Self-adaptive energy-saving regulation and control system and method driven by energy consumption data of building equipment

Abstract

The invention relates to the technical field of building engineering, and discloses a building equipment energy consumption data driven self-adaptive energy-saving regulation and control system and method, and the method comprises global state sensing, energy consumption feature extraction, dynamic modeling prediction, multi-objective optimization solution and hierarchical execution control. Time sequence and space data are fused through Kalman filtering, unified modeling of cross-protocol equipment data is achieved, equipment operation, environment, weather and control instructions are integrated according to a space-time state input matrix, a multi-dimensional decision basis is formed, a Pareto optimal solution set is solved based on an NSGA-II algorithm, a compromise solution is selected through a fuzzy decision method, and the optimal solution set is obtained. According to the method, conflict targets of energy consumption, comfort and equipment life are balanced, a priority arbitration problem is solved by adopting dynamic weight allocation, meanwhile, a security fault-tolerant mechanism is established, a mobile security boundary model and a self-adaptive retry mechanism are introduced, an operation threshold is dynamically adjusted, and the robustness of the system is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of building engineering technology, specifically to an adaptive energy-saving control system and method driven by building equipment energy consumption data. Background Technology

[0002] Building equipment energy consumption data adaptive energy-saving control is an energy-saving technology based on real-time data acquisition, analysis, and intelligent decision-making. It mainly uses Internet of Things sensor networks to monitor the energy consumption data of various equipment in the building in real time, and combines environmental parameters with machine learning, optimization algorithms, and automatic control technology to dynamically adjust the operating status of the equipment. It maximizes energy efficiency while ensuring comfort, and is of great significance in reducing building energy consumption and improving equipment life and operating efficiency. A search revealed a method for regulating and optimizing a building electromechanical equipment system, disclosed in Chinese Patent Publication No. CN110161863A. This method involves collecting monitoring data from various equipment within a building, as well as indoor environmental data. The collected data is stored in a database according to equipment type. Based on the current environment, the DeST model algorithm is used to obtain the target operating state of the equipment. The actual operating state of the equipment is compared and analyzed with the target state, and the operating state is adjusted to approximate the target state. Based on historical energy consumption data, an energy consumption baseline for each piece of equipment is formed. The actual energy consumption data of the equipment is compared with the energy consumption baseline to obtain the optimal threshold for total energy consumption. This optimal threshold is then adjusted in conjunction with indoor environmental comfort requirements. The adjusted optimal threshold range is used to form an equipment operating curve. The equipment operating curve is continuously optimized based on feedback from equipment regulation results and local meteorological parameters. Finally, the equipment operation is regulated according to the optimized operating curve. However, the current building electromechanical equipment system control and optimization method only analyzes energy consumption characteristics through sliding window mean and historical energy consumption baseline, which makes it difficult to capture the synergistic effect of equipment groups and abnormal changes in periodic energy consumption. Furthermore, it relies on the DeST model for single-objective optimization and cannot handle multi-objective conflict problems. Therefore, based on the above-mentioned technical problems, an adaptive energy-saving control system and method driven by building equipment energy consumption data is proposed. Summary of the Invention

[0003] (I) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an adaptive energy-saving control system and method driven by building equipment energy consumption data. It has the advantages of achieving multi-objective optimization and dynamic resource allocation based on multi-modal data feature fusion processing, and solves the problem in the background technology that still needs further improvement in terms of abnormal changes in equipment cycle performance consumption and conflicts in multi-objective control.

[0004] (II) Technical Solution To achieve the above-mentioned goal of multi-objective optimization and dynamic resource allocation based on multi-modal data feature fusion processing, the present invention provides the following technical solution: an adaptive energy-saving control system driven by building equipment energy consumption data, including a data sensing layer, deploying IoT sensors and fusing heterogeneous networks to achieve dynamic optimization of sampling frequency; The data processing center extracts data features and constructs a correlation map of equipment, energy consumption, and environmental parameters; The intelligent analysis layer predicts future energy consumption curves based on energy consumption prediction models to evaluate equipment performance; The dynamic optimization layer constructs a multi-objective optimizer to generate dynamic control strategies. The execution control layer dynamically allocates control commands based on device weights and sets up a safety fault-tolerance protection mechanism.

