Collaborative optimization method and device for distributed micro-grid, and medium

Abstract

The embodiment of the invention discloses a collaborative optimization method and device for a distributed micro-grid and a medium, and relates to the technical field of energy management, and the method comprises the steps: carrying out the real-time data collection through a multi-element sensor network disposed at each distributed energy node in advance, so as to obtain the real-time energy data corresponding to each distributed energy node; according to the edge node corresponding to each distributed energy node, carbon emission of the distributed energy nodes is predicted by using the real-time energy data, and node-level predicted carbon emission parameters are determined, so that single-node optimization reference information corresponding to each distributed energy node is determined; through a preset global center node, collecting single node optimization information corresponding to each distributed energy node for aggregation, and determining global model gradient information; and based on the global model gradient information and the single-node model gradient information, carrying out collaborative optimization on the single-node optimization strategy to determine a collaborative optimization strategy corresponding to each edge node.

Description

Technical Field

[0001] This specification relates to the technical field of energy management, and particularly to a collaborative optimization method, device, and medium for a distributed microgrid. Background Art

[0002] As the core carrier of the intelligent park energy system, the microgrid is highly expected to improve the penetration rate of renewable energy and reduce carbon emissions. With the large-scale access of renewable energy, the collaborative optimization of the distributed microgrid has become a key technology to improve energy utilization efficiency and reduce carbon emissions.

[0003] Traditional methods usually rely on cloud-based centralized computing platforms for global optimization. The centralized architecture needs to upload massive node data (such as photovoltaic output, energy storage SOC, load demand, etc.) to the cloud, with high data transmission latency, making it difficult to meet the minute-level dynamic optimization requirements. Especially when the wind and light output fluctuate violently, the lag of control instructions is likely to cause safety problems such as voltage over-limit. In addition, the node-level raw data (such as user electricity consumption behavior, device operation status) is easily stolen or tampered with during transmission. Existing encryption methods are difficult to balance security and computational real-time performance, and centralized storage increases the risk of single-point attacks. Moreover, most traditional single-node optimization methods adopt single-objective optimization, which cannot meet the optimization requirements.

[0004] Therefore, the collaborative optimization process of the traditional distributed microgrid relies on the centralized computing method, which has the step of uploading the raw data of each node, increasing the optimization response time and the risk of data leakage, and cannot meet the requirements of optimization efficiency and data security in the distributed microgrid scenario. Summary of the Invention

[0005] One or more embodiments of this specification provide a collaborative optimization method, device, and medium for a distributed microgrid to solve the following technical problems: The collaborative optimization process of the traditional distributed microgrid relies on the centralized computing method, which has the step of uploading the raw data of each node, increasing the optimization response time and the risk of data leakage, and cannot meet the requirements of optimization efficiency and data security in the distributed microgrid scenario.

[0006] One or more embodiments of this specification adopt the following technical solutions:

[0007] One or more embodiments of this specification provide a collaborative optimization method for a distributed microgrid. The method includes: performing real-time data collection through a multi-sensor network pre-arranged at each distributed energy node to obtain real-time energy data corresponding to each distributed energy node, where the real-time energy data includes real-time power generation data and real-time environmental data; predicting the carbon emissions of each distributed energy node based on the edge node corresponding to each distributed energy node using the real-time energy data, determining node-level predicted carbon emission parameters, and based on the node-level predicted carbon emission parameters, determining single-node optimization reference information corresponding to each distributed energy node, where the single-node optimization reference information includes single-node model gradient information and single-node optimization strategies; collecting single-node optimization information corresponding to each distributed energy node through a preset global central node to aggregate the single-node model gradient information in the single-node optimization information and determine global model gradient information; and collaboratively optimizing the single-node optimization strategies based on the global model gradient information and the single-node model gradient information to determine collaborative optimization strategies corresponding to each edge node.

[0008] One or more embodiments of this specification provide a collaborative optimization device for a distributed microgrid, including:

[0009] At least one processor; and,

[0010] A memory communicatively connected to the at least one processor; where,

[0011] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the above method.

[0012] A non-volatile computer storage medium provided by one or more embodiments of this specification stores computer-executable instructions, and the computer-executable instructions are set to: execute the above method.

[0013] At least one of the above-mentioned technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects: through the technical solutions of the embodiments of this specification, the traditional method of relying on a cloud-based centralized computing platform for global optimization requires uploading massive node data to the cloud, facing the problem of high data transmission delay, and it is difficult to meet the minute-level dynamic optimization requirements. The embodiments of this specification collect real-time data through a multi-sensor network pre-deployed at each distributed energy node. The data can be preliminarily processed and analyzed at the local node. Each node does not need to upload a large amount of original data to the cloud, reducing the distance and amount of data transmission, and greatly reducing data transmission delay; under the traditional centralized architecture, node-level original data (such as user electricity consumption behavior, equipment operating status) is easy to be stolen or tampered with during transmission, and existing encryption methods are difficult to balance security and computing real-time performance. The technical solution of the embodiments of this specification reduces the transmission of a large amount of original data, reducing the risk of data leakage from the source. In addition, the distributed data collection and processing method avoids the centralized storage brought about. Single point attack risk, even if a node is attacked, it will not affect the data security of the entire system; the traditional centralized computing platform needs to upload a large amount of node data to the cloud for global optimization calculation, the data transmission burden is heavy and the efficiency is low. The embodiment of this specification collects the single node model gradient information in the single node optimization information through the preset global central node for aggregation to determine the global model gradient information. Since the model gradient information transmitted is processed locally by the node, the amount of data is greatly reduced, which reduces the pressure and delay of data transmission. At the same time, the information of each node can be quickly integrated to grasp the optimization direction of the distributed microgrid from a global perspective. Compared with the traditional centralized computing method, the efficiency and accuracy of determining the global model gradient information are greatly improved; the traditional collaborative optimization process relies on centralized computing and uploading of original data. When facing the complex and changeable operating environment of the distributed microgrid, the response speed is slow and the optimization effect is poor. The embodiment of this specification collaboratively optimizes the single node optimization strategy based on the global model gradient information and the single node model gradient information. In complex situations such as drastic fluctuations in wind and solar power output or rapid changes in load demand, the single-node optimization strategy can be quickly adjusted in real time according to the actual situation of each node and the global optimization direction to achieve collaborative work between the nodes; in traditional centralized collaborative optimization, the security issues of data transmission and storage seriously affect the reliability of the system. In the collaborative optimization process, the embodiments of this specification reduce the transmission of original data and reduce the risk of data leakage. The nodes collaborate by exchanging processed model gradient information and other data. These data can be encrypted before transmission, and because the data volume is relatively small, the encryption and decryption process has little impact on system performance. It can achieve efficient collaborative optimization under the premise of ensuring data security, which not only ensures the security of the system, but also improves the efficiency of collaborative optimization, and meets the dual requirements of optimization efficiency and data security in distributed microgrid scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings. In the accompanying drawings:

