Method and equipment for processing low-voltage protection constant value in regional power grid and medium

Abstract

The embodiment of the invention provides a method and equipment for processing a low-voltage protection constant value in a regional power grid, and a medium. The method comprises the steps of determining a dynamic topological feature vector of power grid equipment in a preset time window, dynamic topological feature data corresponding to the dynamic topological feature vector and a contact relation matrix representing the connection tightness degree between the power grid equipment according to the connection relation and feature data of the power grid equipment in a power grid at different time points; inputting the contact relation matrix, the operation parameters in the power grid and the historical fault data into a generative adversarial network to obtain simulated fault data output by the generative adversarial network; according to the extended training data set, an initial target constant value solution set is obtained through multi-target optimization solution; the extended training data set is obtained according to the simulated fault data and the historical fault data; and performing compliance verification on the initial target constant value solution set to obtain a target constant value solution set. The method is used for improving the accuracy of obtaining the constant value, so that the operation effect of the power grid during operation is improved.

Description

Technical Field

[0001] This application relates to the field of power grid protection technology, and in particular to a method, equipment and medium for processing low-voltage protection settings in a regional power grid. Background Technology

[0002] Currently, in power systems, the calculation of low-voltage protection settings for regional power grids is a crucial component of power system relay protection. Its purpose is to configure reasonable voltage protection action thresholds for transformers, lines, and electrical equipment in the distribution network to address abnormal voltage drops caused by short-circuit faults, heavy load starts, or insufficient reactive power. This calculation requires comprehensive consideration of the power grid structure, operating mode, load characteristics, and the voltage level of the upstream system. While ensuring sensitivity and selectivity, it rationally sets the action values ​​and delays for low-voltage protection to avoid false trips or failures to operate, ensuring power supply continuity and equipment safety. Especially with the increasing integration of distributed power sources, it is also necessary to adapt to the complexity of protection coordination brought about by bidirectional power flow changes, enhancing the proactive defense capabilities and self-healing level of the regional power grid.

[0003] Currently, most low-voltage protection setting calculations rely on static power grid models and historical data, with settings determined through expert experience or pre-defined rules. These methods typically include steps such as data acquisition, topology analysis, setting calculation, and verification.

[0004] However, existing methods are unable to reflect the dynamic changes of the power grid in real time, resulting in insufficient accuracy of the determined values, which are difficult to match the current needs of the power grid. Summary of the Invention

[0005] The method, equipment, and medium for processing low-voltage protection settings in a regional power grid provided in this application embodiment are used to improve the accuracy of obtaining settings, thereby improving the operating performance of the power grid during operation.

[0006] In a first aspect, embodiments of this application provide a method for processing low-voltage protection settings within a regional power grid, including:

[0007] Based on the connection relationships and characteristic data of power grid equipment at different points in time, the dynamic topology feature vectors of power grid equipment within a preset time window, the dynamic topology feature data corresponding to the dynamic topology feature vectors, and the connection relationship matrix characterizing the degree of connection between power grid equipment are determined; among them, the dynamic topology feature data is obtained by tensor decomposition analysis of the dynamic topology feature vectors.

[0008] The connection matrix, the operating parameters of the power grid, and historical fault data are input into the generative adversarial network to obtain the simulated fault data output by the generative adversarial network.

[0009] According to the extended training data set, an initial target setting solution set is obtained by multi-objective optimization solution; the extended training data set is obtained according to the simulation fault data and the historical fault data;

[0010] The initial target setting solution set is subjected to compliance verification to obtain a target setting solution set, which is used to adjust the operation parameters of the power grid equipment according to the operation state of the power grid equipment in the power grid.

[0011] In the second aspect, the embodiments of the present application provide a processing device for low-voltage protection setting values in a regional power grid, which comprises:

[0012] A determination module is configured to determine, according to the connection relationship and the characteristic data of the power grid equipment in the power grid at different time points, a dynamic topology characteristic vector of the power grid equipment within a preset time window, dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, and a tie-line relationship matrix representing the connection closeness between the power grid equipment; wherein the dynamic topology characteristic data is obtained by tensor decomposition analysis of the dynamic topology characteristic vector;

[0013] A obtaining module is configured to input the tie-line relationship matrix, the operation parameters in the power grid and the historical fault data into a generative adversarial network to obtain simulation fault data output by the generative adversarial network;

[0014] A multi-objective optimization module is configured to obtain an initial target setting solution set by multi-objective optimization solution according to an extended training data set; the extended training data set is obtained according to the simulation fault data and the historical fault data;

[0015] A compliance verification module is configured to perform compliance verification on the initial target setting solution set to obtain a target setting solution set, which is used to adjust the operation parameters of the power grid equipment according to the operation state of the power grid equipment in the power grid.

[0016] In the third aspect, the embodiments of the present application provide an electronic device, which comprises a memory and a processor.

[0017] The memory stores computer execution instructions;

[0018] The processor executes the computer execution instructions stored in the memory, so that the processor executes the first aspect and / or various possible implementation manners of the first aspect.

[0019] In the fourth aspect, the embodiments of the present application provide a computer readable storage medium, which stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the first aspect and / or various possible implementation manners of the first aspect.

[0020] In a fifth aspect, an embodiment of the present application provides a computer program product, comprising a computer program which, when executed by a processor, implements the first aspect and / or various possible implementation manners of the first aspect.

[0021] The method, device and medium provided by the embodiments of the present application for processing low-voltage protection setting values in a regional power grid, by collecting connection relationships and operation characteristic data of power grid equipment at different time points in real time, constructing and analyzing a dynamic topology characteristic vector to obtain dynamic topology characteristic data, and then generating a tie-line matrix representing the connection tightness between power grid equipment to accurately reflect the time-varying characteristics of the power grid structure; inputting the tie-line matrix and real-time operation parameters and historical fault data into a generative adversarial network to generate simulated fault data conforming to the actual distribution, and expanding the fault scene training set; based on the expanded data set, performing multi-objective optimization to solve an initial setting value solution set that takes into account the action speed, selectivity and reliability, and combining device constraints, topology adaptability and operation rules to perform compliance checking, and finally outputting a target setting value solution set that meets the actual operation requirements; thereby, using the generative adversarial network to generate high-fidelity simulated fault data in combination with historical fault data and dynamic operation information, expanding the training set to cover more actual operating conditions, and generating a target setting value solution set through multi-objective optimization and compliance checking, thereby improving the response capability and adaptability of the setting value to the dynamic changes of the power grid, and solving the problem that the existing method cannot meet the flexible operation requirements of the power grid due to the difficulty in real-time reflecting topology changes and operation states, resulting in insufficient setting value accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0023] Figure 1 A scene schematic diagram of the method for processing low-voltage protection setting values in a regional power grid provided by the present application;

[0024] Figure 2 A flowchart of the method for processing low-voltage protection setting values in a regional power grid provided by the present application Figure 1 ;

[0025] Figure 3 A flowchart of the method for processing low-voltage protection setting values in a regional power grid provided by the present application Figure 2 ;

[0026] Figure 4 A structure schematic diagram of the processing device for low-voltage protection setting values in a regional power grid provided by the present application;

[0027] Figure 5 A structure schematic diagram of the electronic device provided by the present application.