[0005] An adaptive energy-saving control method driven by building equipment energy consumption data includes the following steps: S1. Global Status Awareness: Collects multi-source heterogeneous data from devices and performs fusion processing to correlate building equipment operation data with environmental status; S2. Energy consumption feature extraction: Extract key indicator features that reflect the operating patterns of equipment from multi-source data. S3. Dynamic Modeling and Prediction: By integrating time-series data with physical laws, a dynamic response model of equipment and environment is constructed to predict changes in equipment energy consumption data; S4. Multi-objective optimization solution: Balancing energy consumption, comfort, and equipment lifespan as conflicting objectives, the optimal solution of the control strategy is solved using the NSGA-II algorithm; S5. Hierarchical execution control: Through priority division and dynamic resource allocation, it ensures that critical instructions are executed first and maintains system stability.

[0006] Preferably, the data preprocessing steps based on multimodal data of building equipment collected by multiple sensors include: 1) Outlier data is detected using the Z-score algorithm to remove noise points. The formula is as follows:

[0007]

[0008] in For the i-th data, The median of the data. The median absolute deviation is determined by the following rule: ; 2) Calculate the interpolated data corresponding to the missing values ​​using the Kriging space interpolation method, which is represented as follows:

[0009] in For known location The measured value, Let i be the weight of the i-th data; Based on the time series ARIMA model, computational interpolation is used to repair missing data. The model is represented as follows:

[0010] in For time series forecast values, For spatial interpolation results, As a weighting factor; 3) The sensor data is calibrated using an adaptive weighted fusion algorithm to eliminate drift errors. The formula is as follows:

[0011] in For the first The sensor for the first The measured value of a physical quantity For the first The weight of each sensor, To calibrate the bias term, the weight update rule is as follows:

[0012] in For the region The true value of the physical quantity Measure the variance of the sensor cluster; Data alignment and fusion processing are performed on the preprocessed multimodal data, including: 1) Based on the PTP clock synchronization and Kriging spatial interpolation methods to unify the timestamps and spatial coordinates of multi-source data respectively, the PTP clock synchronization formula is expressed as:

[0013] in For sensor time, The time to send a synchronization message to the master clock. To receive synchronization message time from the clock, To send a delay request time from the clock, The master clock receives the delay request time; 2) The Kalman filter method is used to fuse multimodal data, as shown below: Prediction step: , Update steps: , in Here is the state transition matrix. To control the input matrix, , These are the covariances of process noise and observation noise, respectively. This is the observation matrix.

[0014] Preferably, energy consumption features are extracted from multimodal data, specifically including: 1) The sliding window statistical method is used to extract time-domain features to capture the dynamic characteristics of equipment energy consumption over time, as shown below: , , , in Let i be the energy consumption value at time i. To adjust the sliding window size, This refers to the current time point; 2) The frequency domain features are extracted using Fast Fourier Transform to analyze the periodicity of energy consumption and anomalous frequency components, as shown below: , Main frequency amplitude: , Spectral entropy: , , in Indicates length is The energy consumption sequence, For the first Complex representation of each frequency component; 3) Spatial features are extracted using graph convolutional network (GCN) embedding to uncover spatial correlation characteristics between devices, as shown below;

[0015]

[0016]

[0017] in Given an adjacency matrix with self-loops, For degree matrix, For the first Layer node feature representation, For trainable weight matrix, This is an activation function whose output is a device embedding vector. .