[0015] Figure 1 It is a schematic flowchart of a collaborative optimization method for a distributed microgrid provided by an embodiment of this specification;

[0016] Figure 2 It is a schematic structural diagram of a collaborative optimization device for a distributed microgrid provided by an embodiment of this specification. Detailed implementation manners

[0017] In order to enable those skilled in the art to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only some embodiments of this specification, rather than all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this specification.

[0018] An embodiment of this specification provides a collaborative optimization method for a distributed microgrid. It should be noted that the execution subject in the embodiments of this specification can be a server or any device with data processing capabilities. Figure 1 It is a schematic flowchart of a collaborative optimization method for a distributed microgrid provided by an embodiment of this specification, as Figure 1 shown, mainly including the following steps:

[0019] Step S101, through the multi-sensor network pre-arranged at each distributed energy node, perform real-time data collection to obtain the real-time energy data corresponding to each distributed energy node.

[0020] Among them, the real-time energy data includes real-time power generation data and real-time environmental data;

[0021] In one embodiment of this specification, a distributed energy node refers to an energy production unit distributed at different geographical locations, and these nodes can be solar power plants, wind farms, small hydropower stations or other types of distributed energy facilities. By collecting real-time data from each node, comprehensive monitoring and management of the entire energy system can be achieved. By deploying a multi-source sensor network at the distributed nodes, real-time power generation data and real-time environmental data corresponding to at least one energy device in the distributed energy node are collected. A variety of different types of sensors are used, and these sensors are distributed on each distributed energy node to collect data in all directions. The real-time power generation data is real-time information about the power generation situation of the distributed energy node, such as the real-time power generation amount of solar panels, the real-time power generation power of wind turbines, etc. The real-time environmental data includes real-time information on the environmental aspects related to energy production, such as light intensity (for solar power generation), wind speed and direction (for wind power generation), temperature, humidity, etc.

[0022] In one embodiment of this specification, the multi-source sensor network may include power generation sensors such as watt-hour meters with a sampling frequency of 0.1 second level, a set of meteorological sensors such as light intensity, wind speed, temperature and humidity, energy storage SOC sensors, and device status monitoring sensors such as inverter efficiency and battery SOC, and collect voltage, current, power and environmental parameter data of distributed power sources (photovoltaic, wind power), energy storage systems and loads at a frequency of ≥1Hz. The data can be transmitted in real time through LoRa wireless networking.

[0023] Step S102, according to the edge node corresponding to each distributed energy node, use the real-time energy data to predict the carbon emissions of the distributed energy node, determine the node-level predicted carbon emission parameters, and based on the node-level predicted carbon emission parameters, determine the single-node optimization reference information corresponding to each distributed energy node.

[0024] Among them, the single-node optimization reference information includes single-node model gradient information and single-node optimization strategies;

[0025] According to the edge node corresponding to each such distributed energy node, use the real-time energy data to predict the carbon emissions of the distributed energy node, and determine the node-level predicted carbon emission parameters, specifically including: obtaining the local historical data corresponding to the edge node, where the local historical data includes historical new energy output data, historical power grid operation data, and historical environmental data; extracting the historical time series features and power grid state features from the local historical data to train a pre-constructed long short-term memory network prediction model to determine a dynamic emission factor prediction model; through the dynamic emission factor prediction model, predict the emission factor of the distributed energy node according to the real-time energy data to determine the corresponding dynamic emission factor; obtain in advance the predicted node carbon emissions corresponding to each such distributed energy node, and according to the dynamic emission factor and the predicted node carbon emissions, determine the node-level carbon emission intensity corresponding to each such distributed energy node to determine the node-level predicted carbon emission parameters, where the node-level predicted carbon emission parameters include the node-level carbon emission intensity and the predicted node carbon emissions.

[0026] In one embodiment of the present specification, each distributed energy node corresponds to an edge node. Data preprocessing, feature extraction and other data processing processes are performed through a GPU-accelerated edge server. The edge node trains a lightweight neural network based on local data. Each distributed energy node has a corresponding edge node. First, obtain the local historical data corresponding to the edge node. The local historical data includes historical new energy output data (such as the power generation of a solar power station and the power generation power of a wind power plant in the past period of time, etc., production data of new energy), historical power grid operation data (such as data such as the voltage, current, and power transmission of the power grid), and historical environmental data (such as environmental parameters such as past light intensity, wind speed, temperature, and humidity). Extract historical time series features and power grid state features from the obtained local historical data. Among them, the historical time series features reflect the change law of data over time, such as the change trend of new energy output in different time periods of a day or a year, etc.; the power grid state features reflect various state information of the power grid operation, such as the load situation of the power grid, power balance, etc. Use these extracted features to train a pre-constructed long short-term memory network (LSTM) prediction model. LSTM is a special recurrent neural network that can effectively process time series data. Through training, the model can learn the patterns and rules in historical data. Through the LSTM prediction model, both a dynamic emission factor prediction model for determining the same dynamic emission factor and a carbon emission prediction model for predicting node carbon emissions can be obtained.

