Microgrid energy-saving scheme generation method and system based on energy storage optimization scheduling

Abstract

The present invention provides a method and system for generating a microgrid energy-saving scheme based on energy storage optimization scheduling, which relates to the technical field of energy storage optimization scheduling, including collecting historical operation data, real-time operation data and external environment data of a microgrid system, and performing multi-scenario prediction through a Monte Carlo simulation method to generate multi-scenario prediction results; based on the multi-scenario prediction results, and using fuzzy decision theory and hierarchical analysis method, selecting an optimal microgrid energy-saving scheme from the solution set; converting the optimal microgrid energy-saving scheme into an intelligent energy-saving operation instruction set, and using blockchain to record the microgrid operation status, energy storage scheduling results and energy transaction information, calculating comprehensive performance indicators based on the data recorded in the blockchain, using a federated learning method to perform distributed evaluation based on the comprehensive performance indicators, and continuously optimizing the microgrid energy-saving scheme according to the evaluation results of the comprehensive performance indicators.

Description

Technical Field

[0001] The present invention relates to energy storage optimization scheduling technology, and in particular to a method and system for generating a microgrid energy-saving solution based on energy storage optimization scheduling. Background Art

[0002] As an effective combination of distributed energy and power grids, microgrids can improve the utilization efficiency of renewable energy and enhance the flexibility and reliability of power grids. However, the intermittent and volatile characteristics of renewable energy in microgrids, as well as the randomness of load demand, bring challenges to the optimized operation and energy-saving scheduling of microgrids. In the prior art, microgrids usually adopt prediction-based optimization scheduling strategies, but the prediction error and model parameter uncertainty affect the optimization effect. In addition, the existing microgrid energy-saving optimization mainly relies on centralized algorithms and lacks distributed collaborative optimization mechanisms. It is difficult to fully tap the computing resources of edge devices and cannot effectively cope with the dynamic changes and uncertainty factors of microgrid operation. Therefore, there is an urgent need for a microgrid energy-saving optimization method that can comprehensively consider prediction uncertainty, distributed optimization, and adaptive dynamic adjustment. Summary of the invention

[0003] The embodiments of the present invention provide a method and system for generating a microgrid energy-saving solution based on energy storage optimization scheduling, which can solve the problems in the prior art.

[0004] According to a first aspect of the embodiments of the present invention,

[0005] A method for generating a microgrid energy-saving solution based on energy storage optimization scheduling is provided, comprising:

[0006] Collect historical operation data, real-time operation data and external environment data of the microgrid system, pre-process and extract features of the collected historical operation data, real-time operation data and external environment data to form a multi-scale spatiotemporal feature representation, and build a joint prediction model for future load demand, renewable energy power generation and electricity market price based on the multi-scale spatiotemporal feature representation, and generate multi-scenario prediction results by performing multi-scenario prediction through Monte Carlo simulation method;

[0007] According to the multi-scenario prediction results, a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy is constructed. The stochastic dynamic optimization model is solved by an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set containing different weight combinations. Fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set.

[0008] The optimal microgrid energy-saving scheme is converted into an intelligent energy-saving operation instruction set, which includes energy storage device control instructions, renewable energy power generation regulation instructions, load management strategies and user behavior suggestions. Distributed transmission and collaborative control are performed through edge computing devices. At the same time, blockchain is used to record the microgrid operating status, energy storage scheduling results and energy trading information. Comprehensive performance indicators are calculated based on the data recorded on the blockchain. Federated learning methods are used for distributed evaluation based on the comprehensive performance indicators. The microgrid energy-saving scheme is continuously optimized based on the evaluation results of the comprehensive performance indicators.

[0009] In an optional embodiment,

[0010] Based on the multi-scale spatiotemporal feature representation, a joint prediction model for future load demand, renewable energy generation and electricity market price is constructed, and a multi-scenario prediction is performed through the Monte Carlo simulation method to generate multi-scenario prediction results including:

[0011] Based on the obtained multi-scale spatiotemporal characteristics, the wavelet transform method is used to extract the time-frequency domain characteristics of load demand data, renewable energy power generation data and electricity market price data to obtain time-frequency domain feature data, and the graph convolutional neural network is used to extract the spatial dependency characteristics of the microgrid topology structure to obtain spatial dependency feature data. At the same time, the attention mechanism is introduced to dynamically adjust the importance weights of the time-frequency domain feature data and the spatial dependency feature data to obtain a spatiotemporal feature data set;

[0012] Construct a spatiotemporal dynamic graph neural network model, represent the microgrid system as a dynamic graph structure that changes over time, wherein nodes represent power equipment and load points in the microgrid, edges represent the interactive relationship between power equipment and load points, and input the spatiotemporal feature data set into the spatiotemporal dynamic graph neural network model, and establish the spatiotemporal dependency relationship between nodes through the message passing and aggregation operations of the graph neural network;

[0013] Taking historical data of load demand data, renewable energy generation data and electricity market price data as training samples, the spatiotemporal dynamic graph neural network model is trained by a supervised learning method, and the parameters of the spatiotemporal dynamic graph neural network model are optimized by minimizing the loss function of multi-time step prediction errors to obtain a trained spatiotemporal dynamic graph neural network model;

[0014] A probability model of uncertain factors is constructed based on electricity market price data and meteorological condition data, and multiple future scenarios are generated through Monte Carlo random sampling. The electricity market price data and meteorological condition data under each future scenario are input into the trained spatiotemporal dynamic graph neural network model to predict the future load demand, renewable energy power generation and electricity price of the microgrid system under each future scenario, and obtain multi-scenario prediction results.

[0015] In an optional embodiment,

[0016] According to the multi-scenario prediction results, a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy is constructed. The stochastic dynamic optimization model is solved by an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set containing different weight combinations. Fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set, including:

[0017] A random dynamic optimization model is constructed based on the multi-scenario prediction results. The random dynamic optimization model takes operating cost, renewable energy utilization rate and load demand satisfaction as optimization targets. A comprehensive objective function is formed by introducing weight coefficients to weightedly combine multiple optimization targets. Decision variables include energy storage device charging and discharging power, renewable energy output power, demand response load adjustment amount and grid interaction power. Constraints include energy storage device capacity limit, charging and discharging power limit, renewable energy output upper limit, load demand satisfaction constraint and grid interaction power upper and lower limits. Random variables are introduced to represent the uncertainty of load demand, renewable energy power generation and electricity price. A scenario-based method is used to discretize random variables and transform the random dynamic optimization model into a deterministic optimization problem.

[0018] An intelligent optimization algorithm with an adaptive parameter adjustment mechanism is used to solve the deterministic optimization problem, initialize the algorithm parameters and randomly generate an initial population. Each individual corresponds to a microgrid optimization scheme. Each individual in the current population is decoded, its objective function value and constraint violation are calculated, the fitness value is determined, the current population is sorted according to the fitness value, and the current optimal solution is recorded.

[0019] Perform selection, crossover and mutation operations on the sorted population to generate a new child population, merge the child population with the parent population, calculate the fitness value of the merged population, re-sort according to the fitness value, select the individual with the highest fitness as the new parent population, repeat the iteration until the preset termination condition is met, output the current optimal solution as the microgrid optimization scheduling plan, and at the same time, divide the latest parent population into multiple child populations and assign them to different computing nodes for parallel computing. Perform selection, crossover and mutation operations on the child population on each computing node, and regularly migrate populations and exchange information between computing nodes, and finally obtain a solution set of microgrid optimization scheduling plans under different weight combinations;

[0020] A fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The analytic hierarchy process is used to determine the weight of each evaluation index. The membership degree of each scheme in the microgrid optimization scheduling scheme solution set under each evaluation index is calculated, and the comprehensive evaluation value of each scheme is calculated based on the index weight. The schemes in the microgrid optimization scheduling scheme solution set are ranked according to the comprehensive evaluation value. Combined with the knowledge of domain experts and decision-making preferences, the microgrid optimization scheduling scheme with the highest comprehensive evaluation value is selected from the sorting results as the optimal microgrid energy-saving scheduling scheme.

[0021] In an optional embodiment,

[0022] A fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The analytic hierarchy process is used to determine the weight of each evaluation index, and the membership degree of each scheme under each evaluation index in the microgrid optimization scheduling solution is calculated. The comprehensive evaluation value of each scheme is calculated in combination with the index weight, including:

[0023] According to the characteristics of each evaluation index and expert knowledge, the cloud model parameters corresponding to each evaluation index are determined, and the cloud model parameters include expected value, entropy and super entropy, wherein the expected value represents the central position of the membership, the entropy represents the discrete degree of the membership, and the super entropy represents the randomness of the membership;

[0024] Using the forward cloud generator of the cloud model, a group of cloud droplets are generated according to the determined cloud model parameters, and each cloud droplet represents the membership value of the evaluation index at different membership levels;

[0025] According to the performance value of each microgrid optimization scheduling scheme under each evaluation index, the reverse cloud generator of the cloud model is used to calculate the probability that the microgrid optimization scheduling scheme belongs to different membership levels, and the membership vector of the current microgrid optimization scheduling scheme is obtained;

[0026] The membership vector of the current microgrid optimization scheduling scheme is weighted and aggregated with the evaluation index weight to obtain the comprehensive membership of the current microgrid optimization scheduling scheme, and the comprehensive membership is used as the comprehensive evaluation value of the current microgrid optimization scheduling scheme.