[0018] Preferably, a unified spatiotemporal state input matrix is ​​constructed to achieve multi-source fusion data modeling and processing, represented as:

[0019] in For equipment operation data, For environmental conditions, For external meteorological data, For control commands, For the number of devices, For data dimension metrics; Spatiotemporal feature encoding based on graph convolutional-temporal attention networks includes: Spatial encoding: , Timing coding: , Fusion Output: , in To add self-connected device adjacency matrices, This is the spatial weight matrix. It is a temporal convolutional network. This is a multi-head attention mechanism; The physical equations of the equipment are incorporated into the data-driven model, and a loss function is constructed to embed physical constraints. The loss function is expressed as follows:

[0020] in The mean square error between the predicted and actual values. For the physical equations of the equipment, , For the weighting factor; Based on temporal convolutional network The formula for constructing a prediction model with dilated convolution is expressed as follows:

[0021] in As a void factor, For convolution kernel weights, For bias terms, It is the ReLU activation function; The attention weights are calculated and expressed as follows: , in For query vector, For key vectors, For vector dimensions; Monte Carlo Dropout forecasting is used to quantify uncertain data, represented as follows:

[0022] in For the number of samples, The parameter distribution guided by Dropout.

[0023] Preferably, the optimal decision-making method for solving multi-objective problems based on the NSGA-II algorithm specifically includes: 1) Establish the core optimization objective function, including: a. Minimize energy consumption costs:

[0024] in For a moment Electricity purchased from the power grid, For time-of-use electricity pricing, This refers to the discharge power of the energy storage battery. This is the battery cycle loss cost coefficient; b. Maximize thermal comfort:

[0025] The optimal solution is obtained when PMV is close to 0. The predicted average vote value for region i is calculated as follows:

[0026] in This refers to the human body's metabolic rate. For heat load; c. Minimize equipment wear and tear:

[0027] in This refers to the number of times the equipment has been started and stopped. This represents the absolute value of the change in equipment load rate. , These are the wear coefficients for start-up / shutdown and load changes, respectively; 2) Set constraints, including: Indoor environmental constraints: , Equipment physical constraints: , Power grid interaction constraints:

[0028] in For equipment power, Limitation of power change rate; 3) The steps for solving the optimal decision using the NSGA-II algorithm include: a. Using real-number encoding, each individual represents a control strategy. The population is initialized as follows:

[0029] in Set the temperature value for region i. For the power adjustment of the air conditioning system, For adjusting the lighting brightness ratio, To schedule the target state of charge of energy storage batteries; b. Set up individuals With individuals The dominance relationship between them, in the individual Dominant Individual When, if and only if ,and The population is divided into multiple non-dominated hierarchical levels and ordered hierarchically. c. For individuals within the same non-dominated layer, calculate their distribution density in the target space, expressed as:

[0030] in , These are the adjacent individuals after sorting by target m. These are the maximum and minimum values ​​of the target m in the current population, respectively; d. Select individuals from the parent generation for competition, choosing individuals with high non-dominant hierarchy and high crowding. (This involves comparing two parent individuals.) , Generate offspring , , is represented as:

[0031]

[0032] in The crossover index, which introduces a perturbation to a specific gene locus in individual x, is expressed as:

[0033]

[0034] in are random numbers and , The variation index is used to merge parent and offspring populations and select new populations based on priority when the non-dominant level is higher or the crowding is greater in the same level.

[0035] 4) Obtain a set of non-dominated solutions as the Pareto optimal solution set. Each solution represents a trade-off strategy. Use fuzzy decision-making to select a compromise solution, expressed as:

[0036] in To achieve the ideal target value, select As the largest solution .

