A data analysis method and system for optimizing the supply of regional power resources

Abstract

The present application provides a data analysis method and system for optimizing the supply of regional power resources, which relates to the technical field of industrial data mining and analysis. The method includes: obtaining the substation operation time series data and energy storage station operation time series data of the corresponding historical first time period of the target area node to be analyzed; determining the preliminary load balancing measures for the corresponding future second time period based on the substation scheduling optimization model; determining the preliminary charge and discharge operation measures for the corresponding second time period based on the energy storage station scheduling optimization model; determining the target load balancing measures and target charge and discharge operation measures for the corresponding second time period based on the measure comprehensive optimization model, and the measure comprehensive optimization model includes a data preprocessing layer, a graph neural network layer, a feature fusion layer and a reinforcement learning layer. Thus, the comprehensive scheduling analysis of the substation and the energy storage station is carried out to meet the scheduling requirements of the modern power system.

Description

Technical Field

[0001] The present application relates to the technical field of industrial data mining and analysis, and in particular to a data analysis method and system for optimizing regional power resource supply. Background Art

[0002] With the continuous growth of electricity demand and the development of new energy technologies, the efficient supply and management of regional power resources have become particularly important. At the same time, with the rapid development of energy storage technology, some regions have configured energy storage stations, especially hot spots where there are even multiple energy storage stations. Energy storage systems are gradually being used in power systems to smooth load fluctuations, improve the utilization rate of renewable energy, and so on.

[0003] In the existing regional power resource dispatching system, the focus is often on the dispatching and management of substations, while ignoring the potential value of energy storage stations in power dispatching. With the development of new energy technologies, the configuration of energy storage stations has gradually increased, especially in some hot spots. The existence of multiple energy storage stations makes the solution of dispatching substations alone unable to meet the needs of modern power systems.

[0004] In addition, in the process of power resource dispatching, cost control is an important but difficult to achieve accurately. Traditional dispatching methods usually find it difficult to optimize the balance between cost and benefit while ensuring the security of power supply. For example, when performing charging and discharging operations, energy storage stations need to consider factors such as fluctuations in electricity market prices, battery life, and maintenance costs, but existing technologies lack accurate cost assessment and optimization methods, resulting in high operating costs and unsatisfactory benefits. Similarly, when dispatching loads, substations often find it difficult to take into account both the reliability and economy of power supply, resulting in unreasonable load distribution and high operating costs.

[0005] Therefore, how to comprehensively dispatch substations and energy storage stations to achieve the optimal allocation of power resources has become a difficult problem that needs to be solved urgently in the industry. Summary of the invention

[0006] The present application provides a data analysis method, system, storage medium, computer program product and electronic device for optimizing regional power resource supply, which is used to at least solve the problem that the current power resource scheduling does not comprehensively consider energy storage stations, resulting in unreasonable load distribution, poor power supply reliability and high operating costs.

[0007] In a first aspect, an embodiment of the present application provides a data analysis method for optimizing the supply of regional power resources, including: obtaining the substation operation time series data and energy storage station operation time series data of a corresponding historical first time period of a target regional node to be analyzed; the data types of the substation operation time series data include: load records, peak-valley loads, operating costs, equipment status data, and historical maintenance records; the data types of the energy storage station operation time series data include: charge and discharge records, equipment status, remaining capacity, operating costs, electricity market prices, and weather data; inputting the substation operation time series data into a substation scheduling optimization model to determine a preliminary load balancing measure for a corresponding future second time period; the substation scheduling optimization model adopts a multi-objective optimization model; inputting the energy storage station operation time series data and the weather forecast information for the corresponding second time period into an energy storage station scheduling optimization model to determine a preliminary charge and discharge operation measure for the corresponding second time period; the energy storage station scheduling optimization model adopts a time series prediction model; inputting the substation operation time series data, the energy storage station operation time series data, the preliminary load balancing measure, and the preliminary charge and discharge operation measure into a measure comprehensive optimization model to determine a target load balancing measure and a target charge and discharge operation measure for the corresponding second time period; the measure comprehensive optimization model includes a data preprocessing layer, a graph neural network layer, a feature fusion layer, and a reinforcement learning layer; the data preprocessing layer is used to preprocess the substation operation time series data, the energy storage station operation time series data, the preliminary load balancing measure, and the preliminary charge and discharge operation measure to obtain corresponding first time series features, second time series features, first measure features, and second measure features; inputting the first time series feature and the second time series feature into the graph neural network layer to obtain corresponding substation dependence features and energy storage station dependence features respectively, specifically including: constructing a graph structure based on the first time series feature and the second time series feature; the graph structure includes multiple graph nodes and edges; each graph node is defined by the time series feature of the corresponding substation or energy storage station, and each edge is defined by a substation and an energy storage station having a power connection relationship; iteratively updating the feature vectors of each graph node based on the graph neural network layer to gradually fuse the information of neighbor nodes and realize the extraction of global features, so as to obtain the substation dependence features and energy storage station dependence features corresponding to each final graph node:

[0008] Initialize the feature vector of each graph node i x i represents the time series feature of the substation or energy storage station corresponding to graph node i;

[0009] At each layer l, each graph node i receives information from its neighbor nodes N(i) and aggregates it:

[0010]

[0011] In the formula, is the message of graph node i at the (l + 1)-th layer, and w ij represents the edge weight corresponding to the edge connecting graph node i and graph node j. is the feature vector of graph node i at the (l + 1)-th layer, and W (l) and b (l) are the weight and bias at the l-th layer, and σ is the activation function.

[0012] It is iteratively updated through multiple graph neural network layers to gradually fuse global information to obtain the final node feature vector:

[0013]

[0014] In the formula, L is the number of layers of the graph neural network layer, and H i is the final feature vector of graph node i, which is used to determine the corresponding substation-dependent feature or energy storage station-dependent feature.

[0015] The feature fusion layer is used to fuse the substation-dependent feature, the energy storage station-dependent feature, the first measure feature, and the second measure feature to obtain a comprehensive feature representation:

[0016] H fusion = Concat(H GNN-substation , H GNN-storage , F substation , F storage )

[0017] In the formula, H fusion represents the comprehensive feature vector, H GNN-substation represents the substation-dependent feature, H GNN-storage represents the energy storage station-dependent feature, F substation represents the first measure feature, and F storage represents the second measure feature.

[0018] Based on the comprehensive feature representation processed by the reinforcement learning layer, the corresponding target load balancing measures and target charge and discharge operation measures are determined, specifically including:

[0019] Each state in the state space of the reinforcement learning layer is defined by the following formula:

[0020] s = H fusi on

[0021] In the formula, s represents the state of the reinforcement learning layer, which is defined by the corresponding comprehensive feature representation.

[0022] The policy update formula is:

[0023] π(a|s) = softmax(Wπ ·s + b π )

[0024] wherein, π(a|s) represents the policy of the reinforcement learning layer, which represents the probability of selecting action a in state s; a is defined by the corresponding load adjustment amplitude and charge-discharge operation amplitude; W π and b π respectively represent the policy weight matrix and the policy bias vector;

[0025] The reward function is:

[0026]

[0027] wherein, R represents the reward value, P profit represents the overall revenue of the power system, P cost represents the operating cost of the power system; L balanced represents the regional power load after load balance adjustment, L total represents the total load of the power system, E loss represents the energy loss; λ 1 , λ 2 , λ 3 respectively represent the weight coefficients for measuring the economic benefits, stability and efficiency of the system;

[0028] Determine the target load balance measure and the target charge-discharge operation measure according to the action with the maximum corresponding reward value.