[0027] Examples of implementing the dynamic emission factor prediction model are as follows: First, by collecting the longitude and latitude information of power grid equipment (such as transformers and lines), the longitude and latitude relationship between the starting point and the ending point is established. If the starting longitude and latitude of equipment B are the same as the ending longitude and latitude of equipment A, it indicates that B is a subordinate equipment of A, thus constructing a topological relationship. After that, the branch power loss is calculated from the end node forward, and then the voltage distribution is deduced from the head end to the end node, so as to obtain the power flow situation of the whole network. Based on the power grid topological structure and real-time power flow data, the carbon emission intensity distribution is calculated according to the node carbon flow tracking method. Using the new energy output prediction data obtained in advance through traditional prediction methods, the carbon emission factor is dynamically corrected through a long short-term memory network (LSTM). Specifically, by collecting historical data, such as the new energy power generation, environmental parameters (light intensity, wind speed, etc.), actual carbon emissions and power grid operation status in the past year, key feature variables are extracted, model construction and training are carried out, a cross-validation strategy is adopted to prevent overfitting, the weights are updated using the backpropagation algorithm, the optimization goal is to minimize the mean square error, the difference between the predicted value and the measured value is regularly compared to evaluate the model accuracy, an incremental learning mechanism is introduced, and the model weights are automatically fine-tuned after new data arrives to keep the prediction ability improving over time, and the prediction error is less than 3%. Through the trained dynamic emission factor prediction model, the emission factor of this distributed energy node is predicted according to the real-time energy data (i.e., real-time power generation data and real-time environmental data, etc.) collected currently. The emission factor is an important indicator to measure the carbon emissions generated per unit of energy production or consumption. By predicting the corresponding dynamic emission factor, the carbon emission intensity under the current energy production situation can be reflected.

[0028] Based on historical data and traditional prediction models, the predicted node carbon emissions of this node in a future period of time are obtained. Combining with the predicted dynamic emission factor, the node-level carbon emission intensity C corresponding to each distributed energy node is calculated and determined. node (t). The calculation method is as follows:

[0029] C node (t) = ∑ i＝1 n (E in,i (t)·EF in,i (t)) - ∑ i＝1 n (E out,i (t)·EF out,i (t));

[0030] Among them, E in,i (t), E out,i (t) are the input energy and output energy of the equipment branch node i corresponding to the power grid node at time t respectively, and EF in,i (t), EF out,i(t) is the predicted dynamic carbon emission factor corresponding to the device branch node i. The node-level carbon emission intensity represents the carbon emissions generated per unit of energy production or consumption at this node. The node-level carbon emission intensity and the predicted node carbon emissions are jointly determined as the node-level predicted carbon emission parameters.

[0031] Through the above technical solution, local historical data including historical new energy output data, historical power grid operation data, and historical environmental data is obtained. The multi-dimensional data provides rich information for carbon emission prediction. Compared with a single data source, it can more comprehensively reflect various factors affecting carbon emissions, thereby improving the accuracy of prediction; through the dynamic emission factor prediction model, the emission factor is predicted based on real-time energy data, which can reflect the changes in carbon emission intensity under different energy production situations in real time. The carbon emission monitoring of distributed energy nodes is no longer static, but can be adjusted in real time with changes in factors such as energy production and environment, which helps to detect abnormal situations and change trends of carbon emissions in a timely manner; determining the node-level carbon emission intensity based on the dynamic emission factor and the predicted node carbon emissions can quantify the carbon emission level of each distributed energy node and achieve refined management of the carbon emissions of the entire distributed microgrid; considering real-time environmental data and historical environmental data enables the prediction model and the determined carbon emission parameters to adapt to different environmental condition changes. Whether it is changes in natural environmental factors such as light intensity and wind speed, or changes in human factors such as the power grid operation state, it can respond in a timely manner through real-time data collection and dynamic model adjustment, maintaining the effectiveness of carbon emission prediction and management, and enhancing the adaptability of the system in a complex and changeable environment; traditional methods lack an accurate prediction mechanism for the carbon emissions of distributed energy nodes and cannot effectively evaluate by combining real-time and variable energy data and environmental data. The embodiments of this specification predict the carbon emissions and determine the node-level predicted carbon emission parameters based on the edge nodes corresponding to each distributed energy node using real-time energy data. The method of calculating based on real-time data and local edge nodes can more accurately reflect the carbon emission situation of each node compared with traditional methods.

[0032] Based on the node-level predicted carbon emission parameters, determine the single-node optimization reference information corresponding to each distributed energy node, specifically including: using a pre-constructed multi-objective optimization model, according to the node-level predicted carbon emission parameters and the node parameters corresponding to the distributed energy node, generate the single-node optimization strategy corresponding to each distributed energy node, where the node parameters include node energy device parameters, and the single-node optimization strategy includes energy device control indicators and grid interaction indicators. The energy device control indicators are used to adjust the output information of the power generation devices of the distributed energy node, and the grid interaction indicators are used to determine the power exchange parameters between the distributed energy node and the superior grid; using the optimization objective function corresponding to the multi-objective optimization model, establish a traceable calculation chain, where the traceable calculation chain is the calculation chain from the input weights corresponding to multiple optimization objectives in the optimization objective function to the output total cost; through the traceable calculation chain, determine the single-objective model gradient corresponding to each optimization objective, and perform normalization processing on multiple single-objective model gradients to determine the single-node model gradient information corresponding to the distributed energy node.

[0033] In one embodiment of the present specification, a multi-objective optimization model is pre-constructed, and various factors are comprehensively considered to generate an optimization strategy. According to the previously determined node-level predicted carbon emission parameters (including node-level carbon emission intensity, predicted node carbon emissions, etc.) and the node parameters corresponding to the distributed energy node (including node energy device parameters, such as the type, capacity, efficiency, etc. of the power generation device). Through the calculation and analysis of the multi-objective optimization model, the single-node optimization strategy corresponding to each distributed energy node is generated. It should be noted that the single-node optimization strategy includes energy device and control indicators and grid interaction indicators. The energy device control indicators are used to adjust the output information of the power generation devices of the distributed energy node. For example, for a solar power station, the working state of the solar panels can be adjusted according to this indicator, and the power generation power can be controlled to achieve the purposes of optimizing energy production, reducing carbon emissions, or meeting specific power demands. The grid interaction indicators are used to determine the power exchange parameters between the distributed energy node and the superior grid, such as determining how much power the node transmits to the grid or how much power it obtains from the grid, so as to achieve power balance and optimal configuration between the distributed energy node and the superior grid, and improve the operation efficiency and stability of the entire power system. Most traditional single-node optimization methods adopt single-objective optimization and cannot meet the complex optimization requirements of distributed microgrids. The single-node optimization reference information determined based on the node-level predicted carbon emission parameters in the embodiments of the present specification takes into account multiple objectives such as energy utilization efficiency, cost, and environmental protection. Through this multi-objective optimization method, the potential of each node can be fully exploited, and under different operating conditions, the optimal configuration of energy production, consumption, and transmission can be achieved, improving the operation efficiency and overall performance of the node and meeting the diverse optimization requirements of distributed microgrids.