[0037] Preferably, the steps for priority allocation and dynamic resource allocation include: 1) Establish a policy-instruction mapping model to transform optimization policies into low-level instructions executable by the device. The model is represented as follows:

[0038] in To optimize the layer output of the ideal control quantity, For the set of instructions allowed by the device, For the safety cost function, This is the safety weighting coefficient; 2) Arbitration of priority conflicts among multiple devices based on dynamic weighted priority sorting is expressed as follows:

[0039] in The energy-saving benefits of equipment k regulation. The comfort impact index is the reciprocal of the PMV deviation. The device's health is rated, with lower scores indicating higher priority. They are dynamic weights and The arbitration rule is to execute instructions with higher priority first, or to sort instructions of the same priority according to the device's response speed. 3) Introduce an adaptive retry mechanism to ensure accurate instruction execution and feedback of status, represented as:

[0040] in This is the current number of retries. Based on the retry interval, To determine the maximum retry interval, a timeout timer is started after the first command is issued. If no response is received within the timeout period, the device is retried according to the exponential backoff strategy. Once the maximum number of retries is reached, the device is marked as faulty. 4) Establish a mobile security boundary model, updating the security threshold based on real-time status. The model is represented as follows:

[0041] in For risk function, The learning rate is the number of consecutive periods when the virus is detected. The conditions for triggering the update of the security threshold are as follows.

[0042] (III) Beneficial Effects Compared with the prior art, the present invention provides an adaptive energy-saving control system and method driven by building equipment energy consumption data, which has the following beneficial effects: 1. The adaptive energy-saving control system and method driven by building equipment energy consumption data eliminates sensor drift error through an adaptive weighted fusion algorithm, and achieves unified modeling of cross-protocol equipment data by fusing temporal and spatial data through Kalman filtering. Furthermore, it integrates equipment operation, environment, meteorology and control commands based on the spatiotemporal state input matrix to form a multi-dimensional decision-making basis.

[0043] 2. The adaptive energy-saving control system and method driven by building equipment energy consumption data is based on the NSGA-II algorithm to solve the Pareto optimal solution set, and a compromise solution is selected through fuzzy decision method to balance the conflicting objectives of energy consumption, comfort and equipment life. Dynamic weight allocation is used to solve the priority arbitration problem. At the same time, a safety fault tolerance mechanism is established and a moving safety boundary model and an adaptive retry mechanism are introduced to dynamically adjust the operation threshold, so as to significantly improve the robustness of the system. Attached Figure Description

[0044] Figure 1 This is a schematic diagram illustrating the structural principle of the adaptive energy-saving control system of the present invention; Figure 2 This is a flowchart of the adaptive energy-saving control method of the present invention. Detailed Implementation

[0045] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1 like Figure 1 The diagram shown is a schematic representation of the structural principle of the adaptive energy-saving control system of the present invention; as shown... Figure 2 The diagram shown is a flowchart of the adaptive energy-saving control method of the present invention.

[0047] In this embodiment, a human-computer interaction layer is constructed, which realizes two-way information transmission between the user and the system through multimodal interaction and intelligent visualization technology, specifically including: 1) Dynamically generate heatmaps based on Kriging interpolation, visualize and render multi-dimensional data, and calculate spatial correlations using a semi-variogram model. The model is represented as follows:

[0048] in For the nugget effect, For sill values, For variable range; 2) An attention-based operation classification model analyzes and identifies the user's optimization intent, represented as:

[0049] in For user operation sequences, , These represent the classifier weights and biases, respectively. 3) Calculate the Mean Absolute Percentage Error (MAPE), collect actual equipment operating data, and quantify the prediction error. The calculation formula is as follows:

[0050] in Let i be the actual energy consumption value at time i. This represents the predicted energy consumption value at the corresponding time point. To evaluate the number of data points within the window, when The model's prediction accuracy is excellent and requires no adjustment; when Time triggers model fine-tuning; when Time-triggered model reconstruction or sensor calibration; 4) A sliding window mechanism is used to achieve online incremental learning, and the prediction model weights are optimized based on error feedback, as shown below:

[0051] in For model parameters, This is the data for the current time window. For learning rate, The loss function has a window update rule as follows: It keeps the window size fixed and updates on a time-based scrolling basis.