[0027] In an optional embodiment,

[0028] The membership vector of the current microgrid optimization scheduling scheme is weighted and aggregated with the evaluation index weight, and the calculation formula for the comprehensive membership of the current microgrid optimization scheduling scheme is as follows:

[0029]

[0030] Among them, S represents the comprehensive membership, n represents the number of evaluation indicators, and w i represents the evaluation index weight, xi represents the specific performance value of the xth microgrid optimization scheduling scheme under the ith evaluation index, E x Indicates the central position of membership, E n Indicates the discrete degree of membership, H e represents the super entropy, which is a measure of the randomness of the membership, α i Represents the dynamic adjustment factor.

[0031] In an optional embodiment,

[0032] Blockchain is used to record the operating status of microgrids, energy storage dispatch results, and energy transaction information. Comprehensive performance indicators are calculated based on the data recorded on blockchain. Federated learning methods are used for distributed evaluation based on the comprehensive performance indicators. The microgrid energy-saving solution is continuously optimized based on the evaluation results of the comprehensive performance indicators, including:

[0033] Construct a microgrid data blockchain. By designing the block data structure and consensus mechanism, the microgrid operation status, energy storage dispatch results and energy transaction information are recorded, and a data access interface and permission control mechanism are provided. Based on the microgrid operation data recorded in the blockchain, a microgrid comprehensive performance indicator system is constructed, and the indicator calculation tasks are decomposed to each edge node.

[0034] Each edge node regularly calculates the comprehensive performance index based on the microgrid operation data recorded in the blockchain according to the index calculation standard, generates the index hash value, and submits the index hash value to the blockchain network for recording and verification. After the verification is passed, the blockchain network aggregates the comprehensive performance index data submitted by each node;

[0035] Using the comprehensive performance indicator data of each edge node recorded in the blockchain, a federated learning model training sample is constructed, and a distributed training of the microgrid comprehensive performance evaluation model is performed using a federated learning algorithm. By aggregating and updating encrypted model parameters, the trained global evaluation model parameters are obtained;

[0036] Based on the trained global evaluation model parameters, the comprehensive performance index data calculated by each edge node is evaluated, and the quantitative evaluation results of the microgrid performance are output. According to the quantitative evaluation results, the global energy-saving optimization strategy is generated through the intelligent optimization algorithm, and the generated global energy-saving optimization strategy is distributed to each edge node through the blockchain. Each edge node generates local control instructions based on the received global energy-saving optimization strategy and the local microgrid operation status, and sends the control instructions to the corresponding device end through the blockchain to guide the device to adjust the operation status, and upload the device execution results to the blockchain;

[0037] According to the device execution results received by the blockchain, each edge node is re-triggered to calculate the comprehensive performance indicators of the microgrid and generate new evaluation results. When the evaluation results are lower than the preset threshold, the energy-saving strategy optimization process is automatically triggered, and the data generated by the strategy optimization process is recorded on the chain to achieve continuous optimization of the microgrid energy-saving plan.

[0038] In an optional embodiment,

[0039] Based on the quantitative evaluation results, the global energy-saving optimization strategy generated by the intelligent optimization algorithm includes:

[0040] Define the state space and action space of the microgrid. The state space includes the charging and discharging status of the energy storage device, the real-time output of renewable energy, load demand and electricity price fluctuations. The action space includes adjusting the charging and discharging of the energy storage system, the interactive power between the grid and the microgrid, and the optimization of load demand response.

[0041] Design a multidimensional reward function, wherein the multidimensional reward function includes an operating cost value, a renewable energy utilization value, a load response degree value, a dynamic response speed value, and a stability value, and establish a deep reinforcement learning model based on the multidimensional reward function, wherein the intelligent agent in the deep reinforcement learning continuously interacts with the environmental state of the microgrid, and the energy-saving optimization strategy of the intelligent agent is updated based on the instant reward value obtained from each interaction;

[0042] The microgrid is modeled as a multi-agent node network, and the microgrid is abstracted as a network topology composed of multiple intelligent nodes. Each of the intelligent nodes is responsible for managing the local state of its corresponding microgrid. By establishing a communication protocol between the intelligent nodes, the intelligent nodes share their local state information with each other, and the shared local state information is integrated into the global state information. At the same time, the obtained quantitative evaluation results and the global state information are introduced into the deep reinforcement learning model, the energy-saving optimization strategy is iteratively optimized, and finally a global energy-saving optimization strategy is generated.

[0043] According to a second aspect of the embodiments of the present invention,

[0044] A microgrid energy-saving solution generation system based on energy storage optimization scheduling is provided, comprising:

[0045] The first unit is used to collect historical operation data, real-time operation data and external environment data of the microgrid system, pre-process and extract features of the collected historical operation data, real-time operation data and external environment data to form a multi-scale spatiotemporal feature representation, and build a joint prediction model of future load demand, renewable energy power generation and electricity market price based on the multi-scale spatiotemporal feature representation, and generate multi-scenario prediction results by performing multi-scenario prediction through Monte Carlo simulation method;

[0046] The second unit is used to construct a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy based on the multi-scenario prediction results, and solve the stochastic dynamic optimization model by using an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set including different weight combinations, and fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set.

[0047] The third unit is used to convert the optimal microgrid energy-saving plan into an intelligent energy-saving operation instruction set, which includes energy storage device control instructions, renewable energy power generation regulation instructions, load management strategies and user behavior suggestions, and performs distributed transmission and collaborative control through edge computing devices. At the same time, blockchain is used to record the microgrid operation status, energy storage scheduling results and energy trading information, and comprehensive performance indicators are calculated based on the data recorded on the blockchain. A federated learning method is used for distributed evaluation based on the comprehensive performance indicators, and the microgrid energy-saving plan is continuously optimized based on the evaluation results of the comprehensive performance indicators.

[0048] According to a third aspect of the embodiments of the present invention,

[0049] An electronic device is provided, comprising:

[0050] processor;

[0051] a memory for storing processor-executable instructions;

[0052] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0053] A fourth aspect of the embodiments of the present invention is:

[0054] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0055] In this embodiment, by constructing a joint prediction model and introducing Monte Carlo simulation, the uncertainty of microgrid operation can be effectively handled, and the robustness and reliability of energy-saving optimization can be improved; the introduction of random dynamic optimization models and adaptive parameter adjustment mechanisms can dynamically adapt to changes in microgrid operating conditions, and enhance the adaptability and flexibility of energy-saving optimization; the method of combining parallel computing and fuzzy decision theory can efficiently solve large-scale optimization problems, balance optimization performance and computing efficiency, and improve the real-time performance of energy-saving optimization; edge computing and blockchain technology are used to achieve distributed intelligent control and secure and reliable data management, and support decentralized collaborative optimization of microgrids; integrated federated learning realizes distributed evaluation, sustainably optimizes energy-saving solutions, and improves the energy-saving performance of microgrids throughout their life cycle; the generated intelligent energy-saving operation instruction set covers multiple levels of energy storage, renewable energy, and load, and can provide a global collaborative energy-saving optimization solution to improve the comprehensive energy-saving benefits of microgrids. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of a flow chart of a method for generating a microgrid energy-saving solution based on energy storage optimization scheduling according to an embodiment of the present invention;

[0057] Figure 2 It is a structural schematic diagram of a microgrid energy-saving solution generation system based on energy storage optimization scheduling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0060] Figure 1 FIG. 1 is a flow chart of a method for generating a microgrid energy-saving solution based on energy storage optimization scheduling according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0061] S101. Collect historical operation data, real-time operation data and external environment data of the microgrid system, pre-process and extract features of the collected historical operation data, real-time operation data and external environment data to form a multi-scale spatiotemporal feature representation, and based on the multi-scale spatiotemporal feature representation, construct a joint prediction model for future load demand, renewable energy power generation and electricity market price, and perform multi-scenario prediction through Monte Carlo simulation method to generate multi-scenario prediction results.

[0062] Among them, historical operation data refers to the operation record of the microgrid in the past period of time, including load demand, charging and discharging of the energy storage system, and the output of distributed energy. This type of data can reflect the long-term operation rules and trends of the microgrid. Real-time operation data refers to the actual operation status of the system at the current moment, including the current load demand, power generation, energy storage, etc. These data can provide a basis for immediate optimization decisions. External environmental data covers external factors that affect the operation of the microgrid, such as weather conditions, temperature, humidity, wind speed, solar radiation intensity, etc. These environmental factors have an important impact on the power generation of renewable energy (such as wind and solar energy). At the same time, electricity market price data is also part of the external environmental data, which will affect the economic operation decisions of the microgrid.

