Resource control decision-making method for distributed energy storage network

Abstract

The present invention belongs to the technical field of energy storage resource regulation and control, and specifically relates to a distributed energy storage network resource regulation and control decision-making method, including the following steps: multi-level data collection and processing; establishing a two-way interactive mechanism with the user end, collecting the user's electricity demand and energy storage willingness, and sending a demand response signal to the user when the power is tight or the electricity price fluctuates, encouraging the user to adjust the electricity behavior; based on the multi-level data and the user's demand response results, constructing a dynamic energy routing model of the distributed energy storage network to predict future energy demand and supply. The present invention can comprehensively utilize multi-level data, improve the accuracy of energy demand prediction, enhance the flexibility and intelligence level of the regulation strategy, adopt a global optimization method for resource regulation, and improve the robustness and reliability of the system.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage resource regulation and control, and in particular relates to a distributed energy storage network resource regulation and control decision-making method. Background Art

[0002] With the transformation of energy structure and the increasing proportion of renewable energy (such as wind energy, solar energy, etc.) in the power system, the challenges faced by traditional centralized power grids are becoming increasingly severe. Renewable energy is intermittent and volatile, which brings great uncertainty to the stable operation of the power grid. Distributed energy storage systems, as an important means of regulating the balance of power supply and demand, can effectively alleviate this problem. However, existing technologies still have many shortcomings in the regulation of distributed energy storage network resources, which are specifically manifested in the following aspects:

[0003] Low data utilization: Existing distributed energy storage network control methods usually rely on a single level of data, such as device layer data or regional layer data, and lack comprehensive utilization of network-wide data. This leads to low data utilization and makes it difficult to fully and accurately reflect the operating status and demand changes of the power grid, affecting the scientificity and effectiveness of control decisions.

[0004] Insufficient prediction accuracy: Traditional energy demand prediction methods mostly rely on simple time series analysis or empirical rules, which are difficult to fully consider complex factors such as changes in user behavior and real-time electricity price fluctuations, resulting in insufficient prediction accuracy. This situation is particularly evident in power grids with a high proportion of renewable energy, which affects the optimization and regulation of energy storage systems.

[0005] Lack of control flexibility: Existing control strategies are mostly static or preset plans, which are difficult to respond to dynamic changes in the power grid environment in real time and lack control flexibility. The user demand response mechanism is imperfect and user participation is low, which limits the full utilization of demand-side resources.

[0006] Traditional optimization methods mostly focus on local optimization and lack a global coordination mechanism. Due to the wide geographical distribution, diverse types and different scales of distributed energy storage systems, how to efficiently regulate resources and make decisions has become a hot topic and difficulty in current research.

[0007] In response to the above problems, a new distributed energy storage network resource regulation decision-making method is urgently needed. Summary of the invention

[0008] Based on the above problems, the purpose of the present invention is to provide a distributed energy storage network resource control decision-making method that can comprehensively utilize multi-level data, improve the accuracy of energy demand forecasting, enhance the flexibility and intelligence level of control strategies, use global optimization methods for resource control, and improve the robustness and reliability of the system.

[0009] The distributed energy storage network resource control decision-making method comprises the following steps:

[0010] S1. Multi-level data collection and processing: Collect multi-level data of distributed energy storage systems, including equipment operation data at the micro level, regional power grid data at the meso level, and whole-network data at the macro level, and process the multi-level data synchronously;

[0011] S2. Two-way interactive user demand response: Establish a two-way interactive mechanism with the user side to collect user electricity demand and energy storage willingness. When power is tight or electricity prices fluctuate, send demand response signals to users to encourage users to adjust their electricity consumption behavior and provide corresponding economic incentives.

[0012] S3. Dynamic energy routing optimization: Based on multi-level data and user demand response results, a dynamic energy routing model of the distributed energy storage network is constructed to predict future energy demand and supply. The establishment of the dynamic energy routing model includes:

[0013] S31. Energy demand forecasting: Based on historical energy demand and supply data, considering data at all levels and feedback from user demand responses, an energy demand forecasting model based on machine learning algorithms is constructed. Input features include historical load, electricity price fluctuations, and user response data. The pre-processed multi-level data and user demand response data are used as model inputs for joint training.

[0014] S32. User demand response analysis: Analyze user response data, predict user behavior changes under different electricity prices and demand response signals, establish a user response model, and quantify user energy storage charging and discharging behaviors under different scenarios;

[0015] S33, Dynamic Optimization Algorithm: Build a dynamic optimization model with the goal of minimizing energy transmission loss and charging and discharging loss, use a distributed algorithm to solve the optimization problem, and ensure computational efficiency and real-time performance;

[0016] S34. Dynamic energy routing model: Based on energy demand prediction and dynamic optimization model, an energy routing model is established. Based on real-time collected data and prediction results, the energy transmission path is dynamically adjusted to optimize the charging and discharging strategy of the energy storage system to ensure efficient energy transmission and use.

[0017] Preferably, the multi-level data collection and processing of S1 specifically includes:

[0018] S11, Micro-level data collection: collect voltage, current, power and temperature data through sensors installed on energy storage devices;

[0019] S12, meso-level data collection: Install monitoring equipment at each node of the regional power grid to collect regional load and regional voltage level data;

[0020] S13, macro-level data collection: collect the load, frequency and voltage data of the whole network through the monitoring unit of the power grid dispatching center;

[0021] S14. Multi-level data synchronization processing: Aggregate the equipment operation data at the micro level, the regional power grid data at the meso level, and the whole network data at the macro level to the central data processing platform, use the timing synchronization technology to align the timestamps of data from different sources, and integrate the multi-level data into a unified data set through the data fusion algorithm.

[0022] Preferably, the two-way interactive user demand response in S2 specifically includes:

[0023] S21. User-side data collection and monitoring: Install smart meters at the user end to monitor and collect the user's electricity consumption data in real time, including electricity consumption, electricity consumption time, and type of electricity consumption equipment. Use smart sensors to collect data on the user's own energy storage equipment, including storage capacity, charging and discharging status, and equipment health status.