[0034] The optimization objective function usually includes multiple optimization objectives, such as minimizing carbon emissions, maximizing energy utilization efficiency, minimizing operating costs, etc. Establishing a traceable calculation chain is a calculation chain from the input weights corresponding to multiple optimization objectives in the optimization objective function (i.e., the importance weights of each optimization objective in the comprehensive optimization) to the output total cost (here the total cost may include comprehensive costs such as the cost of energy production and the environmental cost caused by carbon emissions). Through the traceable calculation chain, the gradient of the single-objective model corresponding to each optimization objective is determined. The single-objective model gradient reflects the degree and direction of the impact of the change in the input weight of a certain optimization objective on the output total cost under the condition that other conditions remain unchanged.

[0035] Specifically, when constructing the traceable calculation chain, it can be achieved through the following steps. Rewrite the optimization problem into a neural network form, and use the automatic differentiation function of TensorFlow to trace the variable influence path to establish a complete traceable calculation chain from the input weight to the output total cost. For example, fix the real-time data at the current moment, such as the photovoltaic output of 500 kW and the electricity price of 1.2 yuan, fine-tune the weight parameters, observe the change in the total cost, increase the economic weight from 0.6 to 0.61, keep other weights unchanged, calculate the change in the total cost ΔCost, and obtain the gradient component through the ratio of the change in the total cost to the weight adjustment step size. Calculate the gradients of multiple dimensions simultaneously through matrix operations to avoid the time consumption of successive adjustments. Normalize the gradients of multiple single-objective models and adjust them to a unified scale range for easy comparison and comprehensive analysis. After normalization, determine the single-node model gradient information corresponding to this distributed energy node, which reflects the relative importance and change trend of each optimization objective in the optimization process of this node.

[0036] Before generating the single-node optimization strategy corresponding to each distributed energy node by using the pre-constructed multi-objective optimization model according to the node-level predicted carbon emission parameters and the node parameters corresponding to this distributed energy node, the method further includes: determining the optimization objective function corresponding to this multi-objective optimization model, where the optimization objective function includes an economic objective term, an environmental protection objective term, and a reliability objective term; dynamically adjusting the weights of the objective terms corresponding to this multi-objective optimization model according to the real-time energy data corresponding to this distributed energy node to determine the weights of the objective terms corresponding to this optimization objective function; and determining the multi-objective optimization model through the weights of the objective terms and this optimization objective function.

[0037] In view of the problem that most traditional optimizations adopt single-objective optimization and cannot meet the optimization requirements, in the embodiments of this specification, an optimization objective function is determined through an economic objective item, an environmental protection objective item, and a reliability objective item to perform multi-objective optimization. First, preset optimization objective weights are loaded, for example, 60% for economy, 30% for environmental protection, and 10% for reliability, and static parameters of nodes are injected, such as equipment capacity tables, constraint threshold libraries, etc. The construction of the objective function of the multi-objective optimization model includes an economic objective item, an environmental protection objective item, and a reliability objective item. The economic objective item is obtained by calculating the weighted sum of equipment operation and maintenance costs, external power purchase costs, and power sales revenues; the environmental protection objective item is based on dynamic carbon emission intensity parameters, and the equipment output and external power purchase quantity are converted into equivalent carbon emission costs; the reliability objective item determines the power supply quality score Q according to the voltage violation probability and the frequency fluctuation range. 质量 It is obtained. The following formula is used for normalization and weighting of each objective item: Among them, the values of C reference and E quota are dynamically updated according to historical operation data.

[0038] In view of the problem that in traditional microgrid node optimization, fixed optimization objective weights are adopted (such as 60% for economy and 40% for environmental protection), which cannot respond to the carbon emission intensity generated by real-time environmental changes, in the embodiments of this specification, α, β, and γ are set as dynamic weights, and according to the real-time environmental data in the real-time energy data corresponding to the distributed energy node, the objective item weights of the multi-objective optimization model are dynamically adjusted to determine the objective item weights corresponding to the optimization objective function. When the environmental parameter mutation exceeds the preset threshold, a weight reallocation module is performed. For example, when it is detected that the irradiance decrease rate > 30% / min, the reliability objective weight is increased to more than 50%, and the optimization model is solved after the weight adjustment.

[0039] In an embodiment of this specification, the improved NSGA-II algorithm is adopted. The Pareto optimal solution set is generated through non-dominated sorting, and the final strategy is selected according to real-time requirements. Generally, the solution with the lowest total cost is selected. Specifically, when the improved NSGA-II algorithm is used to generate the Pareto optimal solution set through non-dominated sorting, an adaptive crossover probability operator is introduced, and the mutation probability is automatically increased when the population diversity is detected to decrease. An elite retention strategy is designed to retain the top 10% of the optimal solutions that meet the reliability objective threshold; a preference solution screening mechanism is established, and according to the real-time electricity price signal and the carbon trading price, the solution with the lowest total operating cost in the Pareto solution set is selected as the final strategy.