[0063] In the feature extraction stage, the raw data needs to be converted into a representative multi-scale spatiotemporal feature representation. Multi-scale spatiotemporal feature representation is a way to analyze data from the perspective of time and space. Specifically, feature extraction on the time scale can capture the laws of different time dimensions by dividing the data into multiple time scales such as hourly, daily, and weekly levels; features on the spatial scale refer to the associated features extracted from data of multiple microgrid nodes or different geographical locations, which are used to reflect the synergy between nodes and regional differences.

[0064] In an optional embodiment,

[0065] Based on the multi-scale spatiotemporal feature representation, a joint prediction model for future load demand, renewable energy generation and electricity market price is constructed, and a multi-scenario prediction is performed through the Monte Carlo simulation method to generate multi-scenario prediction results including:

[0066] Based on the obtained multi-scale spatiotemporal characteristics, the wavelet transform method is used to extract the time-frequency domain characteristics of load demand data, renewable energy power generation data and electricity market price data to obtain time-frequency domain feature data, and the graph convolutional neural network is used to extract the spatial dependency characteristics of the microgrid topology structure to obtain spatial dependency feature data. At the same time, the attention mechanism is introduced to dynamically adjust the importance weights of the time-frequency domain feature data and the spatial dependency feature data to obtain a spatiotemporal feature data set;

[0067] Construct a spatiotemporal dynamic graph neural network model, represent the microgrid system as a dynamic graph structure that changes over time, wherein nodes represent power equipment and load points in the microgrid, edges represent the interactive relationship between power equipment and load points, and input the spatiotemporal feature data set into the spatiotemporal dynamic graph neural network model, and establish the spatiotemporal dependency relationship between nodes through the message passing and aggregation operations of the graph neural network;

[0068] Taking historical data of load demand data, renewable energy generation data and electricity market price data as training samples, the spatiotemporal dynamic graph neural network model is trained by a supervised learning method, and the parameters of the spatiotemporal dynamic graph neural network model are optimized by minimizing the loss function of multi-time step prediction errors to obtain a trained spatiotemporal dynamic graph neural network model;

[0069] A probability model of uncertain factors is constructed based on electricity market price data and meteorological condition data, and multiple future scenarios are generated through Monte Carlo random sampling. The electricity market price data and meteorological condition data under each future scenario are input into the trained spatiotemporal dynamic graph neural network model to predict the future load demand, renewable energy power generation and electricity price of the microgrid system under each future scenario, and obtain multi-scenario prediction results.

[0070] For example, first, it is necessary to extract multi-scale spatiotemporal features of load demand data, renewable energy generation data, and electricity market price data. Through wavelet transform, these time series data can be decomposed to obtain features at different frequencies and time scales. Wavelet transform is a signal processing method that can decompose time series into time domain and frequency domain at the same time to obtain time-frequency domain features. Through this decomposition process, multi-scale time-frequency information of load fluctuations, renewable energy output changes, and electricity market price fluctuations in microgrids can be extracted. This information can help the prediction model capture future trends more accurately.

[0071] At the same time, it is also necessary to extract the spatial dependency features in the microgrid topology. Spatial dependency features refer to the mutual influence and collaborative relationship between power equipment and load points in a microgrid. Since there are geographical and electrical connections between nodes in a microgrid (such as energy storage equipment, power generation equipment, and load points), the complex relationship between these nodes can be effectively captured through graph convolutional neural networks. Graph convolutional neural networks can process the interdependence between microgrid devices through graph structure data modeling, transmit and aggregate the status information of neighboring nodes, and thus extract spatial dependency features.

[0072] In order to further enhance the prediction ability of the model, an attention mechanism is also introduced. The attention mechanism can automatically adjust the importance weights of time-frequency domain features and spatial dependency features in different situations according to the dynamic changes of the data. In this way, the model can automatically pay attention to the most important feature information at each time step, so as to better adapt to changes in dynamic environments and improve prediction accuracy.

[0073] Next, we need to build a spatiotemporal dynamic graph neural network model. The spatiotemporal dynamic graph neural network is a deep learning model that can capture temporal and spatial dependencies at the same time. In this model, the microgrid system is represented as a dynamic graph structure. The nodes in the graph represent the power equipment and load points in the microgrid, and the edges represent the physical or logical connections between nodes, such as the transmission lines or the interaction between loads and equipment. Through the message passing mechanism of the spatiotemporal graph neural network, nodes can continuously exchange information with neighboring nodes, thereby building the spatiotemporal dependencies between power equipment and load points in the microgrid. This graph structure changes over time, dynamically updating the status information of nodes and edges, enabling the model to handle complex spatiotemporal dependencies.

[0074] The model is trained based on historical data of load demand, renewable energy generation, and electricity market prices using supervised learning methods. Supervised learning is a method of training a model by giving input data and its corresponding correct output (i.e., label). In this context, historical data is used as input, and the goal is to predict future data for multiple time steps. By minimizing the loss function of multi-time step prediction errors, the parameters of the model can be gradually optimized, and ultimately a trained spatiotemporal dynamic graph neural network model is obtained that can accurately predict future load demand, renewable energy generation, and electricity market prices.

[0075] After completing the model training, considering the high uncertainty of electricity market prices and meteorological conditions, the Monte Carlo simulation method was introduced for multi-scenario forecasting. Monte Carlo simulation is a forecasting method that simulates the possible state of the system under different scenarios through a large number of random sampling. The probability model of uncertain factors constructed based on electricity market price data and meteorological condition data can generate multiple possible future scenarios. These scenarios reflect the various future development paths of the microgrid system under uncertain conditions. Through random sampling, multiple combinations of future electricity market prices and meteorological conditions can be generated. Then, the data of these future scenarios are input into the trained spatiotemporal dynamic graph neural network model to predict the load demand, renewable energy generation and electricity price under each scenario. The final multi-scenario forecast results can provide multi-dimensional support for the future decision-making and scheduling of the microgrid system.

[0076] In this embodiment, the time-frequency domain features are extracted by wavelet transform and the spatial dependency features are captured by graph convolutional neural network, which can accurately describe the spatiotemporal dynamic relationship between nodes in the microgrid system. By introducing the attention mechanism, the model can dynamically adjust the importance weights of spatiotemporal features and enhance its ability to cope with complex environmental changes. Combined with the Monte Carlo simulation method, the model can generate multiple future scenarios to help the microgrid system make refined decisions and dispatch under uncertain conditions. The model can maintain high prediction accuracy in the face of environmental changes, meteorological uncertainties and market price fluctuations, and provide a variety of alternatives for decision-making.

[0077] S102. Based on the multi-scenario prediction results, a random dynamic optimization model including energy storage equipment charging and discharging strategies, renewable energy utilization, demand response and grid interaction strategies is constructed. The random dynamic optimization model is solved by an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set containing different weight combinations. Fuzzy decision theory and hierarchical analysis method are used to screen out the optimal microgrid energy-saving solution from the solution set.

[0078] Among them, multi-scenario forecasting refers to considering multiple possible scenarios in an uncertain future environment to improve the model's adaptability to future changes. The model includes the charging and discharging strategy of the energy storage device, which means that it is necessary to decide when to charge and when to discharge to make the most efficient use of the energy storage system. The energy storage device can be a battery, a superconducting magnetic energy storage system, or other types of energy storage devices. The timing of charging and discharging strategies directly affect the stability and economy of the power grid.

[0079] The grid interaction strategy refers to how to interact with the external grid to achieve goals such as power exchange and load balancing between grids. An effective grid interaction strategy can improve the flexibility and stability of the microgrid and help cope with the volatility of the external grid.

[0080] To solve this model, an optimization algorithm that introduces an adaptive parameter adjustment mechanism is used. This algorithm has the ability to adaptively adjust parameters, allowing the optimization process to find the optimal solution under different conditions. The optimization process of the algorithm is performed under a parallel computing framework, which means that multiple calculations can be performed simultaneously to speed up the solution process and obtain a solution set under different weight combinations.

[0081] After obtaining the solution set, fuzzy decision theory and hierarchical analysis method are used to screen out the optimal microgrid energy-saving solution. Fuzzy decision theory allows decision-making under conditions of uncertainty and ambiguity, and is suitable for dealing with situations where standards are unclear or data is incomplete. Hierarchical analysis method evaluates the pros and cons of each solution by establishing a hierarchical relationship between decision factors, thereby selecting the solution that best meets the needs.

[0082] In an optional embodiment,

[0083] According to the multi-scenario prediction results, a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy is constructed. The stochastic dynamic optimization model is solved by an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set containing different weight combinations. Fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set, including:

[0084] A random dynamic optimization model is constructed based on the multi-scenario prediction results. The random dynamic optimization model takes operating cost, renewable energy utilization rate and load demand satisfaction as optimization targets. A comprehensive objective function is formed by introducing weight coefficients to weightedly combine multiple optimization targets. Decision variables include energy storage device charging and discharging power, renewable energy output power, demand response load adjustment amount and grid interaction power. Constraints include energy storage device capacity limit, charging and discharging power limit, renewable energy output upper limit, load demand satisfaction constraint and grid interaction power upper and lower limits. Random variables are introduced to represent the uncertainty of load demand, renewable energy power generation and electricity price. A scenario-based method is used to discretize random variables and transform the random dynamic optimization model into a deterministic optimization problem.