[0029] In a second aspect, an embodiment of the present application provides a data analysis system for optimizing the supply of regional power resources, which is used to implement the steps of the data analysis method for optimizing the supply of regional power resources in any embodiment of the present application.

[0030] In a third aspect, an electronic device is provided, which 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 steps of the data analysis method for optimizing the supply of regional power resources in any embodiment of the present application.

[0031] In a fourth aspect, an embodiment of the present application provides a storage medium, on which a computer program is stored, and characterized in that the program realizes the steps of the data analysis method for optimizing the supply of regional power resources in any embodiment of the present application when executed by a processor.

[0032] Fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instructions, which when executed by a processor implement the steps of the data analysis method for optimizing regional power resource supply in any embodiment of the present application.

[0033] Through the data analysis method for optimizing regional power resource supply provided by the present application, the following technical effects can be at least achieved:

[0034] (1) By respectively inputting the operation timing data of the substation and the energy storage station into the substation scheduling optimization model and the energy storage station scheduling optimization model, the respective operation characteristics and constraint conditions can be comprehensively considered, and preliminary load balancing and charge-discharge operation measures can be formulated. Through these preliminary measures, the load scheduling ability of the substation or the energy storage station can be significantly improved, and the load balancing ability can be better supported to smooth the load fluctuations. Subsequently, the measure comprehensive optimization model comprehensively optimizes the scheduling measures of the substation and the energy storage station, extracts and fuses the global features through the graph neural network layer, and can accurately capture the dependence relationship between the substation and the energy storage station by iteratively updating the information of the neighbor nodes, make full use of the complementary characteristics of the substation and the energy storage station, realize the overall optimal allocation of power resources, and enhance the global optimization effect on the entire power system.

[0035] (2) In this technical solution, the measure comprehensive optimization model comprehensively considers the timing characteristics and preliminary scheduling measures of the substation and the energy storage station, and obtains a comprehensive feature representation through the feature fusion layer, ensuring the comprehensiveness and scientificity of the scheduling decision. Furthermore, the reinforcement learning layer selects the optimal load balancing measures and charge-discharge operation measures through policy update according to the overall benefits, economic benefits, stability and efficiency of the power system, not only ensuring the reliability of power supply, but also maximizing the economic benefits and significantly reducing the operation cost.

[0036] (3) The timing prediction model is used to predict the charge-discharge operation of the energy storage station, and combined with the weather forecast information, the charge-discharge plan of the energy storage station is flexibly adjusted, so that the system can better adapt to the load change and external environment change, and provide the flexibility and adaptability of the preliminary charge-discharge operation measures predicted by the system. The substation scheduling optimization model adopts a multi-objective optimization model, and outputs preliminary load balancing measures by simultaneously considering multiple objectives, which can effectively avoid the situation of unreasonable load balancing measures.

[0037] Through this technical solution, the substation and the energy storage station are comprehensively scheduled and analyzed, and technical means such as multi-objective optimization, timing prediction, graph neural network and reinforcement learning are adopted, which significantly improves the efficiency, reliability and economy of the power system resource supply, and has strong flexibility and adaptability, meeting the scheduling requirements of modern power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0039] Figure 1 Fig. 5 shows a flowchart of an example of a data analysis method for optimizing regional power resource supply according to an embodiment of the present application;

[0040] Figure 2 Fig. 9 shows a schematic structural connection diagram of an example of a measure comprehensive optimization model according to an embodiment of the present application;

[0041] Figure 3 Fig. 13 shows an operation flowchart of an example of processing time series features using a graph neural network layer according to an embodiment of the present application;

[0042] Figure 4 Fig. 17 shows a schematic structural connection diagram of an example of an energy storage station scheduling optimization model according to an embodiment of the present application;

[0043] Figure 5 Fig. 21 shows a schematic structural block diagram of an example of a data analysis system for optimizing regional power resource supply according to an embodiment of the present application;

[0044] Figure 6 Fig. 25 shows a schematic structural diagram of an embodiment of an electronic device of the present application. Detailed Embodiments

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0046] In the technical solutions of the present application, for the processing of the collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved, etc., they all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0047] Figure 1 Fig. 37 shows a flowchart of an example of a data analysis method for optimizing regional power resource supply according to an embodiment of the present application.

[0048] Regarding the execution entity of the method according to the embodiments of the present application, it can be any controller or processor with computing or processing capabilities. Through multi-level and multi-model optimization means, the integrated scheduling of the substation and the energy storage station is realized, the efficiency and accuracy of power resource supply and management are improved, the operation cost is significantly reduced, and the stability of the system and the energy utilization efficiency are improved.

[0049] In some examples, it can be a data analysis platform, and can be integrated and configured in an electronic device or a terminal in a software, hardware, or software-hardware combination manner, and the types of the terminal or the electronic device can be diverse, such as mobile phones, tablet computers, or desktop computers, etc.

[0050] As Figure 1 shown, in step S110, the substation operation time series data and the energy storage station operation time series data of the corresponding historical first time period of the target area node to be analyzed are obtained.

[0051] In some embodiments, the data analysis platform can call the system log and query and call the system log according to a preset time period of a past time length (for example, one day or one week) to determine the corresponding substation operation time series data and energy storage station operation time series data.

[0052] Taking examples for combination, the substation operation time series data can include load records (load data on a daily, hourly, or finer-grained basis), peak-valley loads (recording the load peaks and valleys on a daily or hourly basis), operation costs (operation and maintenance costs of the substation in different time periods), equipment status data (health status of substation equipment, fault records, etc.), and historical maintenance records (maintenance time, maintenance content, equipment replacement records, etc.). In addition, the energy storage station operation time series data can include charge and discharge records (charge and discharge data on a daily, hourly, or finer-grained basis), equipment status (health status of energy storage equipment, fault records, etc.), remaining capacity (current remaining power of the energy storage equipment), operation costs (operation and maintenance costs of the energy storage station in different time periods), electricity market prices (time series data of market electricity prices), and weather data (meteorological data related to power demand, such as temperature, humidity, etc.).

[0053] Thus, by obtaining multiple types of data, the operation conditions of the substation and the energy storage station can be comprehensively reflected, the accuracy and reliability of the data are improved, and the rich data types provide a solid foundation for subsequent scheduling optimization and prediction, ensuring that the model can make accurate decisions even in complex situations.

[0054] In step S120, the substation operation time series data is input into the substation scheduling optimization model to determine the preliminary load balancing measures for the corresponding future second time period.

[0055] Here, the substation scheduling optimization model adopts a multi-objective optimization model. For example, it comprehensively considers multiple objectives such as load balance, operating cost, equipment status, and maintenance records, and sets constraint conditions such as the maximum load of equipment and the maintenance interval time. By using appropriate optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) for solution, preliminary load balancing measures can be obtained.