[0028] The specific embodiments of the application have been shown and described in considerable detail in order to explain clearly how to make and use the application. The above examples are illustrative only of the application and are not intended to limit the scope of the application in any way. Rather, reference should be made to the appended claims, which fully indicate the scope of the application, wherein: DETAILED DESCRIPTION

[0029] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description herein is intended for illustrating the exemplary embodiments of the application by reference to particular embodiments, but is not intended to limit the scope of the application in any way. Rather, reference should be made to the appended claims to resolve any ambiguities of the application.

[0030] First, the terms related to the present application are explained:

[0031] Principal Component Analysis (PCA) is an unsupervised linear dimensionality reduction method that aims to extract the most important variation direction (i.e., principal component) from high-dimensional data to preserve the core structure of the data and reduce redundancy. The basic principle is to find the orthogonal direction with the largest variance as the first principal component by performing eigenvalue decomposition on the data covariance matrix, and the subsequent principal components are orthogonal and have decreasing variance. By selecting the first several principal components with higher cumulative contribution rate, the original high-dimensional data can be projected into a low-dimensional space, reducing the computational complexity while preserving most of the information.

[0032] LASSO regression (Least Absolute Shrinkage and Selection Operator) is a linear regression method with L1 regularization, mainly used for feature selection and model simplification. The core idea is to add the sum of the absolute values of the regression coefficients as a penalty term in the loss function, forcing some unimportant features to shrink to zero, thereby achieving the purpose of automatically selecting key variables.

[0033] Dynamic Time Warping (DTW) is an algorithm for measuring the similarity between two time series, especially suitable for aligning sequences with different lengths or time shifts and speed changes. It finds the optimal matching path between two sequences by nonlinearly stretching or compressing the time axis, so that the cumulative distance between corresponding points is minimized, thereby achieving elastic alignment of sequences in the time dimension.

[0034] Currently in the power system, the low-voltage protection setting value calculation of the regional power grid is one of the important links to ensure the safe and stable operation of the power grid. The low-voltage protection setting value calculation methods currently used mostly rely on static power grid models and historical data, and the setting values are set through expert experience or preset rules. These methods usually include data acquisition, topology analysis, setting value calculation and verification, etc. The existing technology usually uses traditional methods in processing dynamic topology changes and multi-source data fusion, but also has important reference value in the basic framework of basic data processing and setting value calculation.

[0035] However, the traditional method has shortcomings in dynamic topology feature extraction and multi-objective optimization. The traditional method mostly uses a static topology model, which is difficult to reflect the dynamic changes of the power grid in real time, resulting in insufficient adaptability and accuracy of the protection setting value; at the same time, in the process of multi-objective optimization, the traditional method mostly relies on a single optimization algorithm, and cannot comprehensively consider multiple constraint conditions and objective functions, resulting in insufficient optimality and robustness of the optimization result.

[0036] The processing method for low-voltage protection setting value in the regional power grid provided by the present application improves the adaptability and accuracy of the protection setting value by analyzing the dynamic topology feature vector through tensor decomposition and aligning by combining dynamic time warping; and improves the optimality and robustness of the optimization result by generating a time series feature sequence through a sliding time window aggregation data and performing a Pareto hierarchical optimization.

[0037] Figure 1 The scene schematic diagram of the processing method for low-voltage protection setting value in the regional power grid provided by the present application is shown in Figure 1 The specific application scenario of the present application is a setting value processing system, which can be a server, and the server can be a mobile phone, a computer or the like. The present application does not set the execution subject, as long as the subject can determine the dynamic topology feature vector of the power grid equipment in a preset time window, the dynamic topology feature data corresponding to the dynamic topology feature vector, and the tie-in relationship matrix representing the connection closeness between the power grid equipment according to the connection relationship and the characteristic data of the power grid equipment in the power grid at different time points; wherein the dynamic topology feature data is obtained by tensor decomposition analysis of the dynamic topology feature vector; the tie-in relationship matrix, the operating parameters and the historical fault data in the power grid are input into a generative adversarial network to obtain simulated fault data output by the generative adversarial network; an initial target setting value solution set is obtained by multi-objective optimization according to the expanded training data set; the expanded training data set is obtained according to the simulated fault data and the historical fault data; the initial target setting value solution set is subjected to compliance verification to obtain a target setting value solution set, and the target setting value solution set is used to adjust the operating parameters of the power grid equipment according to the operating state of the power grid equipment.

[0038] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific examples. The following specific examples can be combined with each other, and the same or similar concepts or processes can not be described again in some examples. The embodiments of the present application will be described below with reference to the drawings.

[0039] Figure 2 Flowchart of the method for processing low-voltage protection setting of regional power grid Figure 1 As shown in the method, the method comprises the following steps. Figure 2

[0040] S201, according to the connection relationship and characteristic data of the power grid equipment in the power grid at different time points, determining the dynamic topology characteristic vector of the power grid equipment in a preset time window, the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, and the tie-in relationship matrix representing the connection closeness between the power grid equipment; wherein the dynamic topology characteristic data is obtained by tensor decomposition analysis of the dynamic topology characteristic vector.

[0041] The connection relationship of the power grid equipment at different time points can refer to the dynamic change of the physical connection state between each power grid equipment in the process of power grid operation, which is caused by the opening and closing of switches, the switching of operation modes, and the like.

[0042] The characteristic data of the power grid equipment at different time points can include real-time electrical quantities such as voltage, current, active / reactive power, frequency, phase angle, and non-electrical quantities such as temperature, load rate, protection action state, and the like. These data are collected through SCADA, PMU or intelligent terminal, and can reflect the operation condition and behavior characteristics of the power grid equipment at a specific moment, and are used to describe the dynamic behavior mode of the power grid.

[0043] The dynamic topology characteristic vector can refer to a high-dimensional vector representation that integrates the connection relationship of the power grid equipment and the multi-dimensional operation characteristics. It structurally encodes the topology state (such as the adjacency relationship) and key characteristic data of the equipment within a preset time window, forms a state snapshot evolving with time, and is used to represent the comprehensive topology-operation characteristics of the power grid within the time period.

[0044] The dynamic topology characteristic data can refer to the potential factors or components extracted by tensor decomposition of the dynamic topology characteristic vector. These data can represent the low-dimensional structure rules in the high-dimensional dynamic characteristics, such as the implicit coupling mode between devices, the dominant operation mode or the key influence path.