[0085] An intelligent optimization algorithm with an adaptive parameter adjustment mechanism is used to solve the deterministic optimization problem, initialize the algorithm parameters and randomly generate an initial population. Each individual corresponds to a microgrid optimization scheme. Each individual in the current population is decoded, its objective function value and constraint violation are calculated, the fitness value is determined, the current population is sorted according to the fitness value, and the current optimal solution is recorded.

[0086] Perform selection, crossover and mutation operations on the sorted population to generate a new child population, merge the child population with the parent population, calculate the fitness value of the merged population, re-sort according to the fitness value, select the individual with the highest fitness as the new parent population, repeat the iteration until the preset termination condition is met, output the current optimal solution as the microgrid optimization scheduling plan, and at the same time, divide the latest parent population into multiple child populations and assign them to different computing nodes for parallel computing. Perform selection, crossover and mutation operations on the child population on each computing node, and regularly migrate populations and exchange information between computing nodes, and finally obtain a solution set of microgrid optimization scheduling plans under different weight combinations;

[0087] A fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The analytic hierarchy process is used to determine the weight of each evaluation index. The membership degree of each scheme in the microgrid optimization scheduling scheme solution set under each evaluation index is calculated, and the comprehensive evaluation value of each scheme is calculated based on the index weight. The schemes in the microgrid optimization scheduling scheme solution set are ranked according to the comprehensive evaluation value. Combined with the knowledge of domain experts and decision-making preferences, the microgrid optimization scheduling scheme with the highest comprehensive evaluation value is selected from the sorting results as the optimal microgrid energy-saving scheduling scheme.

[0088] For example, firstly, according to the historical operation data, real-time operation data and external environment data of the microgrid, data preprocessing and feature extraction are carried out through time series analysis, feature engineering and other technical means to construct a multi-scale spatiotemporal feature representation, and fully explore the regular information contained in the microgrid operation data. On this basis, the joint prediction model of load demand, renewable energy generation and electricity market price is constructed by integrating technologies such as deep learning and probabilistic graphical models. This prediction model can not only capture the correlation between the three, but also quantify the uncertainty distribution characteristics of each variable. Considering the limitations of the prediction results of a single scenario, the Monte Carlo simulation method is introduced to sample a large number of random variables in the prediction model to generate multiple possible future scenarios. Each scenario corresponds to a set of predicted values ​​of load demand, renewable energy generation and electricity price, forming a multi-scenario prediction result set, which provides more comprehensive and reliable input data for subsequent optimization decisions.

[0089] After obtaining the multi-scenario prediction results, a random dynamic optimization model is constructed based on this. The optimization model aims to minimize the operating cost of the microgrid, maximize the utilization rate of renewable energy, and maximize the satisfaction of load demand. By introducing weight coefficients, multiple optimization objectives are weighted and combined to form a comprehensive objective function. The decision variables include the charging and discharging power of energy storage equipment, the output power of renewable energy, the demand response load adjustment amount, and the grid interaction power. The variables are coupled through equations such as power balance constraints and energy storage capacity constraints. At the same time, considering the uncertainty factors in the actual operation of the microgrid, random variables are introduced to represent the fluctuation characteristics of load demand, renewable energy generation, and electricity prices. In order to facilitate the solution, the scenario method is used to discretize the random variables, convert each continuous random variable into a finite number of discrete scenarios, and assign the corresponding probability of occurrence to each scenario, thereby converting the original random dynamic optimization model into a deterministic mixed integer programming problem.

[0090] For the above large-scale mixed integer programming problem, traditional exact algorithms are difficult to obtain the optimal solution within an acceptable time, so intelligent optimization algorithms are used to solve it. First, according to the characteristics of the problem, a suitable optimization algorithm is selected, such as heuristic search methods such as genetic algorithm and particle swarm optimization algorithm. Then, a targeted coding strategy is designed to map the microgrid optimization scheduling scheme to individuals or particles in the algorithm. Each individual corresponds to a set of energy storage equipment charging and discharging power, renewable energy output power, demand response load regulation, and grid interaction power. In the initialization stage of the algorithm, the initial solution is randomly generated according to the preset population size, and each individual is decoded to calculate its comprehensive objective function value and constraint violation under multiple scenario conditions, thereby determining the individual's fitness value. In the iterative optimization process, new solutions are generated through genetic operations such as selection, crossover, and mutation, or particle velocity and position updates, and the population is selected according to the fitness value to continuously approach the global optimum. Considering the complexity of the optimization problem and the impact of the algorithm parameter settings on the results, an adaptive parameter adjustment mechanism is introduced to dynamically adjust the algorithm parameters, such as crossover probability and mutation probability, according to the convergence speed of the algorithm and the quality of the solution, in order to balance the algorithm's exploration and mining capabilities. In addition, in order to further improve the solution efficiency and result quality, a parallel computing framework is used to divide the population into multiple sub-populations and assign them to different computing nodes. A relatively independent optimization process is performed in each node, and information exchange and collaboration are carried out between nodes through population migration and other methods. Finally, a solution set of microgrid optimization scheduling solutions under different weight combinations is obtained.

[0091] In order to select the optimal energy-saving scheduling scheme from the set of alternative solutions, a fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The operating cost index measures the total cost expenditure of the microgrid during the optimization period, including the depreciation cost of energy storage equipment, the cost of renewable energy generation, load demand response compensation and the power purchase fee of the power grid; the renewable energy utilization rate index measures the proportion of actual renewable energy generation to the theoretical maximum generation, reflecting the microgrid's level of renewable energy consumption; the load demand satisfaction index measures the satisfaction degree of load demand under the optimization scheme, which can be calculated by comparing the deviation between the optimization result and the actual demand. After determining the evaluation index, the hierarchical analysis method is used to determine the importance weight of each index. The judgment matrix is ​​constructed by pairwise comparison, the maximum eigenvalue and the corresponding eigenvector of the matrix are calculated, and the consistency test is performed to finally obtain the index weight vector. Then, the membership function is used to calculate the satisfaction degree of each alternative scheme under different evaluation indicators, and the weighted average method is used to obtain the comprehensive evaluation value of each scheme in combination with the index weight. Finally, the schemes are sorted according to the comprehensive evaluation value and submitted to domain experts for human-computer interactive decision-making to select the optimal energy-saving scheduling scheme.

[0092] Taking a residential microgrid as an example, the microgrid consists of 20 residential users, a 10kW photovoltaic power generation system, a 15kW wind turbine, a group of 5kWh energy storage batteries and a 10kW grid connection point. Based on historical data such as residential electricity consumption data and environmental monitoring data, a joint prediction model of power load, photovoltaic power generation, wind power generation and grid electricity price is constructed. The model is used to predict multi-scenario data for the next 24 hours. Each scenario contains 24 time steps of load demand, photovoltaic power generation, wind power generation and electricity price data. With this as input, a microgrid optimization scheduling model is constructed, with the goal of minimizing the total operating cost for 24 hours. The decision variables include energy storage charging and discharging power, photovoltaic output power, wind power output power, grid interaction power and load regulation in each period. The genetic algorithm is used to solve the optimization model. The individual code is a 120-dimensional vector, and each dimension corresponds to the value of a decision variable. Through iterative optimization through selection, crossover, mutation and other operations, and parallel execution in 4 computing nodes, a set of microgrid optimization scheduling alternatives are finally obtained. Then, the alternative plans are evaluated comprehensively from the perspectives of operating cost, renewable energy utilization rate and load satisfaction, and the comprehensive score of each plan is calculated. Finally, based on the comprehensive score ranking results, discussions are held with grid dispatchers, user representatives, etc. to weigh the advantages and disadvantages of different plans, select the best microgrid energy-saving optimization plan, and form a microgrid equipment operation plan for the next 24 hours.

[0093] In this embodiment, by introducing a joint probability prediction model and multi-scenario analysis, the challenges brought by the volatility of renewable energy and the uncertainty of load demand are effectively addressed, and the robustness of the optimization results is improved; an adaptive intelligent optimization algorithm is used to improve the global optimization capability while meeting real-time requirements, and a higher-quality optimization solution can be obtained; parallel computing and distributed optimization technology are used to fully utilize heterogeneous computing resources and accelerate the solution of large-scale optimization problems; by constructing a multi-objective fuzzy evaluation system, a human-computer interactive decision-making mechanism is introduced, and multiple optimization objectives are balanced, so that the solution results are closer to actual needs; the optimization plan covers multiple levels such as energy storage scheduling, demand response, and grid interaction, and can achieve better energy-saving effects than a single strategy.