[0040] Through the above technical solutions, dynamically adjusting the weights of target items according to real-time environmental data can enable the multi-objective optimization model to better adapt to different environmental conditions and operating states. For example, in the case of sufficient light and appropriate wind speed, the weight of the environmental protection target item (such as maximizing the utilization of renewable energy) can be appropriately increased to make full use of clean energy; while during peak power demand or when the stability of the power grid is challenged, the weight of the reliability target item can be increased to ensure the stability of power supply. The dynamic adjustment mechanism enables the model to flexibly optimize according to the actual situation, improving the adaptability and robustness of the system. By dynamically adjusting the weights of target items and determining the multi-objective optimization model, the model can more accurately match the actual operating conditions of distributed energy nodes. The real-time environmental data reflects the current real environmental conditions. Based on this, adjusting the weights can enable the model to more accurately consider the influence of various factors when generating single-node optimization strategies, thereby obtaining more practical, operable, and effective optimization results, improving the operating efficiency and performance of the distributed microgrid.

[0041] Using the pre-constructed multi-objective optimization model, combining the node-level predicted carbon emission parameters and node energy device parameters to generate single-node optimization strategies, comprehensively considering the carbon emission situation and the characteristics of the energy devices themselves, avoiding the limitations brought by considering a single factor; the energy device control indicators and grid interaction indicators included in the single-node optimization strategies enable the power generation devices to operate as efficiently as possible under different working conditions, contributing to achieving power balance between distributed energy nodes and the superior power grid; determining the single-objective model gradient corresponding to each optimization objective through a traceable calculation chain can quantify the change trend and influence degree of each optimization objective during the optimization process. The single-node optimization strategies and single-node model gradient information provide important basic information for the collaborative optimization of the distributed microgrid.

[0042] Step S103, through a preset global central node, collect the single-node optimization information corresponding to each distributed energy node to aggregate the single-node model gradient information in the single-node optimization information and determine the global model gradient information.

[0043] In an embodiment of this specification, the global central node is deployed in the regional dispatching center, with a gradient aggregation module and a digital twin verification platform built-in, and communicates with the edge nodes through a 5G private network. Through the global central node, collect the single-node optimization information corresponding to each distributed energy node. It should be noted that the collected content here only includes the single-node optimization information, that is, the single-node model gradient information and the corresponding single-node optimization strategies corresponding to each edge node, without involving the processing of raw data, thus avoiding the leakage of raw data. Moreover, part of the processing process is implemented in the edge nodes, effectively reducing the data processing pressure on the central node and improving the data processing efficiency.

[0044] Aggregate the single-node model gradient information in the single-node optimization information to determine the global model gradient information, which specifically includes: constructing a gradient encrypted transmission channel to encrypt the single-node model gradient information through a preset encryption algorithm to generate a single-node encrypted transmission gradient; obtaining the single-node training data volume corresponding to each piece of the single-node model gradient information to update the node gradient credibility weight of each distributed energy node based on the single-node training data volume, and determining the single-node gradient credibility weight corresponding to each single-node encrypted transmission gradient; aggregating multiple pieces of the single-node model gradient information through the single-node gradient credibility weight and the single-node encrypted transmission gradient to determine the global model gradient information.

[0045] In one embodiment of this specification, a gradient encrypted transmission channel is established under the federated learning framework, and homomorphic encryption is performed on the single-node model gradient information of each distributed energy node to generate a single-node encrypted transmission gradient. Federated learning is a distributed machine learning framework that allows each participating party (here referring to distributed energy nodes) to collaboratively train a model without sharing the original data. Establishing an encrypted transmission channel is to ensure the security and privacy of data when transmitting model gradient information and prevent information from being stolen or tampered with during transmission. In addition, homomorphic encryption allows calculations to be performed on ciphertext, and the calculation results after decryption are the same as those obtained by directly performing the same calculations on the plaintext. Through homomorphic encryption, subsequent operations such as aggregation can still be performed on the single-node model gradient information without revealing the original content, protecting the privacy of each node while enabling the effective utilization of information.

[0046] The adaptive weighted aggregation algorithm is adopted to dynamically assign weights to the gradient vectors according to the node confidence of the nodes. Specifically, the amount of training data corresponding to the gradient information of each single-node model is obtained. Here, the amount of data is the training data amount of the model for obtaining the single-node optimization strategy, which reflects to a certain extent the reliability and accuracy of the node model. The larger the amount of data, the more fully the model usually learns about the relevant situation, and the more credible the model gradient information output by it. The ratio of the training data amount corresponding to each node to the total amount of training data of all nodes is used to update the node gradient credibility weight of each distributed energy node, and the single-node gradient credibility weight corresponding to the encrypted transmission gradient of each single node is determined. It should be noted that determining the weight based on the ratio of the data amount reflects the emphasis on the nodes with a larger data amount. The larger the ratio, the larger the data share of the node in the entire system, and the corresponding weight is also larger, and its model gradient information will be given higher importance in the subsequent aggregation process. Using the single-node gradient credibility weight and the encrypted transmission gradient of the single node, the gradient information of multiple single-node models is weighted and aggregated. Weighted aggregation comprehensively calculates the gradient information of the model according to the weight of each node. The gradient information of the node with a larger weight plays a greater role in the formation of the global model gradient information. Through the weighted aggregation operation, the global model gradient information is finally determined, which comprehensively considers the model gradient information of each distributed energy node and its credibility weight, and can more accurately reflect the optimization direction and trend of the entire distributed energy system.

[0047] After aggregating the single-node model gradient information in the single-node optimization information and determining the global model gradient information, the method further includes: evaluating each single-node model gradient information according to the global model gradient information to determine the specified node model gradient information whose gradient deviation is greater than the preset threshold; sending the specified node model gradient information to the corresponding specified edge node to facilitate the specified edge node to update the local node model.

[0048] In one embodiment of the present specification, based on the determined global model gradient information, each single-node model gradient information is evaluated with this as a reference standard. The global model gradient information synthesizes the information of each distributed energy node and represents the optimization direction and trend of the entire distributed energy system. By comparing the single-node model gradient information with it, the degree of fit between each node's model and the overall system can be measured. During the evaluation process, the single-node model gradient information with a gradient deviation greater than a preset threshold is found and determined as the specified node model gradient information. Here, the gradient deviation refers to the degree of difference between the single-node model gradient information and the global model gradient information, and the preset threshold is a standard value set in advance to determine whether this difference is large enough to require further processing. If the gradient deviation of a certain node exceeds this threshold, it indicates that there is a large difference between the model of this node and the optimization direction of the overall system, and there may be problems such as inaccurate models, abnormal data, or other issues, which need to be adjusted.