[0045] ​The liaison relationship matrix can refer to a weighted matrix reflecting the connection tightness degree between power grid devices, and the elements thereof represent the electrical coupling strength or information interaction probability between any two devices. The matrix is constructed based on dynamic topology feature data and physical connection relationship, and not only contains structural information of whether to be connected, but also quantifies dynamic association of how strong to be connected, which is used to support subsequent modeling of fault propagation path by the generative adversarial network.

[0046] In the embodiments of the present application, the operation data of each power grid device in the power grid at different time points can be collected in real time, and after preprocessing such as data cleaning, noise suppression, signal synchronization and switch state de-bouncing, the dynamic topology feature vector can be generated by combining PMU high-frequency electrical quantities, SCADA steady-state signals and non-electrical quantity information, and fusing device connection relationship and multi-dimensional features. The high-dimensional time sequence vector is subjected to tensor modeling, and the latent structure thereof is analyzed by using tensor decomposition method (such as CP or Tucker decomposition) to realize decoupling and dimensionality reduction of multi-dimensional features, extract implicit factors reflecting dynamic coupling mode of the power grid, form dynamic topology feature data, and align time sequence features with different synchronization or rhythm differences by using dynamic time warping (DTW) to improve feature consistency. Based on the dynamic topology feature data, an initial weighted adjacency matrix is constructed to represent the connection strength between devices, and the information transmission weight between nodes is learned by inputting the graph attention network (GAT), and the consistency of node relationship and actual topology is further fused with the physical connection rules and operation constraints of the power grid to evaluate the consistency of node relationship and actual topology, and generate topology relationship confidence. Finally, the confidence weighted, KL divergence detects abnormal connection and checks the topology rationality, and outputs the liaison relationship matrix reflecting the real connection tightness degree between the power grid devices.

[0047] S202, input the liaison relationship matrix, the operation parameters and the historical fault data in the power grid into the generative adversarial network to obtain simulated fault data output by the generative adversarial network.

[0048] The operation parameters in the power grid can refer to electrical quantity data and non-electrical quantity data reflecting the state of the power grid during normal or abnormal operation of the power grid, which can include real-time electrical quantity data such as bus voltage, line current, active / reactive power, frequency, load rate, and non-electrical quantity data such as transformer tap position, circuit breaker position, switch state signal, etc.

[0049] The historical fault data in the power grid can refer to a complete data set recorded by various real fault events occurred in the power grid in the past, which can include transient waveforms (such as voltage drop, current surge) in scenarios such as short-circuit fault, ground fault, broken line fault, protection action sequence, fault location information, wave recording data and fault clearing time, etc. In some embodiments, these data can come from fault recording device, relay protection device and SCADA system.

[0050] The generative adversarial network can refer to a deep learning model architecture composed of a generator and a discriminator, which are optimized through an adversarial training mechanism: the generator takes the adjacency matrix, operating parameters and random noise as input, and attempts to generate simulated fault data; the discriminator is responsible for distinguishing between generated data and real historical fault data. In the iterative training process, the generator continuously improves the authenticity of the generated samples, and finally outputs high-quality simulated data that approximates real faults in terms of statistical characteristics, time series dynamics and spatial propagation patterns, which can be used to expand the training sample library.

[0051] The simulated fault data can refer to synthetic data generated by the generative adversarial network, representing the response behavior of the power grid under various potential fault scenarios, which can include transient current waveform, voltage sag curve, phase jump, frequency fluctuation and other key fault characteristics. Among them, such data is not from actual accidents, but integrates power grid topology structure, dynamic correlation strength and operating constraints, and can simulate extreme, rare or never-occurring fault situations, thus effectively making up for the lack of historical data, and providing a rich training basis for multi-scenario adaptive optimization of protection settings.

[0052] In the embodiments of the present application, the adjacency matrix, operating parameters in the power grid and historical fault data are input into the generative adversarial network to obtain simulated fault data output by the generative adversarial network, which includes:

[0053] According to the adjacency matrix, operating parameters in the power grid and historical fault data, a key topological feature vector, a target feature vector and graph structure data are obtained; wherein the key topological feature vector is extracted from the normalized adjacency matrix; the target feature vector is obtained by principal component analysis on the adjacency matrix, operating parameters in the power grid and historical fault data; and the graph structure data is extracted from a multi-modal graph constructed according to the adjacency matrix, operating parameters in the power grid and historical fault data;

[0054] The key topological feature vector, the target feature vector and the graph structure data are input into the generator of the generative adversarial network to generate initial simulated fault data;

[0055] According to the distribution characteristics of the initial simulated fault data and the historical fault data, the initial simulated fault data is optimized and adjusted to obtain the simulated fault data.

[0056] The operating parameters of the power grid include electrical quantity data and non-electrical quantity data.

[0057] The tie relationship matrix, the operation parameters of the power grid and the historical fault data are cleaned of abnormal values and noise interference, and the data of different dimensions are uniformly normalized and standardized (so that the mean is 0 and the standard deviation is 1) to construct a standardized data set of uniform scale. On this basis, principal component analysis (PCA) is performed: first, the covariance matrix of the data set is calculated, and the eigenvectors corresponding to the first n largest eigenvalues are selected as principal components to retain the main variation mode in the data; then, based on the principal components, the LASSO regression method is applied to automatically select the feature variables that have the greatest influence on the fault scenario prediction through sparsity constraint, thereby obtaining the target feature vector.

[0058] The tie relationship matrix, the operation parameters of the power grid and the historical fault data are fused. First, different types of data are input into the adapted subnetwork for feature extraction, for example, the tie relationship matrix is input into the graph convolution network to extract the topological features of nodes and edges, the power grid operation parameters are input into the multi-layer perceptron to extract the steady-state features, and the historical fault data are input into the recurrent neural network to extract the time sequence dynamic features. Then, the features output by each subnetwork are spliced to form comprehensive multi-modal data containing spatial structure, operation state and time evolution information. On this basis, the multi-modal data is converted into a graph structure representation, that is, the power grid equipment is taken as a node, the connection relationship between the equipment is taken as an edge, and the attributes of the nodes and edges are valued by the multi-modal feature vectors (for example, the node features include voltage, current, temperature, etc., and the edge features include connection strength, communication delay, etc.). Further, the graph structure data is processed by a multi-layer graph convolution network, the neighbor information is aggregated layer by layer and the node representation is updated, and deep graph structure features are extracted, thereby obtaining the graph structure data.