[0094] In an optional embodiment,

[0095] A fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The analytic hierarchy process is used to determine the weight of each evaluation index, and the membership degree of each scheme under each evaluation index in the microgrid optimization scheduling solution is calculated. The comprehensive evaluation value of each scheme is calculated in combination with the index weight, including:

[0096] According to the characteristics of each evaluation index and expert knowledge, the cloud model parameters corresponding to each evaluation index are determined, and the cloud model parameters include expected value, entropy and super entropy, wherein the expected value represents the central position of the membership, the entropy represents the discrete degree of the membership, and the super entropy represents the randomness of the membership;

[0097] Using the forward cloud generator of the cloud model, a group of cloud droplets are generated according to the determined cloud model parameters, and each cloud droplet represents the membership value of the evaluation index at different membership levels;

[0098] According to the performance value of each microgrid optimization scheduling scheme under each evaluation index, the reverse cloud generator of the cloud model is used to calculate the probability that the microgrid optimization scheduling scheme belongs to different membership levels, and the membership vector of the current microgrid optimization scheduling scheme is obtained;

[0099] The membership vector of the current microgrid optimization scheduling scheme is weighted and aggregated with the evaluation index weight to obtain the comprehensive membership of the current microgrid optimization scheduling scheme, and the comprehensive membership is used as the comprehensive evaluation value of the current microgrid optimization scheduling scheme.

[0100] Exemplarily, firstly, the corresponding cloud model parameters are determined according to the physical meaning, value range and domain expert knowledge of each evaluation index. Taking the operating cost index as an example, it can be divided into five levels: "low", "lower", "medium", "higher" and "high", and the power grid operation experts are asked to give the cost interval and membership distribution characteristics corresponding to each level. Then, through statistical analysis and trial and error fitting methods, the expected value, entropy and super entropy parameters of the operating cost index at five levels are obtained. Similarly, the renewable energy utilization index can be divided into three levels: "low", "medium" and "high", and the load satisfaction index can be divided into four levels: "poor", "medium", "good" and "excellent", and their cloud model parameters are determined respectively.

[0101] After obtaining the cloud model parameters of the evaluation index, the membership distribution of each index is randomly simulated using the forward cloud generator of the cloud model. Based on the cloud model parameters, the forward cloud generator generates a large number of cloud droplet samples through nonlinear transformation of normal random numbers and uniform random numbers. Each cloud droplet corresponds to a point on the membership distribution curve, which consists of a set of expected values, membership values ​​and uncertainties. By repeatedly running the forward cloud generator, a cloud droplet matrix reflecting the membership characteristics of the evaluation index can be obtained, in which each column represents a membership level and each row represents a random realization. The cloud droplet matrix intuitively presents the probability distribution law of the membership of the evaluation index at different levels.

[0102] For each microgrid optimization scheduling scheme to be evaluated, its membership vector is calculated using the reverse cloud generator of the cloud model according to its actual performance under each evaluation index. Taking the operating cost index as an example, assuming that the operating cost of a scheduling scheme is 100 yuan, this value is substituted into the cloud model of the operating cost index, and the probability of it belonging to the five levels of "low", "lower", "medium", "higher", and "high" is calculated by the reverse cloud generator to obtain a five-dimensional membership vector. This vector quantitatively describes the fuzzy evaluation results of the current scheduling scheme from the perspective of operating cost, reflecting its closeness to each level. Similarly, the membership vector of the scheduling scheme under the two indicators of renewable energy utilization rate and load satisfaction can be calculated. The membership vectors of the three indicators are combined to form the membership matrix of the current scheduling scheme, where each column corresponds to an evaluation index and each row corresponds to a membership level under the index.

[0103] Finally, in order to obtain the comprehensive evaluation value of the scheduling scheme, the fuzzy weighted average method is used to fuse the membership matrix and the evaluation index weight vector. First, the maximum membership value in each evaluation index membership vector is taken as the membership representative value of the index to obtain the simplified membership vector. Then, the index weight vector and the simplified membership vector are used to perform a dot product operation to map the membership results of the three indicators to the same evaluation level system and assign them different degrees of importance. The dot product result is the comprehensive membership of the scheduling scheme. The larger the value, the better the comprehensive evaluation result of the scheduling scheme. Repeat the above membership calculation and weighted aggregation steps for all the schemes in the solution set of the microgrid optimization scheduling scheme to obtain the comprehensive membership value of each scheme, and then the schemes are ranked accordingly.

[0104] In an optional embodiment,

[0105] The membership vector of the current microgrid optimization scheduling scheme is weighted and aggregated with the evaluation index weight, and the calculation formula for the comprehensive membership of the current microgrid optimization scheduling scheme is as follows:

[0106]

[0107] Among them, S represents the comprehensive membership, n represents the number of evaluation indicators, and w i represents the evaluation index weight, x i represents the specific performance value of the xth microgrid optimization scheduling scheme under the ith evaluation index, E x Indicates the central position of membership, E n Indicates the discrete degree of membership, H e represents the super entropy, which is a measure of the randomness of the membership, α i Represents the dynamic adjustment factor.

[0108] In this embodiment, the membership distribution of the evaluation index is described by the three parameters of expected value, entropy and super entropy, which can effectively characterize the uncertainty, discreteness and randomness in the evaluation process, making the evaluation results more reasonable and accurate; the membership cloud droplet matrix of the evaluation index is generated by the forward cloud generator, which intuitively displays the probability distribution characteristics of the membership, facilitating random simulation analysis; the membership vector of the scheduling scheme is calculated by the reverse cloud generator, quantitatively describing the closeness of the scheme at different evaluation levels, and providing a basis for the optimization of the scheme; the representative value of the membership and the weight of the index are comprehensively considered, and the importance difference of the evaluation index is balanced, so that the evaluation results are more comprehensive and objective. It can deeply explore the inherent uncertainty of the evaluation index, make full use of the experience and knowledge of experts, and flexibly integrate qualitative analysis and quantitative calculation, providing a new fuzzy evaluation idea for energy-saving optimization of microgrids.

[0109] S103. Convert the optimal microgrid energy-saving plan into an intelligent energy-saving operation instruction set, which includes energy storage device control instructions, renewable energy power generation regulation instructions, load management strategies and user behavior recommendations, and perform distributed transmission and collaborative control through edge computing devices. At the same time, use blockchain to record the microgrid operating status, energy storage scheduling results and energy trading information, calculate comprehensive performance indicators based on the data recorded on the blockchain, use a federated learning method to perform distributed evaluation based on the comprehensive performance indicators, and continuously optimize the microgrid energy-saving plan based on the evaluation results of the comprehensive performance indicators.

[0110] Among them, the optimal microgrid energy-saving solution needs to be converted into an actually executable intelligent energy-saving operation instruction set. These instruction sets include: Energy storage device control instructions: Control the charging and discharging behavior of energy storage devices to ensure charging when the load is low and discharging when the load is high, thereby optimizing the balance of power supply and demand. Renewable energy generation regulation instructions: By adjusting the output of renewable energy generation equipment, maximize the use of renewable resources such as solar energy and wind energy. Load management strategy: Adjust load demand through demand response mechanisms, such as adjusting the user's appliance usage time to reduce load demand during peak hours. User behavior recommendations: Provide energy-saving recommendations to users based on load demand forecasts and renewable energy generation forecasts, such as using high-energy-consuming equipment during low electricity prices or peak renewable energy generation periods.

[0111] In an optional embodiment,

[0112] Blockchain is used to record the operating status of microgrids, energy storage dispatch results, and energy transaction information. Comprehensive performance indicators are calculated based on the data recorded on blockchain. Federated learning methods are used for distributed evaluation based on the comprehensive performance indicators. The microgrid energy-saving solution is continuously optimized based on the evaluation results of the comprehensive performance indicators, including:

[0113] Construct a microgrid data blockchain. By designing the block data structure and consensus mechanism, the microgrid operation status, energy storage dispatch results and energy transaction information are recorded, and a data access interface and permission control mechanism are provided. Based on the microgrid operation data recorded in the blockchain, a microgrid comprehensive performance indicator system is constructed, and the indicator calculation tasks are decomposed to each edge node.

[0114] Each edge node regularly calculates the comprehensive performance index based on the microgrid operation data recorded in the blockchain according to the index calculation standard, generates the index hash value, and submits the index hash value to the blockchain network for recording and verification. After the verification is passed, the blockchain network aggregates the comprehensive performance index data submitted by each node;

[0115] Using the comprehensive performance indicator data of each edge node recorded in the blockchain, a federated learning model training sample is constructed, and a distributed training of the microgrid comprehensive performance evaluation model is performed using a federated learning algorithm. By aggregating and updating encrypted model parameters, the trained global evaluation model parameters are obtained;

[0116] Based on the trained global evaluation model parameters, the comprehensive performance index data calculated by each edge node is evaluated, and the quantitative evaluation results of the microgrid performance are output. According to the quantitative evaluation results, the global energy-saving optimization strategy is generated through the intelligent optimization algorithm, and the generated global energy-saving optimization strategy is distributed to each edge node through the blockchain. Each edge node generates local control instructions based on the received global energy-saving optimization strategy and the local microgrid operation status, and sends the control instructions to the corresponding device end through the blockchain to guide the device to adjust the operation status, and upload the device execution results to the blockchain;

[0117] According to the device execution results received by the blockchain, each edge node is re-triggered to calculate the comprehensive performance indicators of the microgrid and generate new evaluation results. When the evaluation results are lower than the preset threshold, the energy-saving strategy optimization process is automatically triggered, and the data generated by the strategy optimization process is recorded on the chain to achieve continuous optimization of the microgrid energy-saving plan.