[0049] Send the determined specified node model gradient information to the corresponding specified edge node. Each distributed energy node has a corresponding edge node, and the edge node manages and maintains the node's model locally. Sending the specified node model gradient information to the corresponding edge node is to enable the edge node to obtain the abnormal information related to its own node model for model update. After receiving the specified node model gradient information, the specified edge node uses this information to update the local node model. Since these gradient information reflect the deviation between the node model and the overall system optimization direction, by updating the local node model, the model of this node can better align with the global model without the training data of other nodes, improving the accuracy and effectiveness of the node model. By updating the local node model, the models of each node can work better together, optimizing the performance of the entire distributed energy system. For example, in carbon emission prediction and energy optimization, the models of each node are more accurate, which helps to more precisely control carbon emissions and schedule energy, improving the operation efficiency and reliability of the distributed microgrid and achieving the overall optimization of the system.

[0050] Step S104, based on the global model gradient information and the single-node model gradient information, co-optimize the single-node optimization strategy to determine the co-optimization strategy corresponding to each edge node.

[0051] Based on the global model gradient information and the single-node model gradient information, co-optimize the single-node optimization strategy to determine the co-optimization strategy corresponding to each edge node, specifically including: based on the global model gradient information and the single-node model gradient information, determine the gradient deviation information corresponding to each single-node optimization strategy; through the gradient deviation information, optimize the single-node optimization strategy to determine the co-optimization strategy corresponding to each edge node.

[0052] In an embodiment of the present specification, the global model gradient information is obtained by synthesizing the single-node model gradient information of each distributed energy node, representing the optimization direction and trend of the entire distributed energy system; the single-node model gradient information reflects the change of each distributed energy node's own model during the optimization process. Based on the two gradient informations, determine the gradient deviation information corresponding to each single-node optimization strategy. By comparing the differences between each single-node model gradient information and the global model gradient information, calculate the deviation degree between them. This deviation degree can reflect the degree of fit between the optimization strategy of each node and the overall system optimization direction. For example, if the single-node model gradient information of a certain node is quite different from the global model gradient information, it indicates that the optimization strategy of this node may not conform to the overall system optimization goal under the current circumstances and needs to be adjusted. Adjust the single-node optimization strategy to determine the co-optimization strategy corresponding to each edge node. The co-optimization strategy here is obtained by comprehensively considering the optimization requirements of the global system and the characteristics of each node. By optimizing the single-node optimization strategy, the nodes can cooperate better with each other to jointly achieve the optimized operation of the distributed energy system. For example, in terms of energy distribution, the co-optimization strategies of different nodes can ensure more reasonable distribution of energy among the nodes and improve energy utilization efficiency; in terms of carbon emission control, the co-optimization strategies can enable each node to cooperate with each other in the process of reducing carbon emissions to achieve the carbon emission target of the entire system.

[0053] Through the gradient deviation information, optimize the single-node optimization strategy to determine the co-optimization strategy corresponding to each edge node, specifically including: determine the co-optimization target item among multiple optimization target items through the deviation gradient direction feature in the gradient deviation information; determine the optimization adjustment parameter corresponding to the co-optimization target item according to the deviation gradient magnitude feature in the gradient deviation information; based on the co-optimization target item and the optimization adjustment parameter, determine the current co-optimization strategy corresponding to the edge node; use the digital twin system to conduct a stability test according to the current co-optimization strategy corresponding to each edge node to determine the strategy verification status corresponding to the current co-optimization strategy; when the strategy verification status passes, distribute the current co-optimization strategy to the corresponding edge node.

[0054] In one embodiment of this specification, the global central node receives the single-node model gradients of each edge node (including the gradient directions of economic, environmental, and reliability sub-goals), and calculates the direction similarity between them and the global gradient. For example, if the gradient direction of the environmental sub-goal of a certain node deviates from the global direction by more than 30 degrees (cosine similarity < 0.86), it is marked as a "high-deviation target item". When the consistency of the economic sub-goal gradient direction is higher than 90%, this target is locked as a priority collaborative optimization item. In a multi-objective optimization scenario, each sub-goal (such as economy and environmental protection) corresponds to an independent gradient direction. The gradient direction consistency reflects whether the optimization directions of each distributed node converge on this goal. High consistency (> 90%) indicates that the parameter adjustment directions of most nodes are highly consistent on this goal (such as all needing to enhance economic benefits), and low consistency reflects conflicts in the optimization directions among nodes (such as some nodes needing to increase power generation while others need to suppress output). Prioritizing goals with high consistency can avoid the "seesaw effect" in multi-objective optimization, that is, the situation where optimizing goal A deteriorates goal B.

[0055] Determine the adjustment amplitude of the optimization parameter according to the gradient magnitude (i.e., the overall strength of the gradient vector). When the magnitude is at a historical high (exceeding 2 standard deviations above the mean), trigger an aggressive adjustment mode, allowing a maximum 15% relaxation of the constraint conditions. When the magnitude is at a low level, adopt conservative adjustment, restricting the parameter change rate within 5%. For the marked collaborative optimization target item (such as environmental protection), generate an optimization strategy in combination with the adjustment amplitude. Set a dynamic adjustment rule for the upper limit of the charge-discharge power of the energy storage system. During the peak photovoltaic output period at noon, allow the energy storage charging power to increase by 10%. Control the adjustable load in a hierarchical manner. When the carbon emission intensity exceeds the standard, give priority to shutting down non-essential loads (such as landscape lighting).