[0059] The minimum-maximum normalization method is used to linearly transform each element in the tie relationship matrix and map it to the [0, 1] interval to generate a normalized tie relationship matrix. Based on the normalized matrix, key indicators reflecting the topological structure characteristics of the power grid are extracted: the node connection strength is directly represented by the connection weight of the corresponding node, the network sparsity is measured by the proportion of non-zero elements to the total elements, reflecting the connection density between devices, and the average connection weight is the mean of all connection weights, reflecting the overall connection strength of the network. The above features are combined to output a key topological feature vector containing node connection strength, network sparsity, average connection weight and other dimensions.

[0060] Based on the architecture of the generative adversarial network, the generator takes the key topological feature vector, the target feature vector and the graph structure data as input, models the spatial dependence relationship through the graph convolution network, generates simulated transient data including transient current waveform and voltage sag data; the discriminator receives the real transient data (derived from historical fault data) and the simulated data output by the generator, judges its authenticity and outputs a probability value between 0 and 1; through the alternating optimization of the adversarial training process of the generator and the discriminator, the generator continuously learns to minimize the identification loss of the discriminator on the generated data, while the discriminator continuously improves the ability to distinguish between real and generated data, thereby gradually improving the fidelity and quality of the generated data; after training, input the key topological feature vector and the graph structure data into the converged generator to generate multiple sets of high-fidelity transient current waveform and voltage sag data, and merge to form a complete transient data set (i.e. initial simulated fault data).

[0061] From the generated data (initial simulated fault data) and the real historical fault data, four types of key features are extracted: current peak in transient current waveform, voltage sag amplitude, rise time and duration in voltage sag data;

[0062] The mean and standard deviation of the generated data and the real data in each feature dimension are calculated, and the mean difference list of each feature dimension is generated by comparing the mean difference;

[0063] Based on the list, set the adjustment direction and amplitude, if the difference of a certain feature exceeds the preset threshold (such as the current peak difference is greater than 15%), formulate the corresponding adjustment rule (such as reducing the feature value in the generated data by 10%), and fuse the physical constraint condition (such as the transient current peak cannot exceed the upper limit of the circuit breaker action), to form a feature adjustment rule table containing adjustment strategy and engineering limit;

[0064] After the generator outputs the transient data, the feature values of the generated data are linearly scaled according to the rule table (such as the current peak is multiplied by 0.9), and Gaussian noise with a mean of 0 and a standard deviation of 5% of the feature standard deviation is superimposed on the scaled data to enhance the naturalness and diversity of the data, generating transient data with optimized distribution characteristics; use the adjusted generator to generate multiple sets of transient data (i.e. simulated fault data), and randomly extract real transient data from historical fault data, mix the generated data and real data in a ratio of 1:0.5 (such as 10000 groups of generated data and 5000 groups of real data), to finally build an expanded training data set with high quality and high coverage.

[0065] S203、According to the expanded training data set, the initial target setting solution set is obtained by multi-objective optimization; the expanded training data set is obtained according to the simulated fault data and the historical fault data.

[0066] In the present embodiment, the multi-objective optimization can refer to a mathematical optimization method for finding a set of solutions that can make all objectives as good as possible when considering two or more conflicting objective functions at the same time. Since there is often a trade-off relationship between these objectives (for example, improving accuracy may reduce speed), there is usually no single solution that can optimally satisfy all objectives, so the goal of multi-objective optimization is to obtain a non-dominated solution set (i.e., a Pareto optimal solution set), in which each solution performs well in some objectives and cannot be further improved in a certain objective without deteriorating other objectives.

[0067] In the present embodiment, the multi-objective optimization can refer to constructing a multi-dimensional objective function based on the diversified failure and operation scenarios provided by the expanded training data set under the premise of considering multiple performance indicators that are mutually constrained, and searching for a set of non-dominated optimal solutions in the solution space through an optimization algorithm.

[0068] The initial target setting solution set can refer to a set of non-dominated optimal solutions obtained after Pareto hierarchical optimization, wherein each solution represents a specific protection setting parameter configuration (such as overcurrent setting, action delay, etc.) and achieves a relatively optimal balance between multiple performance indicators.

[0069] In the present embodiment, the initial target setting solution set is obtained by multi-objective optimization according to the expanded training data set, including:

[0070] According to the preset sliding time window, determine the data and time sequence feature vectors of the expanded training data set in each sliding time window;

[0071] According to the time sequence feature vectors of the expanded training data set in each sliding time window, obtain a time sequence feature sequence;

[0072] Pareto hierarchical optimization is performed on the time sequence feature sequence to obtain a hierarchical optimized Pareto front;

[0073] According to the hierarchical optimized Pareto front and the engineering constraint conditions, obtain the initial target setting solution set.

[0074] The standardized augmented training data set is generated by first performing standardization preprocessing based on the augmented training data set: identifying and filtering abnormal values by using the 3 sigma principle, filling in missing data by using linear interpolation, suppressing high-frequency noise by using a low-pass filter, and normalizing the data to scale them to the interval [0, 1]; then, starting from the starting point of the time series of the data set, a fixed-length sliding time window is used to move step by step at a preset time step, generating a series of continuous time windows, each window covering a data segment in the corresponding time range; in each time window, statistical features such as mean, variance, maximum, minimum, and peak value are extracted to form the time series feature vector of the window; finally, all time series feature vectors are arranged in order of their corresponding window starting times to form a complete time series feature sequence.

[0075] Each solution in the time series feature sequence (i.e., the fixed value solution of each time window) is sorted by non-dominance, compared with all solutions, and the dominance relationship of each solution is determined, and the solutions are divided into different non-dominant levels, the first layer contains all solutions that are not dominated by any other solution, the second layer contains solutions that are dominated by the first layer solutions but not by other solutions, and the output is the layered solution set; based on the non-dominant sorting, different evolution parameters are assigned to each level, including crossover probability and mutation probability, etc., which are used to guide the optimization process, the parameters of the evolution algorithm are adjusted according to the level of the solution, and a solution set with optimized parameters is generated; the solution set with optimized parameters is optimized by genetic algorithm, in each generation of evolution process, the solution set is selected, crossed and mutated by using non-dominant sorting and layered evolution parameters, and the Pareto frontier is gradually approached, through multiple generations of evolution, a stable Pareto frontier is finally obtained, and the layered optimized Pareto frontier is output.

[0076] It should be noted that the parameters of the evolution algorithm can be dynamically set according to the actual optimization process. The solutions in the high level of the Pareto frontier have better target performance, in order to retain their excellent characteristics, they can be assigned a lower mutation probability; while the solutions in the low level have weaker exploration ability, it is appropriate to give them a higher mutation probability to enhance the global search ability. For example, the initial value of the crossover probability can be set to 0.8, and the initial value of the mutation probability can be set to 0.1; for high-level solutions, the mutation probability can be adjusted to 0.05 to maintain stability, while for low-level solutions, it can be increased to 0.2 to promote the diversity exploration of the solution space, thereby achieving an effective balance between convergence and diversity.