[0024] S22. Establishment of a two-way interactive mechanism: Establish a user interaction platform based on mobile applications and web platforms, through which users can view real-time electricity prices, electricity usage, and energy storage status information, and set personal preferences and response intentions;

[0025] S23. User demand response and regulation: When power is tight or electricity prices fluctuate, demand response signals are sent to users through the user interaction platform, suggesting that users adjust their electricity consumption behavior, including postponing electricity consumption or activating their own energy storage equipment;

[0026] S24. Economic incentive mechanism: According to the grid load and market electricity price, a dynamic electricity price mechanism is implemented. Users charge when electricity prices are low and discharge when electricity prices are high to obtain the price difference. Economic incentives are given to users who participate in demand response.

[0027] S25. Clean and standardize the user response data to generate a demand response data set that reflects the user's electricity consumption behavior and energy storage willingness.

[0028] Preferably, the energy demand prediction model is constructed based on an extreme learning machine (ELM) model, and the energy demand prediction model construction based on the ELM includes:

[0029] Model input construction: including historical load, regional load, electricity price fluctuations, and user response data;

[0030] Model building and joint training:

[0031] Model architecture design: Design the ELM model architecture, determine the number of neurons in the input layer, hidden layer, and output layer, randomly generate and fix the weights and biases of the hidden layer, and only train the weights of the output layer. Integrate the multi-level unified data set and the demand response data set into a comprehensive data set as the model input for joint training. The training is as follows:

[0032] Data partitioning: Divide the comprehensive dataset into training set, validation set, and test set to ensure that the training and evaluation data of the model are independent;

[0033] Random initialization: randomly generate the weights and biases of the ELM hidden layer and keep them fixed;

[0034] Training model: Use the training set to train the model, calculate the output layer weights by the least squares method, minimize the prediction error, evaluate the model performance on the validation set, adjust the model hyperparameters based on the validation results, test the model on the test set, and evaluate the model's generalization ability and prediction accuracy.

[0035] Preferably, the training of the ELM model includes randomly generating hidden layer parameters and solving output weights using the least squares method, specifically including:

[0036] Mapping from input layer to hidden layer:

[0037] Assume the input data set is ,in is the input feature vector of the a-th sample, is the corresponding output, represents n-dimensional real number space, represents m-dimensional real number space;

[0038] The output of the hidden layer is expressed as: , where X is the input matrix, W is the randomly generated weight matrix, and b is the randomly generated bias vector. is the activation function;

[0039] Mapping from hidden layer to output layer: The relationship between hidden layer output and output weight is:

[0040] , where H is the hidden layer output matrix, is the output weight matrix, Y is the target output matrix;

[0041] Solving output weights: Solving by least squares method :

[0042] ,in, is the Moore-Penrose pseudoinverse of H.

[0043] Preferably, the user response model is based on a gradient boosted tree (GBT) model, and the construction of the user response model specifically includes:

[0044] Extract key features from demand response datasets, including response time, response volume, response frequency, and electricity price sensitivity;

[0045] User response model establishment: Use user response data to establish a user response model, as follows:

[0046] Classification and clustering: Classify and cluster users to identify different types of user groups. User groups are divided into high-response users and low-response users.

[0047] Feature engineering: construct feature vectors of user responses, including historical response data and electricity price fluctuation data;

[0048] The GBT model builds a series of weak learners to improve model performance and sets the parameters of the GBT model, including the number of trees, the maximum depth of the tree, the learning rate, and the minimum number of sample splits;

[0049] Training and validation: Use user response datasets to train the model, evaluate model performance through cross-validation, and adjust model parameters;

[0050] Prediction of user behavior changes: Simulate user scenarios under different electricity prices and demand response signals, generate prediction data under multiple scenarios, use the GBT model to predict user behavior changes under different electricity prices and demand response signals, and calculate the user's response probability and response amount under different scenarios:

[0051] Input features: Input simulated electricity prices, demand response signals, and user historical response data into the GBT model.

[0052] Prediction output: The GBT model outputs the user’s response probability in the current situation and response volume ;

[0053] Quantification of energy storage charging and discharging behavior: Simulate the charging and discharging behavior of user energy storage devices in different scenarios, including:

[0054] Response strategy definition: define the user's response strategy under different electricity prices and demand response signals;

[0055] Behavior quantification: Based on the user response model, quantify the user's energy storage charging and discharging behavior in different scenarios, and calculate the specific charging and discharging amount and time;

[0056] Benefit evaluation: Evaluate the economic benefits of users under different response strategies, including electricity bill savings, incentives and rewards, and quantify the comprehensive benefits of response behavior.

[0057] Preferably, the energy storage charging and discharging behavior quantification specifically includes:

[0058] Define user response strategies in different situations:

[0059] During high electricity price periods: users reduce electricity consumption or discharge electricity;

[0060] During low electricity price periods: users increase charging or reduce discharging;

[0061] Behavior quantification: Calculate the energy storage charge and discharge amount of users in different situations and :

[0062] ;

[0063] ;

[0064] Among them, MaxCharge and MaxDischarge are the maximum charging and discharging capabilities of the user's device;

[0065] Benefit evaluation: Calculate the economic benefits of users under different response strategies :

[0066] ;

[0067] in, Indicates the price of electricity;

[0068] Evaluate incentives for users to participate in demand response, including electricity bill discounts and point rewards.

[0069] Preferably, the construction of a dynamic optimization model with the goal of minimizing energy transmission loss and charge and discharge loss specifically includes:

[0070] The optimization objective function is as follows:

[0071] ;

[0072] in, is the power loss of the ith energy transmission path, is the transmission time of the ith energy transmission path, is the charging loss power of the jth energy storage device, is the charging time of the jth energy storage device, is the discharge power loss of the kth energy storage device, is the discharge time of the kth energy storage device; N, M, and L are the number of energy transmission paths, charged energy storage devices, and discharged energy storage devices, respectively;

[0073] Constraints:

[0074] User response behavior constraints: User behavior data predicted by the user response model, including response probabilities under different electricity prices and demand response signals and response volume , incorporate user response data into the optimization model and define user response behavior constraints:

[0075] ;

[0076] ;

[0077] in, and are the response probabilities of energy storage devices j and k in the current situation, , are the maximum charge and discharge capacity of the energy storage device, represents the expected energy of energy storage device j under charging operation, represents the expected energy of the energy storage device k under discharge operation;

[0078] Energy balance constraints: ;in, , , They are energy transfer, charging and discharging energy;

[0079] Energy storage equipment capacity constraints:

[0080] ;

[0081] ;

[0082] in, and They are the upper limits of the charging and discharging capacity of the energy storage device;

[0083] The use of distributed algorithms to solve optimization problems specifically includes:

[0084] Problem decomposition: Decompose the global optimization problem into several sub-problems, each of which corresponds to an energy storage device or transmission line, and use the ADMM algorithm to decompose;

[0085] Sub-problem solving: Each sub-problem is solved independently locally to calculate the corresponding charging power, discharging power and transmission power;

[0086] Global coordination: The global coordinator aggregates the solution results of each sub-problem, updates global variables, and ensures the satisfaction of global energy balance and constraints;

[0087] Iterative update: In each round of iteration, the subproblems and the global coordinator exchange information, update their respective optimization variables, and gradually converge to the global optimal solution.