[0056] Inject the generated strategy into the digital twin system to simulate extreme scenarios, such as simulating a 40% sudden drop in photovoltaic output lasting for 10 minutes to verify whether the energy storage system can increase the discharge power according to the strategy to avoid voltage dips. Another example is to simulate a ±0.5Hz fluctuation in the frequency of the superior power grid to test the ability of the strategy to suppress the power fluctuation at the grid connection point. Verify voltage stability, equipment safety, and the corresponding timeliness. The node voltage deviation continuously does not exceed ±5%, the overload rate of key equipment < 10%, and the transmission delay of the strategy instruction < 200ms. If all test scenarios meet the indicators, mark the strategy as "verified passed" and automatically send it to the corresponding edge node for execution. If voltage over-limit or equipment overload is detected, trigger the strategy rollback mechanism, load the safe strategy version of the previous time period, and regenerate and correct the strategy. Distribute the verified passed strategy to the edge node through an encrypted channel. The strategy file contains a timestamp, a version number, and a digital signature; the energy storage controller receives the charge-discharge power instruction and adjusts the output according to the second-order smoothing algorithm (to prevent power mutations). The adjustable load manager performs hierarchical control according to the priority. For example, shut down the third-level load within 5 seconds after the strategy takes effect.

[0057] By determining the collaborative optimization target term through the deviation gradient direction feature, the key direction that needs to be optimized in the system can be accurately found, avoiding blind optimization; by determining the optimization adjustment parameters based on the deviation gradient magnitude feature, the quantitative control of the optimization intensity is realized, and the adjustment amplitude can be reasonably determined according to the actual deviation; by combining the collaborative optimization target term and the optimization adjustment parameters to determine the collaborative optimization strategy, the overall goal of the system and the actual situation of each edge node can be fully considered, and an operable and collaborative optimization strategy can be formulated, which helps to improve the overall performance and operation efficiency of the distributed energy system and realize the coordinated operation between nodes; by using the digital twin system for stability testing and comprehensively verifying and evaluating it before the implementation of the strategy, potential problems and risks can be discovered in advance, avoiding situations such as system instability and performance degradation in actual applications, greatly improving the reliability and success rate of the strategy, and reducing the implementation cost and risk.

[0058] Through the technical solutions of the embodiments of this specification, the traditional method of relying on a cloud-based centralized computing platform for global optimization requires uploading massive node data to the cloud, facing the problem of high data transmission delay, and it is difficult to meet the minute-level dynamic optimization needs. The embodiments of this specification collect real-time data through a multi-sensor network pre-deployed at each distributed energy node. The data can be preliminarily processed and analyzed at the local node, and each node does not need to upload a large amount of original data to the cloud, reducing the distance and amount of data transmission, greatly reducing data transmission delay; under the traditional centralized architecture, node-level original data (such as user electricity consumption behavior, equipment operating status) is easy to be stolen or tampered with during transmission, and existing encryption methods are difficult to balance security and computing real-time performance. The technical solutions of the embodiments of this specification reduce the transmission of a large amount of original data, reducing the risk of data leakage from the source. In addition, the distributed data collection and processing method avoids the risk of single-point attacks brought by centralized storage. Even if a node is attacked hit, it will not affect the data security of the entire system; the traditional centralized computing platform needs to upload a large amount of node data to the cloud for global optimization calculation, the data transmission burden is heavy and the efficiency is low, the embodiment of this specification collects the single-node model gradient information in the single-node optimization information through the preset global central node for aggregation to determine the global model gradient information, because the model gradient information transmitted is processed locally by the node, the amount of data is greatly reduced, the data transmission pressure and delay are reduced, and the information of each node can be quickly integrated at the same time, and the optimization direction of the distributed microgrid can be grasped from a global perspective. Compared with the traditional centralized computing method, the efficiency and accuracy of the global model gradient information determination are greatly improved; the traditional collaborative optimization process relies on centralized computing and uploading of original data. When facing the complex and changeable operating environment of the distributed microgrid, the response speed is slow and the optimization effect is poor. The embodiment of this specification collaboratively optimizes the single-node optimization strategy based on the global model gradient information and the single-node model gradient information. In complex situations such as drastic fluctuations in wind and solar power output or rapid changes in load demand, the single-node optimization strategy can be quickly adjusted in real time according to the actual situation of each node and the global optimization direction to achieve collaborative work between the nodes; in traditional centralized collaborative optimization, the security issues of data transmission and storage seriously affect the reliability of the system. In the collaborative optimization process, the embodiments of this specification reduce the transmission of original data and reduce the risk of data leakage. The nodes collaborate by exchanging processed model gradient information and other data. These data can be encrypted before transmission, and because the data volume is relatively small, the encryption and decryption process has little impact on system performance. It can achieve efficient collaborative optimization under the premise of ensuring data security, which not only ensures the security of the system, but also improves the efficiency of collaborative optimization, and meets the dual requirements of optimization efficiency and data security in distributed microgrid scenarios.

[0059] The embodiments of this specification also provide a collaborative optimization device for a distributed microgrid, such as Figure 2 shown. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above method.

[0060] The embodiments of this specification also provide a non-volatile computer storage medium storing computer-executable instructions, and the computer-executable instructions are set to: execute the above method.

[0061] The various embodiments in this specification are all described in a progressive manner. For the same or similar parts among the various embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the device, equipment, and non-volatile computer storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0062] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The device and medium provided by the embodiments of this specification correspond one-to-one with the method. Therefore, the device and medium also have beneficial technical effects similar to those of their corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and medium will not be elaborated here.

[0064] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, this specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0065] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the specification. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce a means for implementing the specified function in the process Figure 1 one process or multiple processes and / or blocks Figure 1 or a means for implementing the specified function in multiple blocks.

[0066] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means that implements the specified function in the process Figure 1 one process or multiple processes and / or blocks Figure 1 or a means for implementing the specified function in multiple blocks.

[0067] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the specified function in the process Figure 1 one process or multiple processes and / or blocks Figure 1 or a means for implementing the specified function in multiple blocks.