[0056] It should be noted that the load balancing measures refer to adjusting the load distribution of each substation in the power system to ensure the balance of power supply and demand, aiming to optimize the operation stability and reliability of the power system. Exemplarily, the load balancing measures can include various load distribution operations, such as transferring non-critical loads outward, absorbing increased loads, and incentive measures for users to consume electricity.

[0057] In this embodiment, through the multi-objective optimization model, multiple objectives are comprehensively considered to ensure the rationality of the output preliminary load balancing measures.

[0058] In step S130, the operation time series data of the energy storage station and the weather forecast information for the corresponding second time period are input into the energy storage station scheduling optimization model to determine the preliminary charge and discharge operation measures for the corresponding second time period.

[0059] Here, by adopting a time series prediction model and using time series analysis methods for prediction, corresponding charge and discharge measures are obtained, aiming to maximize the utilization rate of energy storage equipment, reduce energy loss, and improve economic benefits. It should be noted that the energy storage station scheduling optimization model can adopt ARIMA, LSTM, or other types of time series prediction models to predict the charge and discharge requirements of the energy storage station.

[0060] It should be noted that the charge and discharge operation measures refer to realizing the power supply and demand balance of the energy storage station through scheduling the charging and discharging operations of the energy storage station in the power system, and optimizing the system operation efficiency and economic benefits. In some examples, the charge and discharge operation measures can include charging operations and corresponding recommended charging powers, discharging operations and corresponding recommended discharging powers to dynamically adjust the charge and discharge operations of the energy storage station.

[0061] In this embodiment, the energy storage station scheduling optimization model comprehensively considers the historical operation data of the energy storage station and the weather forecast information, and uses this to predict the charge and discharge operation measures. Thus, combined with the weather forecast information, the energy storage operation is adjusted in advance to reduce the impact of renewable energy fluctuations on regional power supply and distribution.

[0062] In step S140, the substation operation time series data, the energy storage station operation time series data, the preliminary load balancing measures, and the preliminary charge and discharge operation measures are input into the measure comprehensive optimization model to determine the target load balancing measures and target charge and discharge operation measures for the corresponding second time period.

[0063] In some embodiments, the measure comprehensive optimization model may adopt an ensemble learning model. By comprehensively learning the substation operation data, energy storage station operation data, and the relationship between the substation and the energy storage station, the optimal load balancing and charge-discharge operation measures can be formulated to improve the overall dispatching effect of the power system.

[0064] Figure 2 FIG. shows a schematic structural connection diagram of an example of the measure comprehensive optimization model according to an embodiment of the present application.

[0065] As Figure 2 shown, the measure comprehensive optimization model 200 includes a data preprocessing layer 210, a graph neural network layer 220, a feature fusion layer 230, and a reinforcement learning layer 240.

[0066] The data preprocessing layer 210 is used to preprocess the substation operation time-series data, energy storage station operation time-series data, preliminary load balancing measures, and preliminary charge-discharge operation measures to obtain the corresponding first time-series features, second time-series features, first measure features, and second measure features.

[0067] In some embodiments, the data preprocessing layer 210 scales the data to a unified range for easy model processing, and then extracts key features from the time-series data, such as average load, peak load, etc.

[0068] The first time-series feature and the second time-series feature are input into the graph neural network (Graph Neural Network, GNN) layer 220 to obtain the corresponding substation-dependent feature and energy storage station-dependent feature respectively.

[0069] It should be noted that GNN is suitable for processing network data with complex relationships. In the power system, there are complex dependencies between substations and energy storage stations. GNN can capture these relationships, such as the dependencies between different substations and the charge-discharge relationships between energy storage stations and substations, thereby increasing the information content of the extracted features.

[0070] Figure 3 FIG. shows an example operation flowchart of processing time-series features using the graph neural network layer according to an embodiment of the present application.

[0071] As Figure 3 shown, in step S310, a graph structure is constructed based on the first time-series feature and the second time-series feature.

[0072] Here, the graph structure includes multiple graph nodes and edges. Each graph node is defined by the time-series feature of the corresponding substation or energy storage station, and each edge is defined by the substation and the energy storage station having a power connection relationship.

[0073] In some embodiments, the graph nodes represent substations and energy storage stations in a power system, and their specific node features are defined by corresponding operating time series features, such as load, voltage, operating status, etc. In addition, the edges can be physical or logical connections between substations and energy storage stations. For example, in the same substation area, there can be connections between one substation and one or more energy storage stations. In addition, each edge is also set with a corresponding edge weight w ij , which can be defined by the strength of the relationship between the connected nodes.

[0074] In step S320, based on the graph neural network layer, the feature vectors of each graph node are iteratively updated to gradually fuse the information of neighbor nodes, realize the extraction of global features, and thus obtain the substation dependence features and energy storage station dependence features corresponding to each graph node finally.

[0075] Here, by iteratively updating the feature vectors of nodes through GNN, the information of neighbor nodes is gradually fused to realize the extraction of global features. Specifically, the feature vector x i of each graph node i is initialized, which represents the time series feature of the substation or energy storage station corresponding to graph node i.

[0076] Then, at each layer l, each graph node i receives information from its neighbor nodes N(i) and aggregates it.

[0077] The message passing formula is:

[0078]

[0079] The aggregation formula is:

[0080]

[0081] In the formula, is the message of graph node i at the l+1 layer, is the feature vector of graph node i at the l+1 layer, W (l) and b (l) are the weights and biases at the l layer, and σ is the activation function.

[0082] Through iterative updates by multiple graph neural network layers, the global information is gradually fused to obtain the final node feature vector:

[0083]

[0084] In the formula, L is the number of layers of the graph neural network layer, H i is the final feature vector of graph node i, which is used to determine the corresponding substation dependence feature or energy storage station dependence feature.

[0085] Through the above-mentioned graph structure construction process and GNN feature extraction process, the complex relationship between the substation and the energy storage station can be effectively captured, enriching the information content of the extracted features. Thus, the GNN layer can effectively capture the complex relationship between the substation and the energy storage station, improving the quality of feature extraction and providing more accurate data support for subsequent strategy optimization.

[0086] The feature fusion layer 230 is used to fuse the substation-dependent features, energy storage station-dependent features, first measure features, and second measure features to obtain a comprehensive feature representation:

[0087] H fusion = Concat(H GNN-substation , H GNN-storage , F substation , F storage ), Equation (4)

[0088] In the formula, H fusion represents the comprehensive feature vector, H GNN-substation represents the substation-dependent feature, H GNN-storage represents the energy storage station-dependent feature, F substation represents the first measure feature, F storage represents the second measure feature.

[0089] Here, the features extracted by the GNN are fused with the preliminary measure features to form a comprehensive feature vector for subsequent comprehensive optimization of measures. In this embodiment, by fusing data features from different sources, the model can simultaneously consider the operating states and preliminary strategies of the substation and the energy storage station, improving the effect of comprehensive optimization.

[0090] The reinforcement learning layer 240 is used to process the comprehensive feature representation to determine the corresponding target load balancing measures and target charge and discharge operation measures.

[0091] Here, the reinforcement learning algorithm is used to process the comprehensive feature vector. Since reinforcement learning is suitable for decision-making optimization in dynamic and complex environments and can dynamically adjust strategies, by continuously adjusting the initial measures, the operating cost, stability, and economic benefits of the power system are maximized, and the target load balancing measures and target charge and discharge operation measures are output.

