； in, Indicates the Initial individuals, represents a uniform random distribution, represents the minimum value of hyperparameter search, Indicates the maximum value of hyperparameter search; According to the fitness function Perform roulette wheel selection to select individuals with high fitness to enter the next generation: ； in, Represents an individual The probability of being selected, represents the number of individuals in the current genetic algorithm population, Indicates the number of Individuals.

2. The artificial intelligence-based intelligent energy-saving control method for green building equipment according to claim 1, characterized in that: The collecting of building environment data and preprocessing thereof include: Deploy sensors in buildings to collect building environment data and equipment operating status data; Environmental data includes indoor and outdoor temperature, humidity, light intensity, and carbon dioxide concentration; Equipment operation data includes HVAC system energy consumption, elevator start and stop frequency, and intelligent lighting brightness settings; Preprocessing includes data cleaning, feature extraction and data normalization.

3. The artificial intelligence-based intelligent energy-saving control method for green building equipment according to claim 2, characterized in that: Generating multiple candidate energy-saving control strategies includes: Energy consumption weighted adaptive crossover is introduced. The crossover weight is allocated according to the energy consumption trend, and low energy consumption individuals give priority to contributing hyperparameters: ； in, Represents a new individual, Indicates incidental individual; Perform dynamic compilation and use exponentially decaying mutation rate to prevent premature convergence, which can be expressed as: ； in, represents the dynamic mutation rate, represents the initial mutation rate, Indicates the current genetic generation; The optimized hyperparameters are used to train multiple candidate neural networks, and the objective function is expressed as: ； in, It represents the energy-saving control strategy set finally generated by the baseline neural network after genetic algorithm optimization; represents the building energy consumption predicted by the baseline neural network after optimizing the hyperparameters, represents the HVAC temperature setpoint output by the neural network according to the optimization strategy, represents the lighting setting value output by the neural network according to the optimization strategy, Represents the weights for controlling temperature and lighting optimization.

4. The artificial intelligence-based intelligent energy-saving control method for green building equipment according to claim 3, characterized in that: Inputting the optimized strategy into the deep reinforcement learning model and performing strategy iteration includes: Build a deep reinforcement learning model, define the state, action, and reward mechanism, and enable the system to learn and optimize control strategies; Status includes indoor temperature and humidity, light intensity, carbon dioxide concentration, HVAC power, lighting power, and elevator power. Actions include HVAC temperature setting, lighting brightness adjustment, and elevator dispatch. The reward function is calculated based on energy consumption, comfort deviation and control policy stability.

5. The artificial intelligence-based intelligent energy-saving control method for green building equipment according to claim 4, characterized in that: After each iteration, the energy-saving efficiency and comfort scores are calculated, and the optimal strategy is selected through the reward mechanism, including: The Q-value function of reinforcement learning is used to evaluate the long-term benefits of different control strategies. The Q-value is iteratively updated through the Bellman equation to continuously optimize the strategy. To avoid local optimality, a greedy strategy is used to balance exploration and exploitation. During training, the convergence of the Q-value is continuously monitored. When the strategy converges, that is, the change in the Q-value falls below the set threshold, the training terminates. The optimal control strategy is selected from the training results and applied to the energy-saving control of building equipment to ensure the lowest energy consumption of the building while maintaining user comfort.

6. An artificial intelligence-based green building equipment intelligent energy-saving control system using the method according to any one of claims 1 to 5, characterized in that: Preprocessing module, collects building environment data and performs preprocessing; The prediction module uses the digital twin model to simulate the energy consumption of building equipment, train the baseline neural network model, and obtain the initial control strategy; The optimization module uses a genetic algorithm to optimize the hyperparameters of the baseline model and generate multiple candidate energy-saving control strategies; The screening module inputs the optimized strategy into the deep reinforcement learning model, performs strategy iteration, calculates the energy-saving efficiency and comfort score after each iteration, and selects the optimal strategy through the reward mechanism.

Images