[0052] Example 2 In this embodiment, policy security verification needs to be performed before actual deployment to ensure that the optimized policy meets security and reliability requirements. Specifically, this includes: 1) Constructing a digital twin scene based on EnergyPlus physical modeling, represented as:

[0053] in Let i be the heat capacity of region i. The heat transfer coefficient of the building envelope. Solar radiation intensity, The heat or cooling capacity of the equipment; 2) Construct an event-driven simulation engine to simulate the policy execution effect in the digital twin, represented as:

[0054] in A building model description file that includes equipment parameters. This is a typical meteorological year data file. Regulatory strategies to be verified 3) The multi-dimensional security assessment steps include: a. Dynamic stability is analyzed using the Lyapunov method, including:

[0055] in Let be the system state vector. , It is a positive definite symmetric matrix. For matrix The smallest eigenvalue, when This indicates that the system is stable; b. Calculate the constraint violation probability, expressed as:

[0056] in For the number of samples taken in Monte Carlo, For the j-th constraint function, This is an indicator function; if the constraint is violated... If it is 1, then it is not a violation of the constraint. =0; c. Assess equipment stress, expressed as:

[0057] in Let k be the real-time power of device k. Rated power of the equipment To accelerate the aging index; 4) Construct a fuzzy risk assessment model to quantify and classify risks. The model is represented as follows:

[0058]

[0059] in All are weights. For temperature overshoot, Temperature sensitivity coefficient; The policy adjustment based on security boundary considerations is expressed as follows:

[0060] in The distance from the strategy to the safety boundary. As a safety margin threshold, This is the gradient descent step size.

[0061] In summary, the adaptive energy-saving control system and method driven by building equipment energy consumption data eliminates sensor drift errors through an adaptive weighted fusion algorithm, and achieves unified modeling of cross-protocol equipment data by fusing temporal and spatial data through Kalman filtering. Furthermore, it integrates equipment operation, environment, meteorology, and control commands based on the spatiotemporal state input matrix to form a multi-dimensional decision-making basis. The Pareto optimal solution set is obtained based on the NSGA-II algorithm, and a compromise solution is selected through fuzzy decision-making to balance the conflicting goals of energy consumption, comfort and equipment life. Dynamic weight allocation is used to solve the priority arbitration problem. At the same time, a safety fault tolerance mechanism is established and a moving safety boundary model and an adaptive retry mechanism are introduced to dynamically adjust the operation threshold, so as to significantly improve the robustness of the system.

[0062] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in these functional modules are technologies known to those skilled in the art and are not improvements to this system. The improvement of this system lies in the interaction or connection between the modules, that is, in improving the overall structure of the system to solve the corresponding technical problems that this system aims to address.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A building equipment energy consumption data-driven adaptive energy-saving control system, characterized in that, include: The data sensing layer is used to deploy IoT sensors and integrate heterogeneous networks to achieve dynamic optimization of the sampling frequency. The data processing hub is used to extract data features and construct a correlation map of equipment-energy consumption-environment parameters; The intelligent analysis layer predicts future energy consumption curves based on energy consumption prediction models, which are used to evaluate equipment performance. The dynamic optimization layer is used to construct a multi-objective optimizer to generate dynamic control strategies; The execution control layer, based on dynamic allocation of device weights, is used to regulate commands and set up security and fault-tolerance protection mechanisms.

2. A data-driven adaptive energy-saving control method for building equipment energy consumption, characterized in that, include: Full-domain status perception, collecting multi-source heterogeneous data from devices and performing fusion processing, and linking building equipment operation data with environmental status; Energy consumption feature extraction involves extracting key indicator features that reflect the operating patterns of equipment from multi-source data. Dynamic modeling and prediction: By integrating time-series data with physical laws, a dynamic response model of equipment and environment is constructed to predict changes in equipment energy consumption data; Multi-objective optimization is used to balance energy consumption, comfort, and equipment lifespan as conflicting objectives, and the optimal solution of the control strategy is obtained based on the NSGA-II algorithm. Hierarchical execution control ensures that critical instructions are executed first and maintains system stability through priority division and dynamic resource allocation.