[0088] Preferably, the construction of the dynamic energy routing model in S34 specifically includes:

[0089] S341, based on the output results of the energy demand prediction and dynamic optimization model, including predicted energy demand, user response behavior prediction data, and optimization strategies for minimizing energy transmission loss and charging and discharging loss;

[0090] Routing model construction: Build an energy routing model to define the selection and scheduling strategies of transmission paths, including:

[0091] Define the nodes and edges in the distributed energy storage network and establish the network topology. The nodes include energy storage equipment, power generation equipment, and load centers, and the edges include energy transmission lines. Calculate the optimal energy transmission path based on the path selection algorithm. Calculate the weight of the transmission path based on the energy demand forecast and dynamic optimization results. The weight includes energy transmission loss and transmission time.

[0092] S342, dynamically adjust the energy transmission path:

[0093] Real-time data collection: Through a distributed sensor network, real-time data on the operation status of the energy storage system and the power grid is collected, including voltage, current, power, energy demand, and user response;

[0094] Path adjustment algorithm: Based on real-time collected data and energy demand forecast results, the path adjustment algorithm is used to dynamically adjust the energy transmission path;

[0095] S343. Optimize the charging and discharging strategy of the energy storage system:

[0096] Based on energy demand prediction and user response behavior model, the charging and discharging strategy of the energy storage system is optimized, including:

[0097] Use real-time data to calibrate energy demand forecast results to improve forecast accuracy; dynamically adjust the charging and discharging time and power of the energy storage system according to the charging and discharging strategy output by the optimization model to minimize energy loss and economic costs; optimize the charging and discharging behavior of user-side energy storage equipment according to the prediction results of the user response model;

[0098] S344, output optimized energy transmission path and energy storage control strategy:

[0099] Output the optimized energy transmission path to ensure that energy can be transmitted to the demand node to meet the load demand; publish the optimized energy storage control strategy to each energy storage device and monitor its execution in real time.

[0100] Preferably, the network topology is expressed as: , where V represents the node set and E represents the edge set;

[0101] The path selection algorithm uses the shortest path algorithm to calculate the optimal energy transfer path:

[0102] ,in, It is the edge The weight of is the path selection variable;

[0103] Path weight calculation: ,in, and is the weight coefficient, is the energy transmission loss power, is the energy transfer time;

[0104] The path adjustment algorithm is based on real-time path updates and is expressed as the formula:

[0105] ,in, is the path weight at time t, and is the transmission loss power and transmission time at time t;

[0106] Adaptive Adjustment: ,in, is the adjustment step size, is the transmission loss power at time t+1, is the actual demand at time t, is the predicted energy demand at time t.

[0107] Beneficial effects of the present invention:

[0108] The present invention collects and integrates multi-level data at the micro, meso and macro levels to establish an energy demand prediction model based on the extreme learning machine (ELM), thereby improving the accuracy of energy demand prediction. Combined with the analysis results of the user response model, the energy storage charging and discharging behavior of users in different situations is quantified to ensure more accurate prediction results. The optimized energy transmission path and energy storage control strategy are dynamically adjusted based on the real-time collected data and prediction results to ensure efficient energy transmission and use, and maximize the utilization rate and economic benefits of the energy storage system.

[0109] The present invention collects the user's electricity demand and energy storage willingness by establishing a two-way interactive mechanism with the user end, sends a demand response signal to the user when the power is tight or the electricity price fluctuates, encourages the user to adjust the electricity behavior and gives corresponding economic incentives. The user response model analyzes the user's behavior changes under different electricity prices and demand response signals, predicts the response probability and response amount, and improves the system's adaptability to changes in user demand.

[0110] The present invention constructs an energy routing model based on energy demand prediction and dynamic optimization model, and uses distributed optimization algorithm to adjust the energy transmission path and the charging and discharging strategy of the energy storage system in real time. Through adaptive adjustment and iterative optimization, the efficiency and reliability of energy transmission and energy storage regulation are ensured, and the flexibility and responsiveness of the system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] Figure 1 A schematic diagram of a decision-making method flow chart of an embodiment of the present invention;

[0112] Figure 2 Schematic diagram of establishing a dynamic energy routing model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0113] The present invention is further described in detail below in conjunction with specific embodiments.

[0114] like Figure 1-Figure 2 As shown, the distributed energy storage network resource control decision-making method includes the following steps:

[0115] S1. Multi-level data collection and processing: Collect multi-level data of distributed energy storage systems, including equipment operation data at the micro level (voltage, current, power, temperature), regional power grid data at the meso level (regional load, regional voltage level), and whole-network data at the macro level (whole-network load, frequency, voltage, etc.), and process multi-level data synchronously;

[0116] S2. Two-way interactive user demand response: Establish a two-way interactive mechanism with the user side to collect user electricity demand and energy storage willingness. When power is tight or electricity prices fluctuate, send demand response signals to users to encourage them to adjust their electricity consumption behavior (such as postponing electricity consumption or activating their own energy storage equipment) and provide corresponding economic incentives.

[0117] S3. Dynamic energy routing optimization: Based on multi-level data and user demand response results, a dynamic energy routing model of the distributed energy storage network is constructed to predict future energy demand and supply. The establishment of the dynamic energy routing model includes:

[0118] S31. Energy demand forecasting: Based on historical energy demand and supply data, considering data at all levels and feedback from user demand responses, an energy demand forecasting model based on machine learning algorithms is constructed. Input features include historical load, electricity price fluctuations, and user response data. The pre-processed multi-level data and user demand response data (i.e., unified data set, comprehensive data set) are used as model inputs for joint training.