[0077] The engineering constraints include physical limitations of grid equipment (such as maximum current carrying capacity), adjustable range of protection device setting value (such as minimum and maximum setting value), and safety standards of grid operation (such as allowable fluctuation range of voltage and frequency), etc., which are mainly extracted from grid design specifications, equipment technical manuals and related operation regulations. Based on the obtained constraints, the feasibility screening is performed on the Pareto hierarchical optimization front solution set: for each candidate solution, the target values of each item are extracted, and whether all preset engineering constraints are met is judged one by one. Through the screening process, the infeasible solutions that do not meet the equipment capacity or operation requirements are eliminated, and a subset containing only compliant solutions is generated, and finally the initial optimal setting value solution set that meets the actual application conditions is output, ensuring that the optimization result has engineering implementability.

[0078] S204, compliance verification is performed on the initial target setting value solution set to obtain a target setting value solution set, and the target setting value solution set is used to adjust the operation parameters of the grid equipment according to the operation state of the grid equipment in the grid.

[0079] The compliance verification can refer to multi-dimensional rule verification on each candidate solution in the initial target setting value solution set to ensure that it meets the engineering feasibility and technical specification requirements of the grid operation. The process includes checking whether the setting value parameters are within the device physical limit range (such as not exceeding the maximum breaking current of the circuit breaker), whether they meet the upper and lower limits of the protection device setting value, whether they meet the selective coordination logic (such as the time difference of upper and lower level protection actions), and whether they adapt to the current grid topology and operation mode. In addition, it also needs to verify whether its action behavior in different fault scenarios meets the dispatching regulations and safety standards. Only the solutions that pass all the verification rules are considered to be compliant, thereby ensuring the implementability and system safety of the final setting value scheme.

[0080] The target setting value solution set can refer to the final feasible solution set retained after compliance verification.

[0081] In the embodiments of the present application, the compliance verification is performed on the initial target setting value solution set to obtain a target setting value solution set, which includes:

[0082] The compliance verification is performed on the initial target setting value solution set to determine a first setting value solution set representing a compliant state and a second setting value solution set representing a boundary non-compliant state in the initial target setting value solution set;

[0083] The second setting value solution set representing the boundary non-compliant state is input to the adversarial generation network for calibration processing to generate a calibrated setting value solution set representing the compliant state;

[0084] The first setting value solution set and the calibrated setting value solution set are screened to obtain a target setting value solution set.

[0085] Wherein, the initial optimal constant value solution set is checked for compliance, firstly, based on the preset static rules (including protection constant value upper and lower limits, device rated parameters and protection action time, etc.), each candidate solution is compared with the rules one by one, and solutions meeting all static constraints are screened out to generate a solution set passing the static rule screening (the first constant value solution set); secondly, combined with real-time power grid topology data, the topology adaptability of each solution is matched to verify its effectiveness under the current network structure, and the solutions that are not applicable are excluded to form a solution set passing the dynamic topology screening; at the same time, each solution is checked with the physical limit conditions of the power grid equipment (such as maximum current carrying capacity, protection device adjustable range, etc.) to ensure that the parameters are within the safe operation boundary of the equipment, and the solution set passing the equipment limit screening is output. Finally, the data fusion technology is used to perform intersection operation on the three types of screened solution sets, and the feasible solutions that meet the static rules, dynamic topology adaptability and equipment constraints at the same time are comprehensively retained to generate a preliminary compliant solution set.

[0086] Among them, the allowed range of the device refers to the range of electrical parameters that the device can safely withstand under normal operating conditions; the device limit conditions include the action constant value range of protection devices such as overcurrent protection and overvoltage protection; the allowed range of the device and the device limit conditions come from the device technical manual and industry standards, etc.

[0087] Each solution in the preliminary compliant solution set is finely evaluated to identify candidate solutions that are close to the constraint boundary on key targets but do not fully meet all compliance requirements, forming a boundary non-compliant solution set (the second constant value solution set); the solution set is standardized to extract its key features, including electrical parameters, topology structure features, protection constant values and device limit conditions, etc., and converted into an input format acceptable to the generative adversarial network to form a pre-processed boundary non-compliant solution set; input this data set into the generative adversarial network, where the generator generates new constant value solutions based on the input features, and the discriminator discriminates their compliance, continuously optimizes the output quality of the generator through the adversarial training mechanism, and finally generates a set of high feasibility, compliant replacement constant value solution set (calibrated constant value solution set); fuse the replacement constant value solution set with the preliminary compliant solution set to ensure that all solutions meet the engineering and operation constraints, forming an optimization candidate set that covers a wider range and has better performance.

[0088] The standardization processing is sequentially performed on each solution in the optimization candidate set (making the mean value 0 and the standard deviation 1), and then the normalization processing is performed, and all feature values are scaled to the interval [0, 1] to generate a pre-processed optimization candidate set; the non-dominated sorting algorithm is used to stratify the solution set: through pairwise comparison, if one solution is not inferior to another solution in all objectives, and is better in at least one objective, it is determined that the former dominates the latter, and the solution not dominated by any other solution constitutes the first layer Pareto front, only the second layer is dominated by the first layer solution and not dominated by the remaining solutions, and so on, and the stratified Pareto front solution set is output; according to the stratification result, a differentiated evolution strategy is implemented, the high-level solution is assigned a lower mutation probability because of its better performance to preserve good characteristics, and the low-level solution is assigned a higher mutation probability to enhance the global exploration ability, the initial values of the crossover probability and the mutation probability can be set according to real-time data, for example, the crossover probability is 0.8 and the mutation probability is 0.1, the mutation probability of the high-level solution is set to 0.05, and the mutation probability of the low-level solution is set to 0.2, forming a solution set with optimization parameters; in each generation evolution process, the high-level solution is preferentially selected based on the non-dominated sorting for selection operation, the crossover operation is realized by feature exchange to generate new individuals, and the mutation operation is performed in combination with the stratified mutation strategy to improve diversity, and the optimal Pareto front is gradually approached through multiple iterations, and the evolved Pareto front solution set is output; finally, combined with the engineering constraint conditions (including the maximum current carrying capacity of the power grid equipment, the setting value range of the protection device, the allowable fluctuation range of voltage and frequency, etc.), each solution in the front solution set is checked for compliance, solutions that do not meet the physical limitations or operation safety requirements are excluded, and only feasible solutions that fully comply with the engineering specifications are retained, and finally the target setting value solution set is output.