3. The adaptive energy-saving control method driven by building equipment energy consumption data according to claim 2, characterized in that, Based on the multimodal data of building equipment collected by multiple sensors, the data preprocessing steps include: The Z-score algorithm is used to detect outlier data and remove noise points. Its formula is as follows: in For the i-th data, The median of the data. The median absolute deviation is determined by the following rule: ; The interpolated data corresponding to the missing values ​​is calculated using the Kriging space interpolation method and is represented as follows: in For known location The measured value, Let i be the weight of the i-th data; Based on the time series ARIMA model, computational interpolation is used to repair missing data. The model is represented as follows: in For time series forecast values, For spatial interpolation results, As a weighting factor; Sensor data is calibrated using an adaptive weighted fusion algorithm to eliminate drift errors; the formula is as follows: in, For the first The sensor for the first The measured value of a physical quantity For the first The weight of each sensor, To calibrate the bias term, the weight update rule is as follows: in, For the region The true value of the physical quantity Measure the variance of the sensor cluster; Data alignment and fusion processing are performed on the preprocessed multimodal data, including: Based on the PTP clock synchronization and Kriging spatial interpolation methods to unify the timestamps and spatial coordinates of multi-source data respectively, the PTP clock synchronization formula is expressed as follows: in For sensor time, The time to send a synchronization message to the master clock. To receive synchronization message time from the clock, To send a delay request time from the clock, The master clock receives the delay request time; The Kalman filter method is used to fuse multimodal data, which is expressed as: Prediction step: , Update steps: , in, Here is the state transition matrix. To control the input matrix, , These are the covariances of process noise and observation noise, respectively. This is the observation matrix.

4. The adaptive energy-saving control method driven by building equipment energy consumption data according to claim 2, characterized in that, Extracting energy consumption features from multimodal data, specifically including: The sliding window statistical method is used to extract time-domain features to capture the time-series dynamic characteristics of device energy consumption, as shown below: ， ， ， in Let i be the energy consumption value at time i. To adjust the sliding window size, This refers to the current time point; Frequency domain features are extracted using Fast Fourier Transform to analyze the periodicity of energy consumption and anomalous frequency components, as shown below: , Main frequency amplitude: , Spectral entropy: , , in Indicates length is The energy consumption sequence, For the first Complex representation of each frequency component; Spatial features are extracted using graph convolutional network (GCN) embedding to uncover spatial correlation characteristics between devices, denoted as: in Given an adjacency matrix with self-loops, For degree matrix, For the first Layer node feature representation, For trainable weight matrix, This is an activation function whose output is a device embedding vector. .

5. The adaptive energy-saving control method driven by building equipment energy consumption data according to claim 2, characterized in that, Construct a unified spatiotemporal state input matrix to achieve multi-source fusion data modeling and processing, represented as: in For equipment operation data, For environmental conditions, For external meteorological data, For control commands, For the number of devices, For data dimension metrics; Spatiotemporal feature encoding based on graph convolutional-temporal attention networks includes: Spatial encoding: , Timing coding: , Fusion Output: , in To add self-connected device adjacency matrices, This is the spatial weight matrix. It is a temporal convolutional network. This is a multi-head attention mechanism; The physical equations of the equipment are incorporated into the data-driven model, and a loss function is constructed to embed physical constraints. The loss function is expressed as follows: in The mean square error between the predicted and actual values. For the physical equations of the equipment, , For the weighting factor; Based on temporal convolutional network The formula for constructing a prediction model with dilated convolution is expressed as follows: in As a void factor, For convolution kernel weights, For bias terms, It is the ReLU activation function; The attention weights are calculated and expressed as follows: , in For query vector, For key vectors, For vector dimensions; Monte Carlo Dropout forecasting is used to quantify uncertain data, represented as follows: in For the number of samples, The parameter distribution guided by Dropout.