[0092] Specifically, each state in the state space of the reinforcement learning layer 240 is defined by the following formula:

[0093] s = H fusi on, Equation (5)

[0094] In the formula, s represents the state of the reinforcement learning layer, which is defined by the corresponding comprehensive feature representation.

[0095] The policy update formula is:

[0096] π(a|s) = softmax(W π ·s + b π ), Equation (6)

[0097] Wherein, π(a|s) represents the policy of the reinforcement learning layer, which represents the probability of selecting action a in state s; a is defined by the corresponding load adjustment amplitude and charge-discharge operation amplitude; W π and b π respectively represent the policy weight matrix and the policy bias vector.

[0098] The reward function is:[[]]

[0099]

[0100] Wherein, R represents the reward value, P profit represents the overall revenue of the power system, P cost represents the operating cost of the power system; L balanced represents the regional power load after load balance adjustment, L total represents the total load of the power system, E loss represents the energy loss; λ 1 , λ 2 , λ 3 respectively represent the weight coefficients for measuring the economic benefits, stability and efficiency of the system.

[0101] Here, by taking the economic benefits as an important component in the reward function, the model can give priority to the costs and revenues of the power system during the decision-making process. The P profit - P cost in the reward function directly reflects the economic benefits of the power system, prompting the model to maximize the revenue and minimize the cost when selecting load balancing measures and charge-discharge operation suggestions.

[0102] By introducing the index of operating stability into the reward function the model can balance the power load, avoid overloading or underloading situations, encourage the model to be more cautious in the selection of load balancing measures, ensure the balance between power supply and demand, and improve the stability of the system.

[0103] Through the system efficiency index E in the reward function loss it reflects the situation of energy loss, prompting the model to consider the efficiency issues in the energy conversion and transmission processes when selecting charge-discharge operation measures, and improving the overall system efficiency.

[0104] Determine the target load balancing measures and target charge-discharge operation measures according to the action with the maximum corresponding reward value.

[0105] Through a reward function that comprehensively considers economic benefits, operation stability, and system efficiency, the reinforcement learning layer can optimize decisions under multi-objective constraints during the process of updating policy actions, ensuring that optimal load balancing measures and charge-discharge operation measures can be output under different scenarios.

[0106] In the embodiments of this application, the measure comprehensive optimization model combines multi-faceted data and preliminary strategies of the substation and the energy storage station, and can flexibly adjust the dispatching optimization strategy considering multiple factors. By comprehensively considering the operating status of the substation and the energy storage station, the market economic environment, and the external weather environment, and globally optimizing the load balancing measures and charge-discharge measures, it can ensure the balance between power supply and demand, reduce the fluctuations of the power system, and avoid overcharging or over-discharging of the energy storage station, improve the stability of the system, and enhance the security of the system.

[0107] In some examples of the embodiments of this application, the substation dispatching optimization model adopts the Enhanced Particle Swarm Optimization (PSO) algorithm.

[0108] More specifically, initialize K sub-populations, each sub-population contains multiple particles, the dimension of each particle is D, and randomly initialize the initial position of each particle and the initial velocity The position of the particle is defined by the load distribution value of the substation, and the velocity of the particle represents the movement rate and direction of the particle in the search space. Here, the update of the velocity determines how the particle moves from the current position to a new position, thereby updating its load distribution measure, and then obtaining the best load distribution strategy based on the optimal particle position obtained by PSO optimization calculation.

[0109] Next, the details of the specific PSO optimization calculation process will be described in detail. First, set the maximum number of iterations T max and the frequency of information exchange between sub-populations T exchange , and perform independent PSO searches on each sub-population to update the velocity and position of each particle in the sub-population.

[0110] In this embodiment, by dividing the particle swarm into multiple sub-populations, each sub-population independently searches in different search spaces. By regularly exchanging information between sub-populations, the search space can be explored more comprehensively, avoiding falling into local optimal solutions.

[0111] More specifically, the particle velocity update formula for the global search stage is:

[0112] v i,j (t + 1) = ω·v i,j (t) + η·(p i,j (t) - x i,j(t)) + μ·(x rand,j (t) - x i,j (t)), Equation (8)

[0113]

[0114] In the formula, ω is the inertia weight, w max and w min are the initial and minimum inertia weights, t is the current iteration number, T max is the maximum iteration number; v i,j (t) represents the value of the velocity of the i-th particle in the j-th dimension at time t, v i,j (t + 1) represents the updated value of the velocity of the i-th particle in the j-th dimension at time t + 1; η represents the first mutation factor, which is used to increase the diversity of particles and help escape from local optima; p i,j (t) is the value of the historical best position of the i-th particle in the j-th dimension; x i,j (t) represents the value of the current position of the i-th particle in the j-th dimension at time t; μ represents the second mutation factor, which is used to introduce a random component and further increase the diversity of particles; x rand,j (t) represents the value of the position of a randomly selected particle in the j-th dimension at time t.

[0115] Here, according to the search stage and performance of the particles, the inertia weight and other parameters are dynamically adjusted, so that the algorithm can adapt to different search environments and optimization requirements, improving the robustness and adaptability of the algorithm, so that the algorithm can perform more global exploration in the early stage of the search and more local search in the later stage. Thus, in the face of sudden load changes and uncertainty factors, the algorithm can quickly adjust the load distribution strategy to ensure the stable operation and optimization effect of the system.

[0116] The formula for updating the particle velocity for the local search stage is:

[0117] v i,j (t + 1) = α·v i,j (t) + β·(p i,j (t) - x i,j (t)) + γ·(g j (t) - x i,j (t)), Equation (10)

[0118] Wherein, α, β, and γ are convergence factors, which respectively control the proportions of the inertia weight, individual cognition, and social cognition; in the formula, α represents the first convergence factor, which is used to control the inertia weight and affects the retention ratio of the particle velocity; β represents the second convergence factor, which is used to control the proportion of individual cognition and affects the role of the particle's own historical optimal position; γ represents the third convergence factor, which is used to control the proportion of social cognition and affects the role of the particle swarm's historical optimal position; g j (t) represents the value of the global optimal position in the sub-swarm at the t-th moment in the j-th dimension. Here, the individual cognition and social cognition parameters control the degree to which the particle is affected by its own optimal position and the global optimal position of the swarm when updating the velocity, and these parameters can be set in the system in advance.

[0119] In this way, in the global search stage, through the mutation velocity update mechanism, the randomness and diversity of the particles are increased to ensure finding a better solution in a wide search space. In the local search stage, through the convergence velocity update mechanism, the search space is reduced, and efforts are concentrated on fine-tuning near the optimal solution to improve the optimization accuracy.

[0120] In this embodiment, multiple velocity update mechanisms are used, and the mutation velocity update formula and the convergence velocity update formula are introduced, which are respectively used in the global search stage and the local search stage. The diversity of the particles is increased through the mutation velocity update to ensure the global search ability, and the search accuracy of the particles is improved through the convergence velocity update to ensure the local optimization effect. For example, in substation scheduling, through an accurate load distribution strategy, the power supply and demand relationship can be better balanced, the peak load pressure can be reduced, and the reliability and stability of the system can be improved.