[0118] Exemplarily, first, a microgrid data blockchain is constructed, and a block data structure and consensus mechanism are designed to record key data such as the microgrid operating status, energy storage dispatch results, and energy transaction information. The block data structure usually consists of two parts: a block header and a block body. The block header records metadata such as the hash value, timestamp, and Merkle root of the previous block, and the block body records the details of the microgrid data transaction. The consensus mechanism adopts a consortium chain mode, with multiple authorized nodes participating in the consensus process. Byzantine fault tolerance algorithms or proof-of-work algorithms are used to ensure data consistency and immutability in the blockchain network. At the same time, the blockchain network provides a data access interface and permission control mechanism, allowing authorized users to read microgrid operating data on demand to ensure data security and privacy. Based on the microgrid operating data recorded in the blockchain, a comprehensive performance indicator system for microgrids is constructed, such as energy utilization efficiency, energy conservation and emission reduction effects, and economic operating costs, and the indicator calculation tasks are decomposed to each edge node to achieve parallel computing.

[0119] As a member of the blockchain network, each edge node regularly calculates the comprehensive performance indicators based on the microgrid operation data recorded in the blockchain according to the indicator calculation standard. The indicator calculation standard specifies the definition, calculation formula, data source, etc. of each indicator to ensure the consistency and comparability of indicator calculation. The edge node performs hash operation on the calculated comprehensive performance indicator data to generate the indicator hash value, and submits the hash value to the blockchain network for recording and verification. Other nodes in the blockchain network verify the submitted hash value to check whether it is consistent with the local calculation result. After verification, the indicator hash value is written into the new block. The blockchain network forms a complete microgrid performance portrait by aggregating the comprehensive performance indicator data submitted by each node, providing data support for subsequent performance evaluation and optimization.

[0120] After obtaining the comprehensive performance index data of the microgrid, the federated learning algorithm is used to perform distributed training on the comprehensive performance evaluation model of the microgrid. Federated learning is a technical framework for machine learning under the premise of protecting data privacy. It realizes model training without exposing the original data through encrypted model parameter aggregation and update. The comprehensive performance index data of each edge node recorded in the blockchain is used as the training sample of federated learning. Each node trains a local evaluation model based on local data, and encrypts the local model parameters and uploads them to the blockchain network. As the coordinator of federated learning, the blockchain network is responsible for receiving the encrypted model parameters of each node, decrypting them for secure aggregation, updating the global evaluation model parameters, and then distributing the updated global parameters to each node. Through multiple rounds of iterative training, a global evaluation model with superior performance is finally obtained, which can accurately evaluate the comprehensive performance of the microgrid.

[0121] Based on the trained global evaluation model parameters, the newly calculated comprehensive performance index data of each edge node is evaluated, and the quantitative evaluation results of the microgrid performance are output. The evaluation results present the performance of the microgrid in different indicator dimensions in the form of intuitive numerical values ​​or levels, and fully reflect the energy-saving optimization potential of the microgrid. According to the quantitative evaluation results, heuristic optimization, evolutionary computing, reinforcement learning and other intelligent optimization algorithms are used to automatically generate a global strategy for microgrid energy-saving optimization. The global energy-saving strategy proposes macro optimization measures such as energy storage scheduling, energy routing, and demand-side response from the perspective of the entire microgrid, and coordinates the operation mode of multiple edge nodes. The blockchain network distributes the generated global energy-saving optimization strategy to each edge node, and the node generates specific device control instructions based on the global strategy and the local microgrid operation status, and securely sends the instructions to the corresponding energy storage equipment, new energy equipment, controllable load equipment, etc. through the blockchain. The device adjusts the operating status according to the received control instructions, and uploads the execution results to the blockchain, and accepts verification and confirmation from the blockchain network.

[0122] In order to achieve continuous optimization of the microgrid energy-saving solution, the edge nodes are re-triggered to calculate the comprehensive performance indicators of the microgrid according to the device execution results received by the blockchain, and the actual effect of the global energy-saving strategy is evaluated. When the evaluation result is lower than the preset threshold, it indicates that there is room for optimization of the current energy-saving solution, and a new round of strategy optimization process is automatically started. All kinds of data generated during the strategy optimization process, such as performance indicators, evaluation results, energy-saving strategies, device instructions, execution results, etc., are recorded and traced through the blockchain, forming a closed loop of microgrid energy-saving optimization.

[0123] Assuming that a microgrid consists of wind power, photovoltaic power, energy storage, diesel generators and multiple loads, a microgrid data management platform based on blockchain is constructed to record the operating data of the microgrid, such as power generation power, power consumption, power balance, and power quality. The platform sets up a comprehensive performance indicator system, including six indicators such as power generation utilization rate, new energy consumption ratio, energy saving and emission reduction, and power supply reliability, and assigns the indicator calculation task to four edge nodes. Each node calculates the performance indicator every 15 minutes, generates an indicator hash value and uploads it to the blockchain, and the blockchain aggregates it to form a complete indicator data set. Then, using the historical indicator data recorded in the blockchain, a training sample with time series characteristics is constructed. Using the federated learning algorithm, the four nodes collaboratively train the recurrent neural network model to obtain a comprehensive performance evaluation model for the microgrid. Whenever new indicator data is generated, the model can be used to carry out performance evaluation and quantify the energy-saving potential of the microgrid. According to the evaluation results, the particle swarm optimization algorithm is used to automatically generate a global energy-saving strategy containing parameters such as battery energy storage power, diesel generator output, and interruptible load power, and send it to each node. The node controls the corresponding device to execute the strategy and uploads the execution result to the blockchain. If it is found that the energy-saving effect is not ideal, for example, the power generation utilization rate is still lower than 90%, a new round of energy-saving optimization will be automatically started to continuously improve the energy-saving level of the microgrid.

[0124] In this embodiment, blockchain technology is used to build a microgrid data management platform to ensure the secure storage, trusted access and full life cycle traceability of key operating data; distributed edge computing is used to efficiently calculate the comprehensive performance indicators of the microgrid and reduce the computing pressure of the data center; a federated learning framework is introduced to achieve distributed training and continuous optimization of the microgrid performance evaluation model while protecting data privacy; fourthly, combined with intelligent optimization algorithms, microgrid energy-saving strategies are automatically generated and optimized, and strategies and instructions are securely distributed using blockchain as a medium to improve energy-saving efficiency; a microgrid energy-saving closed loop is built to organically connect performance evaluation, strategy optimization, instruction issuance, and effect feedback to achieve continuous improvement of microgrid energy-saving solutions.

[0125] In an optional embodiment,

[0126] Based on the quantitative evaluation results, the global energy-saving optimization strategy generated by the intelligent optimization algorithm includes:

[0127] Define the state space and action space of the microgrid. The state space includes the charging and discharging status of the energy storage device, the real-time output of renewable energy, load demand and electricity price fluctuations. The action space includes adjusting the charging and discharging of the energy storage system, the interactive power between the grid and the microgrid, and the optimization of load demand response.

[0128] Design a multidimensional reward function, wherein the multidimensional reward function includes an operating cost value, a renewable energy utilization value, a load response degree value, a dynamic response speed value, and a stability value, and establish a deep reinforcement learning model based on the multidimensional reward function, wherein the intelligent agent in the deep reinforcement learning continuously interacts with the environmental state of the microgrid, and the energy-saving optimization strategy of the intelligent agent is updated based on the instant reward value obtained from each interaction;

[0129] The microgrid is modeled as a multi-agent node network, and the microgrid is abstracted as a network topology composed of multiple intelligent nodes. Each of the intelligent nodes is responsible for managing the local state of its corresponding microgrid. By establishing a communication protocol between the intelligent nodes, the intelligent nodes share their local state information with each other, and the shared local state information is integrated into the global state information. At the same time, the obtained quantitative evaluation results and the global state information are introduced into the deep reinforcement learning model, the energy-saving optimization strategy is iteratively optimized, and finally a global energy-saving optimization strategy is generated.

[0130] Exemplarily, first, the state space and action space of the microgrid are defined to provide a basis for the construction of a deep reinforcement learning model. The state space is a multidimensional vector that describes the operating state of the microgrid, including the charging and discharging state of the energy storage device, the real-time output power of renewable energy, the real-time power of the load demand, and the real-time fluctuation of the electricity price. Among them, the charging and discharging state of the energy storage device reflects the remaining power and current working mode of the energy storage system, the real-time output of renewable energy reflects the power generation power of new energy such as wind power and photovoltaic power, the real-time power of the load demand reflects the power demand on the user side, and the real-time fluctuation of the electricity price reflects the power price signal of the power grid. The action space is a multidimensional vector that describes the optimization control of the microgrid, including adjusting the charging and discharging power of the energy storage system, adjusting the interactive power between the power grid and the microgrid, and adjusting the optimization of the load demand response. Among them, the charging and discharging power regulation of the energy storage system reflects the energy storage and release of the energy storage device in different periods, the interactive power regulation between the power grid and the microgrid reflects the buying and selling of electricity between the microgrid and the main grid, and the optimization of the load demand response reflects the demand-side management of the user-side load.