[0089] In the embodiments of the present application, the method further comprises:

[0090] According to the operating state of the power grid equipment in the power grid, the corresponding adjustment setting value is determined from the target setting value solution set;

[0091] The safety check processing is performed on the adjustment setting value, and the adjustment setting value after the safety check processing is sent to the power grid equipment in the power grid;

[0092] After the power grid equipment in the power grid adjusts according to the adjustment setting value, the corresponding power grid data is collected by the edge nodes distributed in the power grid;

[0093] According to the corresponding power grid data collected by the edge nodes distributed in the power grid and the historical data of the power grid, a corresponding monitoring report is generated and displayed; wherein the data representing the digital thread full-link record in the process of generating the monitoring report are stored in the blockchain.

[0094] The power grid data is collected in real time, preprocessed through data cleaning, signal synchronization and switch state debouncing, and structured operation data is generated. The switch state signal is analyzed, the connection relationship between power grid devices is stored in the form of an adjacency list, and the connected components are identified by traversing the adjacency list through depth-first search (DFS) to realize dynamic identification of the topological structure. The extracted topological features are mapped to a three-dimensional tensor (dimension time x node x feature type), and the CP decomposition method is applied to decompose the tensor to extract independent dynamic topological patterns and generate independent dynamic topological feature vectors. The feature vectors are associated with power grid device parameters, and after standard conversion, standardized entities are formed to describe the topological relationship between devices in the form of "subject-predicate-object" triplets, and a unique digital thread ID is assigned to each data record to construct digital thread metadata. At the same time, for the triggering conditions of the setting value adjustment of the power grid protection device (such as the state change of the tie switch and the load rate exceeding 80%), a structured event description string is generated, the hash value is calculated through the SHA-256 algorithm, and the hash value, timestamp and operator PKI digital certificate are written into the private blockchain. The chain storage data is organized using the Merkle tree structure, and finally the tamper-proof blockchain storage certificate is output, and a traceable and credible setting value adjustment scheme is generated.

[0095] The state space includes independent dynamic topological feature vectors and power grid device operation parameters, the action space generates candidate setting value schemes (such as overcurrent protection setting value range [0.5, 2.0]In) and balances the misoperation rate and the refusal rate, for example, the misoperation rate weight is 0.7, the refusal rate weight is 0.3, the sensitive information (such as device ID) is hidden using the zk-SNARKs algorithm, and only the verification result is disclosed, for example, the scheme meets the constraints.

[0096] The edge node receives the credible setting value adjustment scheme through the MQTT protocol, extracts the key parameters including the protection setting value, the applicable topological range and the blockchain storage reference, etc. using encoding analysis, and verifies the digital signature of the scheme using the ECDSA algorithm to ensure its integrity and source credibility. If the signature verification fails, the scheme is discarded and an alarm is triggered. Then, according to the blockchain storage reference in the scheme, the corresponding storage data is queried from the private chain network to verify whether the scheme has passed the credible storage, and the parsed credible setting value adjustment instruction is output after confirmation.

[0097] Based on the analysis result, the edge node issues instructions (including fixed value parameters, time stamps, etc.), performs fixed value adjustment operation, and records the execution status (success or failure); if negative feedback or timeout is received, a retry mechanism (up to 3 times) is started, and if it still fails, it is marked as an execution exception and reported, forming an execution status log. In the running process, the edge node continuously monitors the power grid state in real time, collects data and pre-processes it, retrieves the same period data (such as the voltage and load curve of the same period in the past 7 days) from the historical database, and the retrieval conditions cover the time range, device ID and topology state label; the fast Fourier transform (FFT) is used to extract the current voltage harmonic component, compare the harmonic amplitude difference with the historical data, calculate the statistical deviation (such as mean difference, standard deviation) of the current load and the historical load, and identify abnormal fluctuation trend; at the same time, the sliding window is used to analyze the time sequence characteristics such as frequency and duration of voltage sag event, evaluate the change trend of power grid stability, and generate a comparative analysis report.

[0098] Further, the power grid state data is aggregated in time dimension (such as every minute), the maximum / minimum voltage, average load and other statistical indicators are extracted, and the standard deviation, peak-valley difference and other characteristic quantities are calculated, the abnormal events (such as voltage out-of-limit, load mutation) are marked in the report combined with the comparative analysis result, and the current topology state (such as the disconnection of tie switch during fault) is associated; finally, the execution status log and the comparative analysis report are fused to generate a comprehensive real-time monitoring report, realizing the closed loop of fixed value execution and dynamic perception of running state.

[0099] The method provided by the embodiment of the application for processing low-voltage protection fixed value in a regional power grid analyzes a high-dimensional dynamic topology feature vector by using a tensor decomposition method, effectively decouples the complex coupling relationship between time, nodes and feature dimensions, extracts independent and interpretable low-rank topology patterns, and reveals the potential evolution law of the power grid structure. Meanwhile, a dynamic time warping technique is introduced to nonlinearly align the topology change process with different synchronization or time rhythm differences, eliminate the time sequence misalignment caused by sampling deviation or event delay, improve the consistency and timeliness of the feature representation, and thus enhance the adaptability and setting accuracy of the protection fixed value in multiple scenarios.

[0100] Further, a sliding time window is used to continuously aggregate the running data, construct a time sequence feature sequence covering different working conditions, and fully depict the evolution process of the power grid dynamic behavior. On this basis, Pareto hierarchical optimization is implemented, the non-dominated sorting mechanism is used to hierarchically screen the candidate solution set, the front solution balanced among multiple objective performances (such as speed, selectivity and sensitivity) is reserved, and the search efficiency is improved by using a differentiated evolution strategy, effectively avoiding local convergence, and improving the global optimality and robustness of the optimization result in a complex and uncertain environment.

[0101] Figure 3The flowchart of the processing method of low-voltage protection setting value in a regional power grid provided in the present application Figure 2 As shown in the embodiment, the embodiment is based on the embodiment Figure 3 The embodiment is based on the embodiment Figure 2 The embodiment is based on the embodiment

[0102] S301, determining a static topology feature vector of a power grid device at different time points in the power grid according to the connection relationship and feature data of the power grid device at different time points in the power grid; the feature data includes electrical quantity data and non-electrical quantity data of the power grid device;

[0103] S302, performing aggregation processing on the static topology feature vectors of the power grid devices at different time points in the power grid to determine a dynamic topology feature vector of the power grid device within a preset time window.

[0104] The feature data includes electrical quantity data and non-electrical quantity data.

[0105] The electrical quantity data includes PMU high-frequency data and SCADA steady-state data.

[0106] The PMU high-frequency data includes phase difference of voltage and current, voltage and current effective value, and frequency fluctuation; the phase difference of voltage and current, the voltage and current effective value, and the frequency fluctuation can be directly obtained through the PMU.

[0107] The SCADA steady-state data includes switch state signal, circuit breaker action signal, and transformer tap information; the switch state signal and the circuit breaker action signal are collected through a smart meter, a DTU or a substation measurement and control device, and uploaded to the SCADA, and the transformer tap information is collected through a transformer on-load voltage regulation controller and accessed to the SCADA.