6. The adaptive energy-saving control method driven by building equipment energy consumption data according to claim 2, characterized in that, Solving optimal decision-making for multi-objective problems based on the NSGA-II algorithm specifically includes: 1) Establish the core optimization objective function, including: a. Minimize energy consumption costs: in For a moment Electricity purchased from the power grid, For time-of-use electricity pricing, This refers to the discharge power of the energy storage battery. This is the battery cycle loss cost coefficient; b. Maximize thermal comfort: The optimal solution is obtained when PMV is close to 0. The predicted average vote value for region i is calculated as follows: in This refers to the human body's metabolic rate. For heat load; c. Minimize equipment wear and tear: in This refers to the number of times the equipment has been started and stopped. This represents the absolute value of the change in equipment load rate. , These are the wear coefficients for start-up / shutdown and load changes, respectively; 2) Set constraints, including: Indoor environmental constraints: , Equipment physical constraints: , Power grid interaction constraints: in For equipment power, Limitation of power change rate; 3) The steps for solving the optimal decision using the NSGA-II algorithm include: a. Using real-number encoding, each individual represents a control strategy. The population is initialized as follows: in Set the temperature value for region i. For the power adjustment of the air conditioning system, For adjusting the lighting brightness ratio, To schedule the target state of charge of energy storage batteries; b. Set up individuals With individuals The dominance relationship between them, in the individual Dominant Individual When, if and only if ,and The population is divided into multiple non-dominated hierarchical levels and ordered hierarchically. c. For individuals within the same non-dominated layer, calculate their distribution density in the target space, expressed as: in , These are the adjacent individuals after sorting by target m. These are the maximum and minimum values ​​of the target m in the current population, respectively. d. Select individuals from the parent generation for competition, choosing individuals with high non-dominant hierarchy and high crowding. (This involves comparing two parent individuals.) , Generate offspring , , represented as: in The crossover index, which introduces a perturbation to a specific gene locus in individual x, is expressed as: in are random numbers and , The variation index is used to merge the parent and offspring populations and select new populations according to the priority of higher non-dominant level or greater crowding in the same level. 4) Obtain a set of non-dominated solutions as the Pareto optimal solution set. Each solution represents a trade-off strategy. Use fuzzy decision-making to select a compromise solution, expressed as: in To achieve the ideal target value, select As the largest solution .

7. The adaptive energy-saving control method driven by building equipment energy consumption data according to claim 2, characterized in that, The steps for prioritization and dynamic resource allocation include: 1) Establish a policy-instruction mapping model to transform optimization policies into low-level instructions that can be executed by the device. The model is represented as follows: in To optimize the layer output of the ideal control quantity, For the set of instructions allowed by the device, For the safety cost function, This is the safety weighting coefficient; 2) Arbitration of priority conflicts among multiple devices based on dynamic weighted priority sorting, expressed as: in The energy-saving benefits of equipment k regulation. The comfort impact index is the reciprocal of the PMV deviation. The device's health is rated, with lower scores indicating higher priority. They are dynamic weights and The arbitration rule is to execute instructions with higher priority first, or to sort instructions of the same priority according to the device's response speed. 3) Introduce an adaptive retry mechanism to ensure accurate instruction execution and feedback of status, represented as: in This is the current number of retries. Based on the retry interval, To determine the maximum retry interval, a timeout timer is started after the first command is issued. If no response is received within the timeout period, the device is retried according to the exponential backoff strategy. Once the maximum number of retries is reached, the device is marked as faulty. 4) Establish a mobile security boundary model, and update the security threshold according to the real-time status. The model is represented as follows: in For risk function, The learning rate is the number of consecutive periods when the virus is detected. The conditions for triggering the update of the security threshold are as follows.

Images