[0121] Exchange the optimal information between sub-swarms regularly, and perform an exchange every T exchange iterations;

[0122] Share the optimal solutions of each sub-swarm and update the global optimal position of each sub-swarm;

[0123]

[0124] In the formula, G j (t) represents the value of the global optimal position in all sub-swarms at the t-th moment in the j-th dimension; k represents the index of the sub-swarm, and its value range is from 1 to K, where K is the total number of sub-swarms; represents the value of the global optimal position in the k-th sub-swarm at the t-th moment in the j-th dimension; f is the fitness function, which is used to evaluate the quality of the particles.

[0125] Update the global optimal position within each sub-swarm, update the global optimal position after information exchange, and determine the preliminary load balancing measures based on the updated global optimal position.

[0126] In this embodiment, by initializing multiple sub-populations, each sub-population conducts searches independently and exchanges information among sub-populations regularly, ensuring that each sub-population can utilize the optimal solutions of other sub-populations, thereby enhancing the global search ability and optimization efficiency. Thus, in the substation scheduling optimization, multi-population collaborative search can ensure finding the globally optimal load distribution strategy within different time periods and avoid resource waste and low operation efficiency caused by local optimal solutions.

[0127] Through the embodiment of this application, the enhanced PSO algorithm adopting multi-population collaborative search and multiple speed update mechanisms can significantly enhance the global search ability and optimization efficiency of the substation scheduling optimization model, enhance the accuracy and reliability of the load distribution strategy, improve the robustness and adaptability of the algorithm, and optimize the economic benefits and operation efficiency of the power system.

[0128] Figure 4 Fig. shows a schematic structural connection diagram of an example of the energy storage station scheduling optimization model according to the embodiment of this application.

[0129] As Figure 4 shown, the energy storage station scheduling optimization model 400 includes a cascaded input layer 410, a convolutional layer 420, an LSTM layer 430, an attention layer 440, and an output layer 450.

[0130] The input layer 410 is used to receive the operation time series data of the energy storage station and weather forecast information.

[0131] The convolutional layer 420 performs a convolution operation on the input time series data using multi-scale convolution kernels to extract corresponding multi-scale convolution features.

[0132]

[0133] In the formula, H conv represents the multi-scale convolution features output by the convolutional layer, and respectively represent the weight matrix and bias term of the convolution kernel with scale s; S represents the set of convolution kernel scales; X represents the input data matrix determined based on the operation time series data of the energy storage station and weather forecast information.

[0134] By extracting the local features of the time series data through the multi-scale convolutional layer, the features of different time windows are captured, improving the model's understanding and prediction ability for complex time series data. In addition, the model can be more efficient in feature extraction, reducing the computational burden of the LSTM layer.

[0135] The LSTM layer 430 is used to process the multi-scale convolution features to obtain corresponding time series modeling convolution features:

[0136] H LSTM = LSTM(Hconv ), Equation (13)

[0137] where H LSTM represents the temporal modeling convolution features output by the LSTM layer;

[0138] The features output by the convolutional layer are processed by the LSTM layer for temporal modeling, effectively capturing the time-dependent relationships and non-linear features in the data.

[0139] The attention layer 440 is used to process the temporal modeling convolution features using the multi-head attention mechanism to obtain the corresponding attention context vector;

[0140] e t = Concat(head 1 ,..., head h ), Equation (14)

[0141]

[0142] where e t represents the attention score at time t, and represent the weight matrix and bias term of the i-th attention head respectively; α t represents the attention weight at time t, and c represents the attention context vector.

[0143] Through the multi-head attention mechanism, the attention layer can dynamically adjust the focus, capture important time points and patterns in the temporal data, improve the ability to capture important features in the temporal data, and further improve the accuracy of the prediction results. In addition, through the multi-head attention mechanism, the parallel computing ability of the model is improved, the impact of long sequence dependencies on the computing efficiency is reduced, and the real-time performance and efficiency of policy optimization are improved.

[0144] The output layer 450 is used to predict the initial charge and discharge operation measures corresponding to the second time period according to the attention context vector:

[0145]

[0146] where represents the charge and discharge decision value; if then the charging operation is executed; if then the discharging operation is executed; W decision and b decision represent the weight matrix and bias term of the decision fully connected layer respectively; represents the charge and discharge policy value, W strategy represents the slope of the policy value conversion, and b strategy represents the intercept of the policy value conversion.

[0147] Based on the above design of the output layer 450, the charge-discharge strategy value not only determines whether the energy storage station conducts charging or discharging operations, but also can precisely control the power of charging and discharging, which directly corresponds to the depth of charge and discharge, making the charge-discharge control measures for the energy storage station clearer, easier to execute, and meeting the actual requirements. In addition, through precise charge-discharge operations, the operating costs and benefits of the energy storage station can be optimized, maximizing economic benefits.

[0148] Through the embodiments of the present application, in the actual scheduling of the energy storage station, according to real-time input data and context information, the model can dynamically adjust the charge-discharge strategy, ensure the accuracy and adaptability of operations, precisely adjust the charge-discharge power for each time period, avoid overcharging or over-discharging, and improve the lifespan and economic benefits of the energy storage device.

[0149] In some examples of the embodiments of the present application, the loss function L of the energy storage station scheduling optimization model is:

[0150] L = λ 1 ·L economic + λ 2 ·L stability + λ 3 ·L efficiency , Equation (20)

[0151] In the formula, L economic represents the economic benefit loss term, L stability represents the operation stability loss term, L efficiency represents the system efficiency loss term; λ 1 , λ 2 , λ 3 respectively represent the corresponding weight coefficients.

[0152]

[0153] In the formula, N represents the number of samples in the data sample set, D profit,i represents the power supply income of the energy storage station corresponding to the i-th sample, D cost,i represents the power supply and distribution operation cost of the energy storage station corresponding to the i-th sample.

[0154]

[0155] In the formula, L balanced,i represents the regional power load after load balancing corresponding to the i-th sample, L lim,i represents the extreme value of the regional power load corresponding to the i-th sample.

[0156]

[0157] In the formula, E loss,i represents the regional power supply energy loss value corresponding to the i-th sample.

[0158] Based on the above loss function design of the embodiments of the present application, by maximizing the revenue and minimizing the cost, the model can optimize the charge and discharge strategies of the energy storage station under different electricity price conditions, effectively control the operating cost, select the best charge and discharge time and strategy, reduce the electricity procurement cost. For example, the energy storage station discharges during the peak electricity price period and charges during the valley electricity price period, so as to maximize the revenue. By balancing the power load, avoiding regional power overload or underload, and improving the stability and reliability of regional power supply and distribution. For example, the energy storage station shares the load of the regional or connected substation to prevent regional power supply and distribution overload. By reducing the energy loss term, the energy storage device operates in the optimal state, improving the charge and discharge efficiency. For example, it charges efficiently under sufficient light conditions.

[0159] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of actions combined. However, those skilled in the art should know that the present application is not limited by the described action sequence, because according to the present application, certain steps can be in other sequences or performed simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application. In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0160] Figure 5 The structural block diagram of an example of a data analysis system for optimizing the supply of regional power resources according to an embodiment of the present application is shown.