[0068] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0069] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0070] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0071] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0072] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A collaborative optimization method for distributed microgrids, characterized in that: The method comprises: Real-time data collection is performed through a multi-sensor network pre-arranged at each distributed energy node to obtain real-time energy data corresponding to each distributed energy node, wherein the real-time energy data includes real-time power generation data and real-time environmental data; According to the edge nodes corresponding to each of the distributed energy nodes, the carbon emissions of the distributed energy nodes are predicted using the real-time energy data to determine the node-level predicted carbon emission parameters, so as to determine the single-node optimization reference information corresponding to each of the distributed energy nodes based on the node-level predicted carbon emission parameters, wherein the single-node optimization reference information includes single-node model gradient information and single-node optimization strategy; Through a preset global central node, the single-node optimization information corresponding to each of the distributed energy nodes is collected to aggregate the single-node model gradient information in the single-node optimization information to determine the global model gradient information; Based on the global model gradient information and the single-node model gradient information, the single-node optimization strategy is collaboratively optimized to determine the collaborative optimization strategy corresponding to each of the edge nodes.

2. A collaborative optimization method for a distributed microgrid according to claim 1, characterized in that: According to the edge node corresponding to each distributed energy node, the carbon emission of the distributed energy node is predicted using the real-time energy data to determine the node-level predicted carbon emission parameters, specifically including: Acquire local historical data corresponding to the edge node, wherein the local historical data includes historical new energy output data, historical power grid operation data and historical environmental data; Extracting historical time series features and grid status features from the local historical data to train a pre-built long short-term memory network prediction model and determine a dynamic emission factor prediction model; By using the dynamic emission factor prediction model, the emission factor of the distributed energy node is predicted according to the real-time energy data to determine the corresponding dynamic emission factor; The predicted node carbon emissions corresponding to each of the distributed energy nodes are obtained in advance, and the node-level carbon emission intensity corresponding to each of the distributed energy nodes is determined according to the dynamic emission factor and the predicted node carbon emissions to determine the node-level predicted carbon emission parameters, wherein the node-level predicted carbon emission parameters include the node-level carbon emission intensity and the predicted node carbon emissions.

3. The collaborative optimization method for a distributed microgrid according to claim 1, characterized in that: Based on the node-level predicted carbon emission parameters, determining the single-node optimization reference information corresponding to each of the distributed energy nodes specifically includes: Using a pre-built multi-objective optimization model, a single-node optimization strategy corresponding to each distributed energy node is generated according to the node-level predicted carbon emission parameters and the node parameters corresponding to the distributed energy node, wherein the node parameters include node energy equipment parameters, and the single-node optimization strategy includes energy equipment control indicators and power grid interaction indicators, the energy equipment control indicators are used to adjust the power generation equipment output information of the distributed energy node, and the power grid interaction indicators are used to determine the power exchange parameters between the distributed energy node and the upper-level power grid; Using the optimization objective function corresponding to the multi-objective optimization model, a traceable calculation chain is established, wherein the traceable calculation chain is a calculation chain from input weights to output total costs corresponding to multiple optimization objectives in the optimization objective function; Through the traceable calculation chain, the single-objective model gradient corresponding to each optimization target is determined, multiple single-objective model gradients are normalized, and the single-node model gradient information corresponding to the distributed energy node is determined.

4. A collaborative optimization method for distributed microgrids according to claim 3, characterized in that: Before generating a single-node optimization strategy corresponding to each distributed energy node by using a pre-built multi-objective optimization model according to the node-level predicted carbon emission parameters and the node parameters corresponding to the distributed energy nodes, the method further includes: Determining an optimization objective function corresponding to the multi-objective optimization model, wherein the optimization objective function includes an economic objective item, an environmental objective item, and a reliability objective item; According to the real-time energy data corresponding to the distributed energy node, dynamically adjust the target item weights corresponding to the multi-objective optimization model to determine the target item weights corresponding to the optimization objective function; A multi-objective optimization model is determined by using the objective item weights and the optimization objective function.

5. A collaborative optimization method for distributed microgrids according to claim 1, characterized in that: Aggregating the single-node model gradient information in the single-node optimization information to determine the global model gradient information specifically includes: Constructing a gradient encryption transmission channel to encrypt the single-node model gradient information through a preset encryption algorithm to generate a single-node encrypted transmission gradient; Obtain the amount of single-node training data corresponding to each of the single-node model gradient information, update the node gradient credibility weight of each of the distributed energy nodes based on the single-node training data amount, and determine the single-node gradient credibility weight corresponding to each of the single-node encrypted transmission gradients; The single-node model gradient information is aggregated through the single-node gradient credibility weight and the single-node encrypted transmission gradient to determine the global model gradient information.

6. A collaborative optimization method for distributed microgrids according to claim 5, characterized in that: After aggregating the single-node model gradient information in the single-node optimization information to determine the global model gradient information, the method further includes: According to the global model gradient information, each of the single node model gradient information is evaluated to determine the designated node model gradient information whose gradient deviation is greater than a preset threshold; The designated node model gradient information is sent to the corresponding designated edge node, so that the designated edge node can update the local node model.

7. A collaborative optimization method for distributed microgrids according to claim 1, characterized in that: Based on the global model gradient information and the single-node model gradient information, the single-node optimization strategy is collaboratively optimized to determine the collaborative optimization strategy corresponding to each edge node, specifically including: Based on the global model gradient information and the single-node model gradient information, determining the gradient deviation information corresponding to each of the single-node optimization strategies; The single-node optimization strategy is optimized through the gradient deviation information to determine the collaborative optimization strategy corresponding to each edge node.

8. A collaborative optimization method for distributed microgrids according to claim 7, characterized in that: The single node optimization strategy is optimized through the gradient deviation information to determine the collaborative optimization strategy corresponding to each edge node, specifically including: Determining a collaborative optimization target item from a plurality of optimization target items by using the deviation gradient direction feature in the gradient deviation information; Determining the optimization adjustment parameter corresponding to the collaborative optimization target item according to the deviation gradient modulus length feature in the gradient deviation information; Determining a current collaborative optimization strategy corresponding to the edge node based on the collaborative optimization target item and the optimization adjustment parameter; Using the digital twin system, a stability test is performed according to the current collaborative optimization strategy corresponding to each edge node to determine the strategy verification status corresponding to the current collaborative optimization strategy; When the strategy verification status passes, the current collaborative optimization strategy is distributed to the corresponding edge node.

9. A collaborative optimization device for distributed microgrids, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the method according to any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured to execute the method according to any one of claims 1 to 8.

Images