[0131] Then, a multi-dimensional reward function is designed to evaluate the benefits of different actions taken by the microgrid in different states. The multi-dimensional reward function comprehensively considers multiple objectives of microgrid operation, including operating cost values, renewable energy utilization rate values, load response degree values, dynamic response speed values, and stability values. The operating cost value measures the energy procurement cost and equipment maintenance cost of the microgrid, the renewable energy utilization rate value measures the microgrid's absorption level of new energy such as wind power and photovoltaics, the load response degree value measures the response degree of the user-side load to the demand response signal, the dynamic response speed value measures the response speed of the microgrid to the grid dispatching instruction, and the stability value measures the fluctuation degree of the voltage, frequency and other indicators of the microgrid. According to the multi-dimensional reward function, a deep reinforcement learning model is established, and the mapping relationship from state to action is fitted through a deep neural network, and the model parameters are optimized using algorithms such as Q learning or policy gradient. In the deep reinforcement learning process, the intelligent agent, as the autonomous decision maker of the microgrid, interacts continuously with the microgrid environment, selects actions according to the current state, and observes the next state and the immediate reward value. Based on the immediate reward value obtained from each interaction, the intelligent agent continuously updates its energy-saving optimization strategy and learns the optimal decision through a trial and error mechanism.

[0132] Furthermore, the microgrid is modeled as a multi-agent node network to enhance the collaborative optimization capability between different microgrid nodes. The microgrid is abstracted as a network topology composed of multiple agent nodes, and each agent node corresponds to a key device or subsystem in the microgrid, such as energy storage device node, new energy device node, load device node, etc. Each agent node trains a local deep reinforcement learning model based on its local state information and reward function, and is responsible for managing the local state and action of its corresponding microgrid. By establishing a communication protocol between agent nodes, the agent nodes can share their local state information with each other, and integrate the shared local state information into global state information. At the same time, the quantitative results and global state information obtained from the microgrid performance evaluation are introduced into the deep reinforcement learning model of each agent node as additional state input or reward signal, guiding the agent node to optimize the local strategy while taking into account the global benefit, so as to achieve distributed collaborative optimization. Through the interconnection and strategy iteration of multi-agent nodes, the global optimal energy-saving optimization strategy is finally generated.

[0133] In this embodiment, by defining the state space and action space, the microgrid energy optimization problem is formalized as a Markov decision process, laying the foundation for reinforcement learning; a multi-dimensional reward function is designed to comprehensively characterize the multi-objective characteristics of the microgrid operation, and guide the intelligent agent to learn strategies that take into account multiple optimization objectives; a deep reinforcement learning algorithm is applied to fully mine the optimization knowledge in massive historical operation data, so that the intelligent agent has autonomous learning capabilities and can adapt to the dynamic changes of the microgrid; by introducing a multi-agent framework, the centralized microgrid scheduling is transformed into distributed multi-node autonomy, thereby improving the optimization efficiency and robustness; through communication and collaboration between intelligent agent nodes, the global benefit is maximized on the basis of the local intelligent agent strategy, thereby improving the overall benefit of the microgrid.

[0134] Figure 2 FIG. 1 is a structural diagram of a microgrid energy-saving solution generation system based on energy storage optimization scheduling according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0135] The first unit is used to collect historical operation data, real-time operation data and external environment data of the microgrid system, pre-process and extract features of the collected historical operation data, real-time operation data and external environment data to form a multi-scale spatiotemporal feature representation, and build a joint prediction model of future load demand, renewable energy power generation and electricity market price based on the multi-scale spatiotemporal feature representation, and generate multi-scenario prediction results by performing multi-scenario prediction through Monte Carlo simulation method;

[0136] The second unit is used to construct a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy based on the multi-scenario prediction results, and solve the stochastic dynamic optimization model by using an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set including different weight combinations, and fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set.

[0137] The third unit is used to convert the optimal microgrid energy-saving plan into an intelligent energy-saving operation instruction set, which includes energy storage device control instructions, renewable energy power generation regulation instructions, load management strategies and user behavior suggestions, and performs distributed transmission and collaborative control through edge computing devices. At the same time, blockchain is used to record the microgrid operation status, energy storage scheduling results and energy trading information, and comprehensive performance indicators are calculated based on the data recorded on the blockchain. A federated learning method is used for distributed evaluation based on the comprehensive performance indicators, and the microgrid energy-saving plan is continuously optimized based on the evaluation results of the comprehensive performance indicators.

[0138] According to a third aspect of the embodiments of the present invention,

[0139] An electronic device is provided, comprising:

[0140] processor;

[0141] a memory for storing processor-executable instructions;

[0142] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0143] A fourth aspect of the embodiments of the present invention is:

[0144] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0145] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a microgrid energy-saving solution based on energy storage optimization scheduling, characterized in that: include: Collect historical operation data, real-time operation data and external environment data of the microgrid system, pre-process and extract features of the collected historical operation data, real-time operation data and external environment data to form a multi-scale spatiotemporal feature representation, and build a joint prediction model for future load demand, renewable energy power generation and electricity market price based on the multi-scale spatiotemporal feature representation, and generate multi-scenario prediction results by performing multi-scenario prediction through Monte Carlo simulation method; According to the multi-scenario prediction results, a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy is constructed. The stochastic dynamic optimization model is solved by an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set containing different weight combinations. Fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set. The optimal microgrid energy-saving scheme is converted into an intelligent energy-saving operation instruction set, which includes energy storage device control instructions, renewable energy power generation regulation instructions, load management strategies and user behavior suggestions. Distributed transmission and collaborative control are performed through edge computing devices. At the same time, blockchain is used to record the microgrid operating status, energy storage scheduling results and energy trading information. Comprehensive performance indicators are calculated based on the data recorded on the blockchain. Federated learning methods are used for distributed evaluation based on the comprehensive performance indicators. The microgrid energy-saving scheme is continuously optimized based on the evaluation results of the comprehensive performance indicators.

2. The method according to claim 1, characterized in that Based on the multi-scale spatiotemporal feature representation, a joint prediction model for future load demand, renewable energy generation and electricity market price is constructed, and a multi-scenario prediction is performed through the Monte Carlo simulation method to generate multi-scenario prediction results including: Based on the obtained multi-scale spatiotemporal characteristics, the wavelet transform method is used to extract the time-frequency domain characteristics of load demand data, renewable energy power generation data and electricity market price data to obtain time-frequency domain feature data, and the graph convolutional neural network is used to extract the spatial dependency characteristics of the microgrid topology structure to obtain spatial dependency feature data. At the same time, the attention mechanism is introduced to dynamically adjust the importance weights of the time-frequency domain feature data and the spatial dependency feature data to obtain a spatiotemporal feature data set; Construct a spatiotemporal dynamic graph neural network model, represent the microgrid system as a dynamic graph structure that changes over time, wherein nodes represent power equipment and load points in the microgrid, edges represent the interactive relationship between power equipment and load points, and input the spatiotemporal feature data set into the spatiotemporal dynamic graph neural network model, and establish the spatiotemporal dependency relationship between nodes through the message passing and aggregation operations of the graph neural network; Taking historical data of load demand data, renewable energy generation data and electricity market price data as training samples, the spatiotemporal dynamic graph neural network model is trained by a supervised learning method, and the parameters of the spatiotemporal dynamic graph neural network model are optimized by minimizing the loss function of multi-time step prediction errors to obtain a trained spatiotemporal dynamic graph neural network model; A probability model of uncertain factors is constructed based on electricity market price data and meteorological condition data, and multiple future scenarios are generated through Monte Carlo random sampling. The electricity market price data and meteorological condition data under each future scenario are input into the trained spatiotemporal dynamic graph neural network model to predict the future load demand, renewable energy power generation and electricity price of the microgrid system under each future scenario, and obtain multi-scenario prediction results.