[0119] S32. User demand response analysis: Analyze user response data, predict user behavior changes under different electricity prices and demand response signals, establish a user response model, and quantify user energy storage charging and discharging behaviors under different scenarios;

[0120] S33, Dynamic Optimization Algorithm: Build a dynamic optimization model with the goal of minimizing energy transmission loss and charging and discharging loss, use a distributed algorithm to solve the optimization problem, and ensure computational efficiency and real-time performance;

[0121] S34. Dynamic energy routing model: Based on energy demand prediction and dynamic optimization model, an energy routing model is established. Based on real-time collected data and prediction results, the energy transmission path is dynamically adjusted to optimize the charging and discharging strategy of the energy storage system to ensure efficient energy transmission and use. The model input includes multi-level data collected in real time and user demand response results, and the output is the optimized energy transmission path and energy storage control strategy.

[0122] The main steps of the dynamic energy routing model are as follows:

[0123] Node and edge definition: Establish the topology of the energy storage network, define the nodes in the network and the edges connecting these nodes.

[0124] Path selection algorithm: Use the shortest path algorithm (such as Dijkstra) to calculate the optimal path from the start node to the target node.

[0125] Path weight calculation: Calculate the transmission loss and time of the path and use it as the weight for path selection.

[0126] Real-time path update: Based on the real-time collected data, the path weight is dynamically adjusted and the optimal path is recalculated to ensure efficient energy transmission.

[0127] Charge and discharge optimization: Based on energy demand forecast and user response behavior model, optimize the charging and discharging strategy of the energy storage system to minimize energy loss and economic cost.

[0128] Response behavior control: According to the prediction results of the user response model, the charging and discharging behavior of the user-side energy storage equipment is optimized to improve the flexibility and responsiveness of the system.

[0129] The multi-level data acquisition and processing in S1 provides basic data, which plays a key role in energy demand forecasting and dynamic optimization in S3.

[0130] The two-way interactive user demand response of S2 directly affects the execution of user demand response analysis and dynamic optimization algorithm in S3 by collecting user electricity demand and response willingness, ensuring that the model can reflect changes in user behavior.

[0131] The dynamic energy routing optimization model of S3 generates dynamic energy routing solutions based on the data of S1 and S2 using prediction and optimization algorithms.

[0132] S1's multi-level data collection and processing specifically includes:

[0133] S11, Micro-level data collection: The voltage, current, power and temperature data are collected through sensors installed on the energy storage device, and the collected data are transmitted to the local edge computing node for preliminary processing using the wireless sensor network. The data is denoised and filtered on the edge computing node, and a pre-processed equipment operation data set is generated;

[0134] S12, meso-level data collection: Install monitoring equipment at each node of the regional power grid to collect regional load and regional voltage level data, use the regional communication network to transmit the data collected by the monitoring equipment to the regional control center, integrate the data at the regional control center, and generate a regional power grid data set;

[0135] S13, macro-level data collection: The monitoring unit of the power grid dispatching center collects the load, frequency and voltage data of the whole network, and transmits the data to the database of the power grid dispatching center through the power-specific communication network. The data is aggregated and processed in the power grid dispatching center to generate a data set for the whole network;

[0136] S14. Multi-level data synchronization processing: Aggregate the equipment operation data at the micro level, the regional power grid data at the meso level, and the whole network data at the macro level to the central data processing platform, use the timing synchronization technology to align the timestamps of data from different sources to ensure the timeliness and consistency of the data, and integrate the multi-level data into a unified data set through the data fusion algorithm, covering the multi-dimensional information of equipment, region and the whole network.

[0137] Convert data from different data sources into a unified unit, normalize data of different ranges to the same range (such as 0 to 1) to make them have the same scale for subsequent processing and fusion. Use the weighted average method to assign different weights to data at different levels according to the importance and credibility of the data source, calculate the weighted average, fuse the data at the micro, meso and macro levels, and generate a unified data set. The fusion process needs to consider the correlation and complementarity between the data to ensure the integrity and consistency of the data after fusion.

[0138] The two-way interactive user demand response in S2 specifically includes:

[0139] S21. User-side data collection and monitoring: Install smart meters at the user end to monitor and collect the user's electricity consumption data in real time, including electricity consumption, electricity consumption time, and type of electricity-consuming equipment. Collect data on the user's own energy storage equipment through smart sensors, including storage capacity, charging and discharging status, and equipment health status. Use the Internet of Things technology to transmit the user-side data through the wireless network to the central data processing platform for unified storage and management;

[0140] S22. Establishment of a two-way interactive mechanism: Establish a user interaction platform based on mobile applications (APP) and web platforms, through which users can view real-time electricity prices, electricity usage, and energy storage status information, and set personal preferences and response intentions;

[0141] S23. User demand response and regulation: When power is tight or electricity prices fluctuate, demand response signals are sent to users through the user interaction platform, suggesting that users adjust their electricity usage, including postponing electricity use or activating their own energy storage devices. Multiple response options are provided for users to choose from, such as immediate response, partial response, temporary non-response, etc., and users are allowed to set automatic response rules;

[0142] S24. Economic incentive mechanism: According to the grid load and market electricity price, a dynamic electricity price mechanism is implemented. Users charge when the electricity price is low and discharge when the electricity price is high to obtain the price difference. Economic incentives are given to users who participate in demand response, including electricity fee discounts, point rewards, and cash back.

[0143] Through the user interaction platform, the effects of demand response and the economic incentives obtained are fed back to users, encouraging users to continue to participate in demand response and improve the responsiveness of the overall system.

[0144] S25. Clean and standardize the user response data to generate a demand response data set that reflects the user's electricity consumption behavior and energy storage willingness.

[0145] The construction of the energy demand prediction model is based on the extreme learning machine ELM model. The construction of the energy demand prediction model based on ELM includes:

[0146] Model input construction: including historical load, regional load, electricity price fluctuations, and user response data;

[0147] Model building and joint training:

[0148] Model architecture design: Design the ELM model architecture, determine the number of neurons in the input layer, hidden layer, and output layer, randomly generate and fix the weights and biases of the hidden layer, and only train the weights of the output layer. Integrate the multi-level unified data set and the demand response data set into a comprehensive data set as the model input for joint training. The training is as follows:

[0149] Data partitioning: Divide the comprehensive dataset into training set, validation set, and test set to ensure that the training and evaluation data of the model are independent;

[0150] Random initialization: randomly generate the weights and biases of the ELM hidden layer and keep them fixed;

[0151] Training model: Use the training set to train the model, calculate the output layer weights by the least squares method, minimize the prediction error, evaluate the model performance on the validation set, adjust the model hyperparameters (the number of hidden layer neurons) based on the validation results, test the model on the test set, and evaluate the model's generalization ability and prediction accuracy;

[0152] After the model is deployed, real-time data is continuously collected, and model parameters are regularly updated using online learning methods to improve the model's adaptability and prediction accuracy. The model's prediction performance is monitored in real time, and the model structure and parameter settings are continuously optimized through error analysis and feedback mechanisms.