[0108] The non-electrical quantity data includes device temperature and environmental parameters; the device temperature is obtained through a temperature sensor, and the environmental parameters are obtained using an environmental monitoring sensor.

[0109] In the embodiments of the present application, after the feature data is acquired, the feature data can be preprocessed, and the preprocessing includes data cleaning, noise suppression, signal synchronization and alignment, and switch state de-bouncing; the data cleaning is to filter transient spikes or noise in the voltage and current based on the 3σ principle, and the missing values are filled using linear interpolation; the noise suppression is to use a low-pass filter, and the cutoff frequency is set according to the fundamental frequency of the power grid, for example, the cutoff frequency is set to 100 Hz for a 50 Hz power grid, to eliminate high-frequency noise interference, and the window sliding average is performed on the SCADA steady-state data, for example, the window length is 5-10 sampling points, to smooth short-term fluctuations; the signal synchronization and alignment is to align the time stamps of the PMU high-frequency data and the SCADA steady-state data, synchronize in the time dimension, and correct the phase deviation of the voltage and current signals based on the synchronous clock signal of the PMU to perform phase correction. The switch state de-bouncing is to set a state retention time threshold, for example, 50 ms, and if the switch state frequently switches within the threshold, for example, due to mechanical vibration or signal interference, it is determined to be de-bouncing and a stable state value (such as the last valid state) is output.

[0110] In combination with the preprocessed data and the switch state signal, generating the dynamic topology feature vector can include:

[0111] 1. According to the type and number of power grid devices, the fixed dimension of the dynamic topology feature vector is defined, for example, each device corresponds to a group of electrical quantity features (such as voltage, current and frequency), a switch state signal and a group of non-electrical quantity features (such as temperature, temperature and humidity), and the electrical quantity features, the switch state signal and the non-electrical quantity features of each power grid device are spliced in order to form a local feature sub-vector of the device;

[0112] 2. The local feature sub-vectors of all devices are spliced in order according to the power grid device ID to generate a static topology feature vector;

[0113] 3. The static topology feature vectors in each time window are aggregated to generate a dynamic topology feature vector in the time window.

[0114] S303, performing tensor decomposition on the dynamic topology feature vector of the power grid device in the preset time window to obtain decoupled features;

[0115] S304, performing alignment processing on the decoupled features according to dynamic time warping to obtain dynamic topology feature data.

[0116] Specifically, the dynamic topological feature vector is represented as a higher-order tensor. This higher-order tensor is then decomposed into a core tensor and multiple factor matrices using higher-order singular value decomposition, generating decoupled core tensors and factor matrices. Dynamic time warping is applied to find the optimal path in the time dimension, minimizing the cumulative distance between two time series for time series alignment, generating aligned dynamic topological feature data. Finally, the decoupled core tensor and factor matrices are fused with the aligned dynamic topological feature data to generate the final dynamic topological feature data.

[0117] S305. Based on the dynamic topology feature data corresponding to the dynamic topology feature vector, obtain the connection relationship matrix that characterizes the degree of connection between power grid equipment.

[0118] Specifically, based on the dynamic topology feature data corresponding to the dynamic topology feature vector, the topology information reflecting the dynamic changes of the network structure is extracted by quantifying the connection status and electrical correlation strength between power grid devices at different times. The connection relationship between devices is modeled as a weighted graph structure, where nodes represent power grid devices and the weight of the edges represents the degree of connection tightness. Finally, a connection relationship matrix that evolves over time is constructed to characterize the dynamic tightness of the connection between power grid devices.

[0119] In this embodiment of the application, a connection matrix characterizing the degree of interconnection between power grid devices is obtained based on the dynamic topology feature data corresponding to the dynamic topology feature vector, including:

[0120] Generate an adjacency matrix based on the power grid equipment in the power grid and the equipment connection relationships between the power grid equipment;

[0121] The connection weights between power grid devices are determined based on the electrical and non-electrical quantity characteristics in the dynamic topology feature data.

[0122] Generate a weighted adjacency matrix based on the connection weights and adjacency matrices between power grid devices;

[0123] Based on the matrix nodes in the weighted adjacency matrix and the connection relationships between matrix nodes, a node relationship graph is obtained;

[0124] Based on the actual topology of the power grid, determine the confidence level of the connections between nodes in the node relationship diagram;

[0125] Based on the confidence of the connections between nodes in the node relationship graph, the connection relationships between nodes in the node relationship graph are traversed to obtain the connection matrix.

[0126] Based on the type and quantity of equipment in the power grid, a preliminary adjacency matrix describing their connection relationships is constructed. Then, combining electrical quantities (such as voltage, current, and frequency) and non-electrical quantities (such as communication status and interaction records) from the dynamic topology feature data, the connection strength between devices is comprehensively evaluated from four dimensions: electrical connection tightness, communication real-time performance, historical interaction frequency, and electrical feature matching degree. Each indicator is graded into a score of 1 to 5 based on actual operating characteristics, with higher scores indicating stronger connections. Specifically: 5 points for direct electrical connection, 4 points for short-distance connection (<1km), 3 points for medium distance (1-5km), 2 points for long distance (>5km), and 1 point for no direct electrical connection; 5 points for communication delay less than 10ms, 4 points for 10-50ms, 3 points for 50-100ms, 2 points for 100-500ms, and 1 point for more than 500ms or communication interruption; 5 points for multiple interactions per minute, 4 points for multiple interactions per hour, 3 points for 1-2 interactions per hour, 2 points for several interactions per day, and 1 point for none or less than ten interactions per week; 5 points for electrical parameter deviation less than 5%, 4 points for 5%-10%, 3 points for 10%-20%, 2 points for 20%-50%, and 1 point for greater than 50%.

[0127] Subsequently, the scores of each indicator are mapped to corresponding weight ranges: 5 points correspond to 1.0, 4 points to 0.8–0.9, 3 points to 0.6–0.7, 2 points to 0.3–0.5, and 1 point to 0.1–0.2. For each device connection, the intersection of the weight ranges of the four indicators is taken as its comprehensive weight range (for example, if electrical connection tightness is 4 points (0.8–0.9), communication real-time performance is 3 points (0.6–0.7), historical interaction frequency is 5 points (1.0), and electrical feature matching degree is 4 points (0.8–0.9), then the intersection is 0.8–0.9), and the median of this range (e.g., 0.85) is taken as the final connection weight. The calculated weights are filled into the preliminary adjacency matrix to generate a weighted adjacency matrix, providing a refined topological input foundation for subsequent graph structure modeling and intelligent analysis.