[0161] As Figure 5 shown, the data analysis system 500 for optimizing the supply of regional power resources includes a data acquisition unit 510, a first measure determination unit 520, a second measure determination unit 530, and a comprehensive measure optimization unit 540.

[0162] The data acquisition unit 510 is used to acquire the substation operation time series data and the energy storage station operation time series data of the corresponding historical first time period of the target regional node to be analyzed; the data types of the substation operation time series data include: load records, peak-valley loads, operating costs, equipment status data, and historical maintenance records; the data types of the energy storage station operation time series data include: charge and discharge records, equipment status, remaining capacity, operating costs, electricity market prices, and weather data.

[0163] The first measure determination unit 520 is used to input the substation operation time series data into the substation scheduling optimization model to determine the preliminary load balancing measures for the corresponding future second time period; the substation scheduling optimization model adopts a multi-objective optimization model;

[0164] The second measure determination unit 530 is configured to input the operation time series data of the energy storage station and the weather forecast information corresponding to the second time period into the energy storage station scheduling optimization model to determine the preliminary charge and discharge operation measures corresponding to the second time period; the energy storage station scheduling optimization model adopts a time series prediction model.

[0165] The comprehensive measure optimization unit 540 is configured to input the operation time series data of the substation, the operation time series data of the energy storage station, the preliminary load balancing measure, and the preliminary charge and discharge operation measure into the measure comprehensive optimization model to determine the target load balancing measure and the target charge and discharge operation measure corresponding to the second time period.

[0166] The measure comprehensive optimization model includes a data preprocessing layer, a graph neural network layer, a feature fusion layer, and a reinforcement learning layer;

[0167] The data preprocessing layer is configured to preprocess the operation time series data of the substation, the operation time series data of the energy storage station, the preliminary load balancing measure, and the preliminary charge and discharge operation measure to obtain corresponding first time series features, second time series features, first measure features, and second measure features;

[0168] Input the first time series feature and the second time series feature into the graph neural network layer to obtain corresponding substation dependence features and energy storage station dependence features respectively, specifically including:

[0169] Based on the first time series feature and the second time series feature, construct a graph structure; the graph structure includes multiple graph nodes and edges; each graph node is defined by the operation data of the corresponding substation or energy storage station, and each edge is defined by the substation and the energy storage station with an electrical connection relationship;

[0170] Based on the graph neural network layer, iteratively update the feature vectors of each graph node to gradually fuse the information of neighbor nodes, realize the extraction of global features, and thus obtain the substation dependence features and energy storage station dependence features corresponding to each final graph node:

[0171] Initialize the feature vector h of each graph node i i (0) = x i ; x i represents the time series feature of the substation or energy storage station corresponding to the graph node i;

[0172] At each layer l, each graph node i receives information from its neighbor nodes N(i) and aggregates them:

[0173]

[0174] where is the message of graph node i at the (l + 1)-th layer, w ij represents the edge weight corresponding to the edge connecting graph node i and graph node j, is the feature vector of graph node i at the (l + 1)-th layer, W (l) and b (l) are the weight and bias at the l-th layer, and σ is the activation function;

[0175] It is iteratively updated through multiple graph neural network layers to gradually fuse the global information to obtain the final node feature vector:

[0176]

[0177] In the formula, L is the number of layers of the graph neural network layer, H i is the final feature vector of graph node i, which is used to determine the corresponding substation-dependent feature or energy storage station-dependent feature;

[0178] The feature fusion layer is used to fuse the substation-dependent feature, the energy storage station-dependent feature, the first measure feature, and the second measure feature to obtain a comprehensive feature representation:

[0179] H fusion = Concat(H GNN-substation , H GNN-storage , F substation , F storage )

[0180] In the formula, H fusion represents the comprehensive feature vector, H GNN-substation represents the substation-dependent feature, H GNN-storage represents the energy storage station-dependent feature, F substation represents the first measure feature, F storage represents the second measure feature;

[0181] Based on the reinforcement learning layer to process the comprehensive feature representation to determine the corresponding target load balancing measure and target charge and discharge operation measure, specifically including:

[0182] Each state in the state space of the reinforcement learning layer is defined by the following formula:

[0183] s = H fusion

[0184] In the formula, s represents the state of the reinforcement learning layer, which is defined by the corresponding comprehensive feature representation;

[0185] The policy update formula is:

[0186] π(a|s) = softmax(W π ·s + bπ )

[0187] where π(a|s) represents the policy of the reinforcement learning layer, which represents the probability of selecting action a in state s; a is defined by the corresponding load adjustment amplitude and charge-discharge operation amplitude; W π and b π represent the policy weight matrix and the policy bias vector respectively;

[0188] The reward function is:

[0189]

[0190] where R represents the reward value, P profit represents the overall revenue of the power system, P cost represents the operating cost of the power system; L balanced represents the regional power load after load balance adjustment, L total represents the total load of the power system, E loss represents the energy loss; λ 1 , λ 2 , λ 3 represent the weight coefficients for measuring the economic benefits, stability, and efficiency of the system respectively;

[0191] Determine the target load balance measure and the target charge-discharge operation measure according to the action with the maximum corresponding reward value.

[0192] In some embodiments, the embodiments of the present application provide a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to perform the steps of any one of the above data analysis methods for optimizing the supply of regional power resources.

[0193] In some embodiments, the embodiments of the present application further provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is enabled to execute the steps of any one of the above data analysis methods for optimizing the supply of regional power resources.

[0194] In some embodiments, the embodiments of the present application further provide an electronic device, which 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 to enable the at least one processor to execute the steps of the data analysis method for optimizing the supply of regional power resources.

[0195] Figure 6 FIG. is a schematic hardware structure diagram of an electronic device that executes a data analysis method for optimizing the supply of regional power resources according to another embodiment of the present application. As shown in Figure 6 shown, the device includes:

[0196] One or more processors 610 and a memory 620. Figure 6 Taking one processor 610 as an example.

[0197] The device that executes the data analysis method for optimizing the supply of regional power resources may further include: an input device 630 and an output device 640.

[0198] The processor 610, the memory 620, the input device 630, and the output device 640 may be connected through a bus or other means. Figure 6 Taking connection through a bus as an example.

[0199] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the data analysis method for optimizing the supply of regional power resources in the embodiments of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, that is, implements the data analysis method for optimizing the supply of regional power resources in the above method embodiments.

[0200] The memory 620 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the electronic device. In addition, the memory 620 may include a high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 620 may optionally include a memory remotely provided with respect to the processor 610, and these remote memories can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0201] The input device 630 can receive input digital or character information, and generate signals related to the user settings and function control of the electronic device. The output device 640 may include a display device such as a display screen.

[0202] The one or more modules are stored in the memory 620 and, when executed by the one or more processors 610, perform the data analysis method for optimizing the supply of regional power resources in any of the above method embodiments.

[0203] The above product can execute the method provided by the embodiments of the present application, and has the corresponding functional modules and beneficial effects for executing the method. For the technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiments of the present application.

[0204] The electronic devices in the embodiments of the present application exist in various forms, including but not limited to:

[0205] (1) Mobile communication devices: These devices are characterized by having mobile communication functions and mainly aim to provide voice and data communication. Such terminals include: smart phones, multimedia phones, functional phones, and low-end phones, etc.

[0206] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have the characteristic of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, etc.