[0153] The training of the ELM model includes randomly generating hidden layer parameters and solving the output weights using the least squares method, specifically including:

[0154] Mapping from input layer to hidden layer:

[0155] Assume the input data set is ,in is the input feature vector of the a-th sample, is the corresponding output, represents n-dimensional real number space, represents m-dimensional real number space;

[0156] The output of the hidden layer is expressed as: , where X is the input matrix, W is the randomly generated weight matrix, and b is the randomly generated bias vector. is the activation function (tanh is used);

[0157] Mapping from hidden layer to output layer: The relationship between hidden layer output and output weight is:

[0158] , where H is the hidden layer output matrix, is the output weight matrix, Y is the target output matrix;

[0159] Solving output weights: Solving by least squares method :

[0160] ,in, is the Moore-Penrose pseudoinverse of H.

[0161] Application of the ELM model in the present invention: The steps of constructing an energy demand prediction model based on ELM are as follows:

[0162] Input features including historical load, electricity price fluctuations, user response data, meteorological data, regional load and voltage levels are selected from a comprehensive dataset and standardized.

[0163] ELM model training:

[0164] Input layer: Let the input feature vector be ;

[0165] Hidden layer: randomly generate weight matrix W and bias vector b, and calculate hidden layer output H;

[0166] Output layer: Use the training data to solve the output weights through the least squares method ;

[0167] The validation set was used to evaluate the model performance and the model hyperparameters (number of neurons in the hidden layer) were adjusted based on the validation results.

[0168] Continuously collect real-time data and use incremental training to regularly update model parameters.

[0169] Deploy the trained ELM model for real-time energy demand forecasting, monitor the model’s forecasting performance in real time, and continuously optimize the model structure and parameter settings through error analysis and feedback mechanism.

[0170] Through the above calculation steps, an energy demand prediction model based on extreme learning machine (ELM) can be constructed based on historical energy demand and supply data, combined with data at all levels and feedback from user demand responses. The preprocessed multi-level data and user demand response data are used as model inputs for joint training to improve the prediction accuracy and adaptability of the model, providing reliable data support for dynamic energy routing optimization.

[0171] The user response model is based on the gradient boosted tree (GBT) model. The construction of the user response model specifically includes:

[0172] Extract key features from demand response datasets, including response time, response volume, response frequency, and electricity price sensitivity;

[0173] User response model establishment: Use user response data to establish a user response model, as follows:

[0174] Classification and clustering: Classify and cluster users to identify different types of user groups. User groups are divided into high-response users and low-response users. K-means clustering is used to achieve this.

[0175] Feature engineering: construct feature vectors of user responses, including historical response data and electricity price fluctuation data;

[0176] The GBT model builds a series of weak learners (decision trees) to improve model performance and sets the parameters of the GBT model, including the number of trees, the maximum depth of the tree, the learning rate, and the minimum number of sample splits;

[0177] Training and validation: Use user response datasets to train the model, evaluate model performance through cross-validation, and adjust model parameters;

[0178] Training based on the GBT algorithm, the specific steps include:

[0179] Initial model: Train the first weak learner, usually a simple regression tree.

[0180] Residual calculation: Calculate the prediction residual of the current model, that is, the difference between the true value and the predicted value.

[0181] Iterative training: In each iteration, a new regression tree is trained to fit the residuals from the previous iteration, and the overall model is updated.

[0182] Model combination: The prediction results of all weak learners are weighted averaged to obtain the final prediction model.

[0183] GBT model training detailed steps:

[0184] Initial model: , usually, the initial model Set to the mean of the target values ​​in the training set, is the model parameter; A is the number of samples;

[0185] Residual calculation: For each iteration , calculate the residual :

[0186] ,in, is the loss function, such as the squared error ;

[0187] Fitting the residuals: training a new regression tree Fitting residuals ;

[0188] Update the model to: ,in, is the learning rate, usually set to a small value, set to 0.1;

[0189] The final model is the weighted sum of all weak learners:

[0190] .

[0191] Prediction of user behavior changes: Simulate user scenarios under different electricity prices and demand response signals, generate prediction data under multiple scenarios, use the GBT model to predict user behavior changes under different electricity prices and demand response signals, and calculate the user's response probability and response amount under different scenarios:

[0192] Input features: Input simulated electricity prices, demand response signals, and user historical response data into the GBT model.

[0193] Prediction output: The GBT model outputs the user’s response probability in the current situation and response volume ;

[0194] Quantification of energy storage charging and discharging behavior: Simulate the charging and discharging behavior of user energy storage devices in different scenarios, including:

[0195] Response strategy definition: define the user's response strategy under different electricity prices and demand response signals, such as postponing electricity use, charging in advance, delaying discharge, etc.;

[0196] Behavior quantification: Based on the user response model, quantify the user's energy storage charging and discharging behavior in different scenarios, and calculate the specific charging and discharging amount and time;

[0197] Benefit evaluation: Evaluate the economic benefits of users under different response strategies, including electricity bill savings, incentives and rewards, and quantify the comprehensive benefits of response behavior.

[0198] The quantification of energy storage charging and discharging behavior specifically includes:

[0199] Define user response strategies in different situations:

[0200] During high electricity price periods: users reduce electricity consumption or discharge electricity;

[0201] During low electricity price periods: users increase charging or reduce discharging;

[0202] Behavior quantification: Calculate the energy storage charge and discharge amount of users in different situations and :

[0203] ;

[0204] ;

[0205] Among them, MaxCharge and MaxDischarge are the maximum charging and discharging capabilities of the user's device;

[0206] Benefit evaluation: Calculate the economic benefits of users under different response strategies :

[0207] ;

[0208] in, Indicates the price of electricity;

[0209] Evaluate incentives for users to participate in demand response, including electricity bill discounts and point rewards.