3. The method according to claim 1, characterized in that According to the multi-scenario prediction results, a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy is constructed. The stochastic dynamic optimization model is solved by an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set containing different weight combinations. Fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set, including: A random dynamic optimization model is constructed based on the multi-scenario prediction results. The random dynamic optimization model takes operating cost, renewable energy utilization rate and load demand satisfaction as optimization targets. A comprehensive objective function is formed by introducing weight coefficients to weightedly combine multiple optimization targets. Decision variables include energy storage device charging and discharging power, renewable energy output power, demand response load adjustment amount and grid interaction power. Constraints include energy storage device capacity limit, charging and discharging power limit, renewable energy output upper limit, load demand satisfaction constraint and grid interaction power upper and lower limits. Random variables are introduced to represent the uncertainty of load demand, renewable energy power generation and electricity price. A scenario-based method is used to discretize random variables and transform the random dynamic optimization model into a deterministic optimization problem. An intelligent optimization algorithm with an adaptive parameter adjustment mechanism is used to solve the deterministic optimization problem, initialize the algorithm parameters and randomly generate an initial population. Each individual corresponds to a microgrid optimization scheme. Each individual in the current population is decoded, its objective function value and constraint violation are calculated, the fitness value is determined, the current population is sorted according to the fitness value, and the current optimal solution is recorded. Perform selection, crossover and mutation operations on the sorted population to generate a new child population, merge the child population with the parent population, calculate the fitness value of the merged population, re-sort according to the fitness value, select the individual with the highest fitness as the new parent population, repeat the iteration until the preset termination condition is met, output the current optimal solution as the microgrid optimization scheduling plan, and at the same time, divide the latest parent population into multiple child populations and assign them to different computing nodes for parallel computing. Perform selection, crossover and mutation operations on the child population on each computing node, and regularly migrate populations and exchange information between computing nodes, and finally obtain a solution set of microgrid optimization scheduling plans under different weight combinations; A fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The analytic hierarchy process is used to determine the weight of each evaluation index. The membership degree of each scheme in the microgrid optimization scheduling scheme solution set under each evaluation index is calculated, and the comprehensive evaluation value of each scheme is calculated based on the index weight. The schemes in the microgrid optimization scheduling scheme solution set are ranked according to the comprehensive evaluation value. Combined with the knowledge of domain experts and decision-making preferences, the microgrid optimization scheduling scheme with the highest comprehensive evaluation value is selected from the sorting results as the optimal microgrid energy-saving scheduling scheme.

4. The method according to claim 3, characterized in that A fuzzy evaluation index system including operating cost, renewable energy utilization rate and load demand satisfaction is constructed. The analytic hierarchy process is used to determine the weight of each evaluation index, and the membership degree of each scheme under each evaluation index in the microgrid optimization scheduling solution is calculated. The comprehensive evaluation value of each scheme is calculated in combination with the index weight, including: According to the characteristics of each evaluation index and expert knowledge, the cloud model parameters corresponding to each evaluation index are determined, and the cloud model parameters include expected value, entropy and super entropy, wherein the expected value represents the central position of the membership, the entropy represents the discrete degree of the membership, and the super entropy represents the randomness of the membership; Using the forward cloud generator of the cloud model, a group of cloud droplets are generated according to the determined cloud model parameters, and each cloud droplet represents the membership value of the evaluation index at different membership levels; According to the performance value of each microgrid optimization scheduling scheme under each evaluation index, the reverse cloud generator of the cloud model is used to calculate the probability that the microgrid optimization scheduling scheme belongs to different membership levels, and the membership vector of the current microgrid optimization scheduling scheme is obtained; The membership vector of the current microgrid optimization scheduling scheme is weighted and aggregated with the evaluation index weight to obtain the comprehensive membership of the current microgrid optimization scheduling scheme, and the comprehensive membership is used as the comprehensive evaluation value of the current microgrid optimization scheduling scheme.

5. The method according to claim 4, characterized in that The membership vector of the current microgrid optimization scheduling scheme is weighted and aggregated with the evaluation index weight, and the calculation formula for the comprehensive membership of the current microgrid optimization scheduling scheme is as follows: Among them, S represents the comprehensive membership, n represents the number of evaluation indicators, and w i represents the evaluation index weight, x i represents the specific performance value of the xth microgrid optimization scheduling scheme under the ith evaluation index, E x Indicates the central position of membership, E n Indicates the discrete degree of membership, H e represents the super entropy, which is a measure of the randomness of the membership, α i Represents the dynamic adjustment factor.

6. The method according to claim 1, characterized in that Blockchain is used to record the operating status of microgrids, energy storage dispatch results, and energy transaction information. Comprehensive performance indicators are calculated based on the data recorded on blockchain. Federated learning methods are used for distributed evaluation based on the comprehensive performance indicators. The microgrid energy-saving solution is continuously optimized based on the evaluation results of the comprehensive performance indicators, including: Construct a microgrid data blockchain. By designing the block data structure and consensus mechanism, the microgrid operation status, energy storage dispatch results and energy transaction information are recorded, and a data access interface and permission control mechanism are provided. Based on the microgrid operation data recorded in the blockchain, a microgrid comprehensive performance indicator system is constructed, and the indicator calculation tasks are decomposed to each edge node. Each edge node regularly calculates the comprehensive performance index based on the microgrid operation data recorded in the blockchain according to the index calculation standard, generates the index hash value, and submits the index hash value to the blockchain network for recording and verification. After the verification is passed, the blockchain network aggregates the comprehensive performance index data submitted by each node; Using the comprehensive performance indicator data of each edge node recorded in the blockchain, a federated learning model training sample is constructed, and a distributed training of the microgrid comprehensive performance evaluation model is performed using a federated learning algorithm. By aggregating and updating encrypted model parameters, the trained global evaluation model parameters are obtained; Based on the trained global evaluation model parameters, the comprehensive performance index data calculated by each edge node is evaluated, and the quantitative evaluation results of the microgrid performance are output. According to the quantitative evaluation results, the global energy-saving optimization strategy is generated through the intelligent optimization algorithm, and the generated global energy-saving optimization strategy is distributed to each edge node through the blockchain. Each edge node generates local control instructions based on the received global energy-saving optimization strategy and the local microgrid operation status, and sends the control instructions to the corresponding device end through the blockchain to guide the device to adjust the operation status, and upload the device execution results to the blockchain; According to the device execution results received by the blockchain, each edge node is re-triggered to calculate the comprehensive performance indicators of the microgrid and generate new evaluation results. When the evaluation results are lower than the preset threshold, the energy-saving strategy optimization process is automatically triggered, and the data generated by the strategy optimization process is recorded on the chain to achieve continuous optimization of the microgrid energy-saving plan.

7. The method according to claim 6, characterized in that Based on the quantitative evaluation results, the global energy-saving optimization strategy generated by the intelligent optimization algorithm includes: Define the state space and action space of the microgrid. The state space includes the charging and discharging status of the energy storage device, the real-time output of renewable energy, load demand and electricity price fluctuations. The action space includes adjusting the charging and discharging of the energy storage system, the interactive power between the grid and the microgrid, and the optimization of load demand response. Design a multidimensional reward function, wherein the multidimensional reward function includes an operating cost value, a renewable energy utilization value, a load response degree value, a dynamic response speed value, and a stability value, and establish a deep reinforcement learning model based on the multidimensional reward function, wherein the intelligent agent in the deep reinforcement learning continuously interacts with the environmental state of the microgrid, and the energy-saving optimization strategy of the intelligent agent is updated based on the instant reward value obtained from each interaction; The microgrid is modeled as a multi-agent node network, and the microgrid is abstracted as a network topology composed of multiple intelligent nodes. Each of the intelligent nodes is responsible for managing the local state of its corresponding microgrid. By establishing a communication protocol between the intelligent nodes, the intelligent nodes share their local state information with each other, and the shared local state information is integrated into the global state information. At the same time, the obtained quantitative evaluation results and the global state information are introduced into the deep reinforcement learning model, the energy-saving optimization strategy is iteratively optimized, and finally a global energy-saving optimization strategy is generated.

8. A microgrid energy-saving solution generation system based on energy storage optimization scheduling, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect historical operation data, real-time operation data and external environment data of the microgrid system, pre-process and extract features of the collected historical operation data, real-time operation data and external environment data to form a multi-scale spatiotemporal feature representation, and build a joint prediction model of future load demand, renewable energy power generation and electricity market price based on the multi-scale spatiotemporal feature representation, and generate multi-scenario prediction results by performing multi-scenario prediction through Monte Carlo simulation method; The second unit is used to construct a stochastic dynamic optimization model including energy storage equipment charging and discharging strategy, renewable energy utilization, demand response and grid interaction strategy based on the multi-scenario prediction results, and solve the stochastic dynamic optimization model by using an optimization algorithm that introduces an adaptive parameter adjustment mechanism. The model is executed under a parallel computing framework to obtain a solution set including different weight combinations, and fuzzy decision theory and hierarchical analysis method are used to select the optimal microgrid energy-saving solution from the solution set. The third unit is used to convert the optimal microgrid energy-saving plan into an intelligent energy-saving operation instruction set, which includes energy storage device control instructions, renewable energy power generation regulation instructions, load management strategies and user behavior suggestions, and performs distributed transmission and collaborative control through edge computing devices. At the same time, blockchain is used to record the microgrid operation status, energy storage scheduling results and energy trading information, and comprehensive performance indicators are calculated based on the data recorded on the blockchain. A federated learning method is used for distributed evaluation based on the comprehensive performance indicators, and the microgrid energy-saving plan is continuously optimized based on the evaluation results of the comprehensive performance indicators.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images