[0207] (3) Portable entertainment devices: These devices can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, and smart toys and portable vehicle navigation devices.

[0208] (4) Other airborne electronic devices with data interaction functions, such as in-vehicle device installed in a vehicle.

[0209] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0210] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, also by hardware. Based on such an understanding, the essence of the above technical solutions or the part that contributes to the related technologies can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0211] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than limiting them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A data analysis method for optimizing regional power resource supply, comprising: Obtaining substation operation timing data and energy storage station operation timing data corresponding to the first historical time period of the target area node to be analyzed; The data types of the substation operation time series data include: load records, peak and valley loads, operation costs, equipment status data and historical maintenance records; the data types of the energy storage station operation time series data include: charge and discharge records, equipment status, remaining capacity, operation costs, electricity market prices and weather data; Inputting the substation operation timing data into a substation dispatch optimization model to determine preliminary load balancing measures corresponding to a second time period in the future; the substation dispatch optimization model adopts a multi-objective optimization model; Inputting the energy storage station operation timing data and the weather forecast information corresponding to the second time period into the energy storage station scheduling optimization model to determine preliminary charging and discharging operation measures corresponding to the second time period; The energy storage station scheduling optimization model adopts a time series prediction model; Inputting the substation operation timing data, the energy storage station operation timing data, the preliminary load balancing measures and the preliminary charge-discharge operation measures into a comprehensive optimization model to determine a target load balancing measure and a target charge-discharge operation measure corresponding to the second time period; The comprehensive optimization model of the measures includes a data preprocessing layer, a graph neural network layer, a feature fusion layer and a reinforcement learning layer; The data preprocessing layer is used to preprocess the substation operation timing data, the energy storage station operation timing data, the preliminary load balancing measures and the preliminary charging and discharging operation measures to obtain corresponding first timing characteristics, second timing characteristics, first measure characteristics and second measure characteristics; The first time series feature and the second time series feature are input into the graph neural network layer to obtain corresponding substation dependency features and energy storage station dependency features, respectively, including: Based on the first time series feature and the second time series feature, a graph structure is constructed; the graph structure includes a plurality of graph nodes and edges; each of the graph nodes is defined by the time series feature of the corresponding substation or energy storage station, and each of the edges is defined by the substation and energy storage station having a power connection relationship; Based on the graph neural network layer, the feature vector of each graph node is iteratively updated to gradually integrate the information of neighboring nodes and extract global features, thereby obtaining the substation dependency features and energy storage station dependency features corresponding to each graph node: Initialize the feature vector of each graph node i x i Represents the time series characteristics of the substation or energy storage station corresponding to the graph node i; At each level l, each graph node i is derived from its neighbor nodes Receive information and aggregate it: In the formula, is the message of graph node i at layer l+1, w ij represents the edge weight corresponding to the edge connecting graph node i and graph node j, is the feature vector of graph node i at layer l+1, W (l) and b (l) is the weight and bias of the lth layer, σ is the activation function; Iterative updates are performed through multi-layer graph neural network layers, gradually integrating global information to obtain the final node feature vector: Where L is the number of graph neural network layers, H i is the final feature vector of graph node i, which is used to determine the corresponding substation dependency feature or energy storage station dependency feature; The feature fusion layer is used to fuse the substation dependency feature, the energy storage station dependency feature, the first measure feature, and the second measure feature to obtain a comprehensive feature representation: H fusion =Concat(H GNN-substation ,H GNN-storage ,F substation ,F storage ) In the formula, H fusion represents the comprehensive feature vector, H GNN-substation represents the substation dependency characteristics, H GNN-storage represents the energy storage station dependency characteristics, F substation represents the first measure feature, F storage represents the second measure characteristic; Processing the comprehensive feature representation based on the reinforcement learning layer to determine corresponding target load balancing measures and target charging and discharging operation measures specifically includes: Each state in the state space of the reinforcement learning layer is defined by the following formula: s=H fusion Where s represents the state of the reinforcement learning layer, which is defined by the corresponding comprehensive feature representation; The strategy update formula is: π(a|s)=softmax(W π ·s+b π ) Where π(a|s) represents the strategy of the reinforcement learning layer, which indicates the probability of selecting action a in state s; a is defined by the corresponding load adjustment amplitude and charge / discharge operation amplitude; W π and b π Represent the strategy weight matrix and strategy bias vector respectively; The reward function is: In the formula, R represents the reward value, P profit represents the overall benefit of the power system, P cost Represents the operating cost of the power system; L balanced represents the regional power load after load balancing adjustment, L total Represents the total load of the power system, E loss represents energy loss; λ1, λ2, λ3 represent weight coefficients used to measure system economic benefits, system stability and system efficiency respectively; According to the action with the largest corresponding reward value, the target load balancing measure and the target charging and discharging operation measure are determined.

2. The method according to claim 1, wherein: The substation dispatch optimization model adopts the enhanced particle swarm optimization PSO algorithm and is used to determine the load balancing measures in the following ways: Initialize K sub-populations, each of which contains multiple particles, the dimension of each particle is D, and randomly initialize the initial position of each particle and initial velocity The position of the particle is defined by the load distribution value of the substation, and the particle's velocity represents the moving speed and direction of the particle in the search space; Set the maximum number of iterations T max and the frequency of information exchange between subgroups T exchange , and perform independent PSO searches for each sub-population to update the velocities and positions of each particle in the sub-population: The particle velocity update formula used in the global search phase is: v u,q (t+1)=ω·v u,q (t)+η·(p u,q (t)-x u,q (t))+μ·(x rand,q (t)-x u,q (t)) Where ω is the inertia weight, w max and w min is the initial and minimum inertia weight, t is the current iteration number, T max is the maximum number of iterations; v u,q (t) represents the velocity of the u-th particle in the q-th dimension at time t, v u,q (t+1) represents the updated value of the velocity of the uth particle in the qth dimension at time t+1; η represents the first variation factor, which is used to increase the diversity of particles and help escape from the local optimum; p u,q (t) is the value of the historical optimal position of the u-th particle in the q-th dimension; x u,q (t) represents the value of the current position of the u-th particle in the q-th dimension at time t; μ represents the second variation factor, which is used to introduce random components to further increase the diversity of particles; x rand,q (t) represents the value of the randomly selected particle position at time t in the qth dimension; The particle velocity update formula used in the local search phase is: v u,q (t+1)=α·v u,q (t)+β·(p u,q (t)-x u,q (t))+γ·(g q (t)-x u,q (t)) In the formula, α, β, and γ are convergence factors, which control the ratio of inertia weight, individual cognition, and social cognition respectively; in the formula, α represents the first convergence factor, which is used to control the inertia weight and affect the ratio of particle velocity maintenance; β represents the second convergence factor, which is used to control the ratio of individual cognition and affect the role of the particle's own historical optimal position; γ represents the third convergence factor, which is used to control the ratio of social cognition and affect the role of the particle group's historical optimal position; g q (t) represents the value of the global optimal position in the sub-population at time t in the qth dimension; Regularly exchange optimal information between subgroups, every T exchange Perform one swap per iteration; Share the optimal solutions of each subgroup and update the global optimal position of each subgroup; In the formula, G q (t) represents the value of the global optimal position of all sub-populations in the qth dimension at time t; k represents the index of the sub-population, ranging from 1 to K, where K is the total number of sub-populations; represents the value of the global optimal position in the kth subpopulation at time t in the qth dimension; f is the fitness function, which is used to evaluate the quality of particles; The global optimal position is updated within each sub-group, and the global optimal position is updated after information exchange, and a preliminary load balancing measure is determined based on the updated global optimal position.