[0210] Through the above specific steps, a user response model can be established based on user response data, and the gradient boosting tree (GBT) algorithm can be used to predict user behavior changes under different electricity prices and demand response signals, and the user's energy storage charging and discharging behavior under different scenarios can be quantified, providing a scientific basis and optimization strategy for distributed energy storage network resource regulation.

[0211] The construction of a dynamic optimization model with the goal of minimizing energy transmission loss and charging and discharging loss specifically includes:

[0212] The optimization objective function is as follows:

[0213] The optimization objective function is as follows:

[0214] ;

[0215] in, is the power loss of the ith energy transmission path, is the transmission time of the ith energy transmission path, is the charging loss power of the jth energy storage device, is the charging time of the jth energy storage device, is the discharge power loss of the kth energy storage device, is the discharge time of the kth energy storage device; N, M, and L are the number of energy transmission paths, charged energy storage devices, and discharged energy storage devices, respectively;

[0216] Constraints:

[0217] User response behavior constraints: User behavior data predicted by the user response model, including response probabilities under different electricity prices and demand response signals and response volume , incorporate user response data into the optimization model and define user response behavior constraints:

[0218] ;

[0219] ;

[0220] in, and are the response probabilities of energy storage devices j and k in the current situation, , are the maximum charge and discharge capacity of the energy storage device, represents the expected energy of energy storage device j under charging operation, represents the expected energy of the energy storage device k under discharge operation;

[0221] Energy balance constraints: ;in, , , They are energy transfer, charging and discharging energy;

[0222] Energy storage equipment capacity constraints:

[0223] ;

[0224] ;

[0225] in, and They are the upper limits of the charging and discharging capacity of the energy storage device;

[0226] Using distributed algorithms to solve optimization problems specifically includes:

[0227] Problem decomposition: Decompose the global optimization problem into several sub-problems, each of which corresponds to an energy storage device or transmission line, using the ADMM (Alternating Direction Method of Multipliers) algorithm;

[0228] Sub-problem solving: Each sub-problem is solved independently locally to calculate the corresponding charging power, discharging power and transmission power, as follows:

[0229] ;

[0230] ;

[0231] ;

[0232] in, is the maximum transmission power of energy transmission;

[0233] Global coordination: The global coordinator aggregates the solution results of each sub-problem, updates global variables, and ensures the satisfaction of global energy balance and constraints;

[0234] Iterative update: In each round of iteration, the sub-problems and the global coordinator exchange information, update their respective optimization variables, and gradually converge to the global optimal solution. The iterative update formula is as follows:

[0235] ;

[0236] ;

[0237] .

[0238] in, is the Lagrange multiplier, x and z represent variables, c is the vector of constant terms on the right side representing the constraints, E and B are matrices used to map variables x and z to the constraints, is the regularization parameter, , is the objective function with respect to variables x and z.

[0239] Through the above steps, a dynamic optimization model with the goal of minimizing energy transmission loss and charging and discharging loss can be constructed, and a distributed algorithm can be used to solve the optimization problem to achieve efficient regulation and optimization of the distributed energy storage network.

[0240] The construction of the dynamic energy routing model in S34 specifically includes:

[0241] S341. Model input: output results based on energy demand prediction and dynamic optimization models, including predicted energy demand, user response behavior prediction data, and optimization strategies to minimize energy transmission loss and charging and discharging loss; model input data includes energy demand prediction results , user response behavior prediction data and And the output strategy of the dynamic optimization model and ;

[0242] Routing model construction: Build an energy routing model to define the selection and scheduling strategies of transmission paths, including:

[0243] Node and edge definition: define nodes and edges in the distributed energy storage network and establish the network topology. Nodes include energy storage equipment, power generation equipment, and load centers, and edges include energy transmission lines.

[0244] Path selection algorithm: Based on the path selection algorithm, the optimal energy transmission path is calculated;

[0245] Path weight calculation: Calculate the weight of the transmission path based on energy demand prediction and dynamic optimization results. The weight includes energy transmission loss and transmission time.

[0246] S342, dynamically adjust the energy transmission path:

[0247] Real-time data collection: Through a distributed sensor network, real-time data on the operation status of the energy storage system and the power grid is collected, including voltage, current, power, energy demand, and user response;

[0248] Path adjustment algorithm: Based on real-time collected data and energy demand forecast results, the path adjustment algorithm is used to dynamically adjust the energy transmission path;

[0249] S343. Optimize the charging and discharging strategy of the energy storage system:

[0250] Based on energy demand prediction and user response behavior model, the charging and discharging strategy of the energy storage system is optimized, including:

[0251] Use real-time data to calibrate energy demand forecast results to improve forecast accuracy; dynamically adjust the charging and discharging time and power of the energy storage system according to the charging and discharging strategy output by the optimization model to minimize energy loss and economic costs; optimize the charging and discharging behavior of user-side energy storage equipment according to the prediction results of the user response model to improve the flexibility and responsiveness of the overall system;

[0252] S344, output optimized energy transmission path and energy storage control strategy:

[0253] Output the optimized energy transmission path to ensure that energy can be transmitted to the demand node to meet the load demand; publish the optimized energy storage control strategy to each energy storage device and monitor its execution in real time.

[0254] The network topology is expressed as: , where V represents the node set and E represents the edge set;

[0255] The path selection algorithm uses the shortest path algorithm to calculate the optimal energy transfer path:

[0256] ,in, It is the edge The weights of (such as energy loss), is the path selection variable (if the edge 1 if in the path, 0 otherwise ;

[0257] Path weight calculation: ,in, and is the weight coefficient, is the energy transmission loss power, is the energy transfer time;

[0258] The path adjustment algorithm is based on real-time path updates and is expressed as the formula:

[0259] ,in, is the path weight at time t, and is the transmission loss power and transmission time at time t;

[0260] Adaptive Adjustment: ,in, is the adjustment step size, is the transmission loss power at time t+1, is the actual demand at time t, is the predicted energy demand at time t.

[0261] Optimize the charging and discharging strategy of the energy storage system, and the charging and discharging optimization formula:

[0262] ;

[0263] Constraints:

[0264] ;

[0265] ;

[0266] Prediction calibration: Real-time data calibration formula:

[0267] ,in, is the calibration factor, is the actual energy demand at time t;

[0268] Output optimized energy transmission path and energy storage control strategy:

[0269] Path output:

[0270] Optimal path formula: ;

[0271] Strategy Release:

[0272] Energy storage control strategy output:

[0273] ;

[0274] .

[0275] in, , They are charging and discharging strategies respectively.

Claims

1. A distributed energy storage network resource control decision-making method, characterized in that: The following steps are involved: S1. Multi-level data collection and processing: Collect multi-level data of distributed energy storage systems, including equipment operation data at the micro level, regional power grid data at the meso level, and whole-network data at the macro level, and process the multi-level data synchronously; S2. Two-way interactive user demand response: Establish a two-way interactive mechanism with the user side to collect user electricity demand and energy storage willingness. When power is tight or electricity prices fluctuate, send demand response signals to users to encourage users to adjust their electricity consumption behavior and provide corresponding economic incentives. S3. Dynamic energy routing optimization: Based on multi-level data and user demand response results, a dynamic energy routing model of the distributed energy storage network is constructed to predict future energy demand and supply. The establishment of the dynamic energy routing model includes: S31. Energy demand forecasting: Based on historical energy demand and supply data, considering data at all levels and feedback from user demand responses, an energy demand forecasting model based on machine learning algorithms is constructed. Input features include historical load, electricity price fluctuations, and user response data. The pre-processed multi-level data and user demand response data are used as model inputs for joint training. S32. User demand response analysis: Analyze user response data, predict user behavior changes under different electricity prices and demand response signals, establish a user response model, and quantify user energy storage charging and discharging behaviors under different scenarios; S33, Dynamic Optimization Algorithm: Build a dynamic optimization model with the goal of minimizing energy transmission loss and charging and discharging loss, use a distributed algorithm to solve the optimization problem, and ensure computational efficiency and real-time performance; S34. Dynamic energy routing model: Based on energy demand prediction and dynamic optimization model, an energy routing model is established. Based on real-time collected data and prediction results, the energy transmission path is dynamically adjusted to optimize the charging and discharging strategy of the energy storage system to ensure efficient energy transmission and use.

2. The distributed energy storage network resource control decision-making method according to claim 1 is characterized in that: The multi-level data collection and processing of S1 specifically includes: S11, Micro-level data collection: collect voltage, current, power and temperature data through sensors installed on energy storage devices; S12, meso-level data collection: Install monitoring equipment at each node of the regional power grid to collect regional load and regional voltage level data; S13, macro-level data collection: collect the load, frequency and voltage data of the whole network through the monitoring unit of the power grid dispatching center; S14. Multi-level data synchronization processing: Aggregate the equipment operation data at the micro level, the regional power grid data at the meso level, and the whole network data at the macro level to the central data processing platform, use the timing synchronization technology to align the timestamps of data from different sources, and integrate the multi-level data into a unified data set through the data fusion algorithm.

3. The distributed energy storage network resource control decision-making method according to claim 2 is characterized in that: The two-way interactive user demand response in S2 specifically includes: S21. User-side data collection and monitoring: Install smart meters at the user end to monitor and collect the user's electricity consumption data in real time, including electricity consumption, electricity consumption time, and type of electricity consumption equipment. Use smart sensors to collect data on the user's own energy storage equipment, including storage capacity, charging and discharging status, and equipment health status. S22. Establishment of a two-way interactive mechanism: Establish a user interaction platform based on mobile applications and web platforms, through which users can view real-time electricity prices, electricity usage, and energy storage status information, and set personal preferences and response intentions; S23. User demand response and regulation: When power is tight or electricity prices fluctuate, demand response signals are sent to users through the user interaction platform, suggesting that users adjust their electricity consumption behavior, including postponing electricity consumption or activating their own energy storage equipment; S24. Economic incentive mechanism: According to the grid load and market electricity price, a dynamic electricity price mechanism is implemented. Users charge when electricity prices are low and discharge when electricity prices are high to obtain the price difference. Economic incentives are given to users who participate in demand response. S25. Clean and standardize the user response data to generate a demand response data set that reflects the user's electricity consumption behavior and energy storage willingness.

4. The distributed energy storage network resource control decision-making method according to claim 1 is characterized in that: The energy demand prediction model is constructed based on an extreme learning machine (ELM) model. The energy demand prediction model based on the ELM model includes: Model input construction: including historical load, regional load, electricity price fluctuations, and user response data; Model building and joint training: Model architecture design: Design the ELM model architecture, determine the number of neurons in the input layer, hidden layer, and output layer, randomly generate and fix the weights and biases of the hidden layer, and only train the weights of the output layer. Integrate the multi-level unified data set and the demand response data set into a comprehensive data set as the model input for joint training. The training is as follows: Data partitioning: Divide the comprehensive dataset into training set, validation set, and test set to ensure that the training and evaluation data of the model are independent; Random initialization: randomly generate the weights and biases of the ELM hidden layer and keep them fixed; Training model: Use the training set to train the model, calculate the output layer weights by the least squares method, minimize the prediction error, evaluate the model performance on the validation set, adjust the model hyperparameters based on the validation results, test the model on the test set, and evaluate the model's generalization ability and prediction accuracy.

5. The distributed energy storage network resource control decision-making method according to claim 4 is characterized in that: The training of the ELM model includes randomly generating hidden layer parameters and solving output weights using the least squares method, specifically including: Mapping from input layer to hidden layer: Assume the input data set is ,in is the input feature vector of the a-th sample, is the corresponding output, represents n-dimensional real number space, represents m-dimensional real number space; The output of the hidden layer is expressed as: , where X is the input matrix, W is the randomly generated weight matrix, and b is the randomly generated bias vector. is the activation function; Mapping from hidden layer to output layer: The relationship between hidden layer output and output weight is: , where H is the hidden layer output matrix, is the output weight matrix, Y is the target output matrix; Solving output weights: Solving by least squares method : ,in, is the Moore-Penrose pseudoinverse of H.