3. The method according to claim 1, wherein: The energy storage station scheduling optimization model includes a cascaded input layer, a convolutional layer, an LSTM layer, an attention layer and an output layer; The input layer is used to receive the energy storage station operation timing data and the weather forecast information; The convolution layer uses a multi-scale convolution kernel to perform a convolution operation on the input time series data and extract the corresponding multi-scale convolution features: In the formula, H conv Represents the multi-scale convolution features output by the convolution layer, and They respectively represent the weight matrix and bias term of the convolution kernel with a scale of s; S represents the convolution kernel scale set; X represents the input data matrix determined based on the operating time data of the energy storage station and the weather forecast information; The LSTM layer is used to process the multi-scale convolutional features to obtain corresponding time series modeling convolutional features: H LSTM =LSTM(H conv ) In the formula, H LSTM Represents the temporal modeling convolutional features output by the LSTM layer; The attention layer is used to process the temporal modeling convolutional features using a multi-head attention mechanism to obtain a corresponding attention context vector; e t =Concat(head1,…,head h ) In the formula, e t represents the attention score at time t, and Represent the weight matrix and bias term of the i-th attention head respectively; α t represents the attention weight at time t, and c represents the attention context vector; The output layer is used to predict the initial charging and discharging operation measures corresponding to the second time period according to the attention context vector: In the formula, Indicates the charge and discharge decision value; if Then perform charging operation; if Then the discharge operation is performed; W decision and b decision Represent the weight matrix and bias term of the decision fully connected layer respectively; Indicates the charge and discharge strategy value, W strategy represents the strategy value conversion slope, and b strategy represents the strategy value transformation intercept.

4. The method according to claim 3, wherein: The loss function of the energy storage station scheduling optimization model is for: In the formula, represents the economic benefit loss item, represents the operational stability loss term, represents the system efficiency loss term; Respectively represent the corresponding weight coefficients; In the formula, N represents the number of samples in the data sample set, D profit,i represents the power supply revenue of the energy storage station corresponding to the i-th sample, D cost,i represents the power supply and distribution operation cost of the energy storage station corresponding to the i-th sample; Where, L balanced,i represents the regional power load after load balancing corresponding to the i-th sample, L lim,i Indicates the regional power load extreme value corresponding to the i-th sample; In the formula, E loss,i Indicates the regional power supply energy loss value corresponding to the i-th sample.

5. A data analysis system for optimizing regional power resource supply, comprising: A data acquisition unit, used to acquire substation operation time sequence data and energy storage station operation time sequence data corresponding to a first time period of the history of the target area node to be analyzed; The data types of the substation operation time series data include: load records, peak and valley loads, operation costs, equipment status data and historical maintenance records; the data types of the energy storage station operation time series data include: charge and discharge records, equipment status, remaining capacity, operation costs, electricity market prices and weather data; A first measure determination unit, configured to input the substation operation timing data into a substation dispatch optimization model to determine a preliminary load balancing measure corresponding to a future second time period; The substation dispatch optimization model adopts a multi-objective optimization model; A second measure determination unit, configured to input the energy storage station operation timing data and the weather forecast information corresponding to the second time period into the energy storage station scheduling optimization model to determine preliminary charging and discharging operation measures corresponding to the second time period; The energy storage station scheduling optimization model adopts a time series prediction model; A comprehensive measures optimization unit, used for inputting the substation operation timing data, the energy storage station operation timing data, the preliminary load balancing measures and the preliminary charge-discharge operation measures into a comprehensive measures optimization model to determine a target load balancing measure and a target charge-discharge operation measure corresponding to the second time period; The comprehensive optimization model of the measures includes a data preprocessing layer, a graph neural network layer, a feature fusion layer and a reinforcement learning layer; The data preprocessing layer is used to preprocess the substation operation timing data, the energy storage station operation timing data, the preliminary load balancing measures and the preliminary charging and discharging operation measures to obtain corresponding first timing characteristics, second timing characteristics, first measure characteristics and second measure characteristics; The first time series feature and the second time series feature are input into the graph neural network layer to obtain corresponding substation dependency features and energy storage station dependency features, respectively, including: Based on the first time series feature and the second time series feature, a graph structure is constructed; the graph structure includes a plurality of graph nodes and edges; each of the graph nodes is defined by the operation data of the corresponding substation or energy storage station, and each of the edges is defined by the substation and energy storage station having a power connection relationship; Based on the graph neural network layer, the feature vector of each graph node is iteratively updated to gradually integrate the information of neighboring nodes and extract global features, thereby obtaining the substation dependency features and energy storage station dependency features corresponding to each graph node: Initialize the feature vector of each graph node i x i Represents the time series characteristics of the substation or energy storage station corresponding to the graph node i; At each level l, each graph node i is derived from its neighbor nodes Receive information and aggregate it: In the formula, is the message of graph node i at layer l+1, w ij represents the edge weight corresponding to the edge connecting graph node i and graph node j, is the feature vector of graph node i at layer l+1, W (l) and b (l) is the weight and bias of the lth layer, σ is the activation function; Iterative updates are performed through multi-layer graph neural network layers, gradually integrating global information to obtain the final node feature vector: Where L is the number of graph neural network layers, H i is the final feature vector of graph node i, which is used to determine the corresponding substation dependency feature or energy storage station dependency feature; The feature fusion layer is used to fuse the substation dependency feature, the energy storage station dependency feature, the first measure feature, and the second measure feature to obtain a comprehensive feature representation: H fusion =Concat(H GNN-substation ,H GNN-storage ,F substation ,F storage ) In the formula, H fusion represents the comprehensive feature vector, H GNN-substation represents the substation dependency characteristics, H GNN-storage represents the energy storage station dependency characteristics, F substation represents the first measure feature, F storage represents the second measure characteristic; Processing the comprehensive feature representation based on the reinforcement learning layer to determine corresponding target load balancing measures and target charging and discharging operation measures specifically includes: Each state in the state space of the reinforcement learning layer is defined by the following formula: s=H fusion Where s represents the state of the reinforcement learning layer, which is defined by the corresponding comprehensive feature representation; The strategy update formula is: π(a|s)=softmax(W π ·s+b π ) Where π(a|s) represents the strategy of the reinforcement learning layer, which indicates the probability of selecting action a in state s; a is defined by the corresponding load adjustment amplitude and charge / discharge operation amplitude; W π and b π Represent the strategy weight matrix and strategy bias vector respectively; The reward function is: In the formula, R represents the reward value, P profit represents the overall benefit of the power system, P cost Represents the operating cost of the power system; L balanced represents the regional power load after load balancing adjustment, L total Represents the total load of the power system, E loss represents energy loss; λ1, λ2, λ3 represent weight coefficients used to measure system economic benefits, system stability and system efficiency respectively; According to the action with the largest corresponding reward value, the target load balancing measure and the target charging and discharging operation measure are determined.

Images