6. The distributed energy storage network resource control decision-making method according to claim 1 is characterized in that: The user response model is based on the gradient boosted tree (GBT) model. The construction of the user response model specifically includes: Extract key features from demand response datasets, including response time, response volume, response frequency, and electricity price sensitivity; User response model establishment: Use user response data to establish a user response model, as follows: Classification and clustering: Classify and cluster users to identify different types of user groups. User groups are divided into high-response users and low-response users. Feature engineering: construct feature vectors of user responses, including historical response data and electricity price fluctuation data; The GBT model builds a series of weak learners to improve model performance and sets the parameters of the GBT model, including the number of trees, the maximum depth of the tree, the learning rate, and the minimum number of sample splits; Training and validation: Use user response datasets to train the model, evaluate model performance through cross-validation, and adjust model parameters; Prediction of user behavior changes: Simulate user scenarios under different electricity prices and demand response signals, generate prediction data under multiple scenarios, use the GBT model to predict user behavior changes under different electricity prices and demand response signals, and calculate the user's response probability and response amount under different scenarios: Input features: Input simulated electricity prices, demand response signals, and user historical response data into the GBT model. Prediction output: The GBT model outputs the user’s response probability in the current situation and response volume ; Quantification of energy storage charging and discharging behavior: Simulate the charging and discharging behavior of user energy storage devices in different scenarios, including: Response strategy definition: define the user's response strategy under different electricity prices and demand response signals; Behavior quantification: Based on the user response model, quantify the user's energy storage charging and discharging behavior in different scenarios, and calculate the specific charging and discharging amount and time; Benefit evaluation: Evaluate the economic benefits of users under different response strategies, including electricity bill savings, incentives and rewards, and quantify the comprehensive benefits of response behavior.

7. The distributed energy storage network resource control decision-making method according to claim 6 is characterized in that: The energy storage charging and discharging behavior quantification specifically includes: Define user response strategies in different situations: During high electricity price periods: users reduce electricity consumption or discharge electricity; During low electricity price periods: users increase charging or reduce discharging; Behavior quantification: Calculate the energy storage charge and discharge amount of users in different situations and : ； ； Among them, MaxCharge and MaxDischarge are the maximum charging and discharging capabilities of the user's device; Benefit evaluation: Calculate the economic benefits of users under different response strategies : ； in, Indicates the price of electricity; Evaluate incentives for users to participate in demand response, including electricity bill discounts and point rewards.

8. The distributed energy storage network resource control decision-making method according to claim 7 is characterized in that: The construction of a dynamic optimization model with the goal of minimizing energy transmission loss and charge and discharge loss specifically includes: The optimization objective function is as follows: ； in, is the power loss of the ith energy transmission path, is the transmission time of the ith energy transmission path, is the charging loss power of the jth energy storage device, is the charging time of the jth energy storage device, is the discharge power loss of the kth energy storage device, is the discharge time of the kth energy storage device; N, M, and L are the number of energy transmission paths, charged energy storage devices, and discharged energy storage devices, respectively; Constraints: User response behavior constraints: User behavior data predicted by the user response model, including response probabilities under different electricity prices and demand response signals and response volume , incorporate user response data into the optimization model and define user response behavior constraints: ； ； in, and are the response probabilities of energy storage devices j and k in the current situation, , are the maximum charge and discharge capacity of the energy storage device, represents the expected energy of energy storage device j under charging operation, represents the expected energy of the energy storage device k under discharge operation; Energy balance constraints: ;in, , , They are energy transfer, charging and discharging energy; Energy storage equipment capacity constraints: ； ； in, and They are the upper limits of the charging and discharging capacity of the energy storage device; The use of distributed algorithms to solve optimization problems specifically includes: Problem decomposition: Decompose the global optimization problem into several sub-problems, each of which corresponds to an energy storage device or transmission line, and use the ADMM algorithm to decompose; Sub-problem solving: Each sub-problem is solved independently locally to calculate the corresponding charging power, discharging power and transmission power; Global coordination: The global coordinator aggregates the solution results of each sub-problem, updates global variables, and ensures the satisfaction of global energy balance and constraints; Iterative update: In each round of iteration, the subproblems and the global coordinator exchange information, update their respective optimization variables, and gradually converge to the global optimal solution.

9. The distributed energy storage network resource control decision-making method according to claim 8 is characterized in that: The dynamic energy routing model construction in S34 specifically includes: S341, based on the output results of the energy demand prediction and dynamic optimization model, including predicted energy demand, user response behavior prediction data, and optimization strategies for minimizing energy transmission loss and charging and discharging loss; Routing model construction: Build an energy routing model to define the selection and scheduling strategies of transmission paths, including: Define the nodes and edges in the distributed energy storage network and establish the network topology. The nodes include energy storage equipment, power generation equipment, and load centers, and the edges include energy transmission lines. Calculate the optimal energy transmission path based on the path selection algorithm. Calculate the weight of the transmission path based on the energy demand forecast and dynamic optimization results. The weight includes energy transmission loss and transmission time. S342, dynamically adjust the energy transmission path: Real-time data collection: Through a distributed sensor network, real-time data on the operation status of the energy storage system and the power grid is collected, including voltage, current, power, energy demand, and user response; Path adjustment algorithm: Based on real-time collected data and energy demand forecast results, the path adjustment algorithm is used to dynamically adjust the energy transmission path; S343. Optimize the charging and discharging strategy of the energy storage system: Based on energy demand prediction and user response behavior model, the charging and discharging strategy of the energy storage system is optimized, including: Use real-time data to calibrate energy demand forecast results to improve forecast accuracy; dynamically adjust the charging and discharging time and power of the energy storage system according to the charging and discharging strategy output by the optimization model to minimize energy loss and economic costs; optimize the charging and discharging behavior of user-side energy storage equipment according to the prediction results of the user response model; S344, output optimized energy transmission path and energy storage control strategy: Output the optimized energy transmission path to ensure that energy can be transmitted to the demand node to meet the load demand; publish the optimized energy storage control strategy to each energy storage device and monitor its execution in real time.

10. The distributed energy storage network resource control decision-making method according to claim 9, characterized in that: The network topology is expressed as: , where V represents the node set and E represents the edge set; The path selection algorithm uses the shortest path algorithm to calculate the optimal energy transfer path: ,in, It is the edge The weight of is the path selection variable; Path weight calculation: ,in, and is the weight coefficient, is the energy transmission loss power, is the energy transfer time; The path adjustment algorithm is based on real-time path updates and is expressed as the formula: ,in, is the path weight at time t, and is the transmission loss power and transmission time at time t; Adaptive Adjustment: ,in, is the adjustment step size, is the transmission loss power at time t+1, is the actual demand at time t, is the predicted energy demand at time t.

Images