[0128] The calculated connection weights are filled into the initial adjacency matrix to generate a weighted adjacency matrix, which is then input into a graph attention network. This network dynamically calculates the relationship weights between each node and its neighboring nodes through an attention mechanism, effectively capturing the importance of local connections. It also aggregates neighborhood information through multi-layer stacking and iteratively updates the feature representations of each node. Based on this, the learned attention weights are used to identify key connection paths and core nodes, outputting a node relationship graph reflecting the current operating state. Subsequently, the network combines power grid physical connection rules (such as electrical connectivity, communication timeliness, and equipment coupling characteristics) and topological constraints (such as voltage level consistency). The system performs rule constraints and topology consistency checks on the node relationship graph (including loop legality and switch state logic) to determine whether the connections conform to the actual power grid operation logic. At the same time, it uses KL divergence to quantify the probability distribution difference between the node relationship graph and the designed topology in the power grid equipment ledger, and calculates the topology confidence score. That is, by comparing the consistency between the real-time inference results and the standard connections, the reliability of the topology structure is evaluated. Connections with confidence scores below a set threshold (such as 0.7) or that violate physical rules will be marked as abnormal or low-confidence connections and adjusted. Finally, the system outputs a power grid topology structure that includes reliable connection relationships and their confidence score evaluation results.

[0129] It should be noted that physical connection rules are used to characterize the inherent correlation characteristics of connections between power grid equipment. Electrical connection tightness reflects the impedance characteristics between equipment and the fault current propagation path; communication real-time performance reflects the timeliness of protection and control signal interaction; historical interaction frequency characterizes the dynamic coupling degree of equipment during operation; and electrical characteristic matching degree measures the similarity of equipment in operating parameters such as voltage, current, and frequency. Topology constraints are a set of rules used to verify whether the equipment connection relationships conform to the actual power grid operating logic. These include constraints such as physical connection feasibility, electrical connectivity, voltage level consistency, topology loop legality, and switch state logic, collectively ensuring that the identified topology has engineering rationality and physical feasibility.

[0130] It should be noted that the formula for obtaining topological confidence is:

[0131] C = A / B;

[0132] Where C represents the topology confidence level, 0≤C≤1, C=1 indicates that all connections are consistent with the ledger, C=0 indicates that no connection is consistent with the ledger, A is the number of connections that are completely consistent with the real-time topology and the ledger design, B is the total number of all connection relationships defined in the ledger, C≥0.9 is high confidence, 0.7≤C≤0.9 is medium confidence, and C<0.7 is low confidence.

[0133] Based on the power grid topology and confidence assessment results, all connections undergo dual verification: First, each connection is checked to ensure it conforms to physical connection rules; if a violation is found, it is marked as an abnormal connection. Simultaneously, its confidence level is assessed to determine if it falls below a set threshold (e.g., 0.7); if so, it is marked as a low-confidence connection. All marked abnormal and low-confidence connections are removed, generating a cleaned and verified topology. Subsequently, this verified topology and the standard connection relationships in the equipment design ledger are transformed into binary event sets (i.e., connections exist or do not exist). For example, if bus A and circuit breaker 1 are connected in the real-time topology, it is recorded as an existing event; otherwise, it is recorded as non-existent. This constructs a real-time topology event set and a reference topology event set, providing structured input for subsequent probabilistic analysis.

[0134] Based on two event sets, the number of events present in the real-time topology and the total number of events are statistically analyzed, and their occurrence probabilities are calculated to obtain the probability distribution of the real-time topology, which is then compared with the probability distribution of the reference topology. The difference between the two is quantified using KL divergence, identifying high-deviation connections that statistically significantly deviate from design expectations, further verifying their reliability. For example, if a connection is retained after verification (high confidence), but KL divergence shows its behavior pattern is severely inconsistent with historical benchmarks, it requires further review. Finally, a matrix is ​​initialized, and the confidence levels of each verified connection are used as weights to fill the corresponding elements. After traversing all connections and completing the filling, a weight threshold (e.g., 0.1) is set to remove weak connections to reduce matrix complexity. The final output is a compact and semantically clear connection matrix, used to characterize the tightness and reliability of dynamic connections between devices.

[0135] Figure 4 This is a schematic diagram of the processing device for low-voltage protection settings in the regional power grid provided in this application, as shown below. Figure 4 As shown, the low-voltage protection setting processing device 40 provided in this embodiment includes:

[0136] The determination module 401 is used to determine the dynamic topology feature vector of the power grid equipment within a preset time window, the dynamic topology feature data corresponding to the dynamic topology feature vector, and the connection relationship matrix characterizing the degree of connection between the power grid equipment, based on the connection relationship and feature data of the power grid equipment at different time points in the power grid; wherein, the dynamic topology feature data is obtained by tensor decomposition analysis of the dynamic topology feature vector.

[0137] Module 402 is used to input the connection matrix, the operating parameters in the power grid and historical fault data into the generator adversarial network to obtain the simulated fault data output by the generator adversarial network.

[0138] The multi-objective optimization module 403 is used to obtain the initial objective set of solutions by solving the multi-objective optimization problem based on the expanded training dataset; the expanded training dataset is obtained based on simulated fault data and historical fault data.

[0139] The compliance verification module 404 is used to perform compliance verification on the initial target set of values ​​to obtain the target set of values. The target set of values ​​is used to adjust the operating parameters of the power grid equipment according to the operating status of the power grid equipment.

[0140] In one possible implementation, the determining module 401 can also be specifically used for:

[0141] Based on the connection relationships and characteristic data of power grid equipment at different points in time, determine the static topology feature vectors of power grid equipment at different points in time; the feature data includes electrical quantity data and non-electrical quantity data of power grid equipment.

[0142] The static topology feature vectors of power grid equipment at different points in time are aggregated to determine the dynamic topology feature vectors of power grid equipment within a preset time window.

[0143] Tensor decomposition is performed on the dynamic topology feature vectors of power grid equipment within a preset time window to obtain dynamic topology feature data corresponding to the dynamic topology feature vectors.

[0144] Based on the dynamic topology feature data corresponding to the dynamic topology feature vector, a connection relationship matrix representing the degree of connection between power grid equipment is obtained.

[0145] In one possible implementation, the determining module 401 can also be specifically used for:

[0146] Tensor decomposition is performed on the dynamic topological feature vectors of power grid equipment within a preset time window to obtain decoupling features;

[0147] Based on dynamic time warping, the decoupled features are aligned to obtain dynamic topological feature data.

[0148] In one possible implementation, the determining module 401 can also be specifically used for:

[0149] Generate an adjacency matrix based on the power grid equipment in the power grid and the equipment connection relationships between the power grid equipment;

[0150] The connection weights between power grid devices are determined based on the electrical and non-electrical quantity characteristics in the dynamic topology feature data.

[0151] Generate a weighted adjacency matrix based on the connection weights and adjacency matrices between power grid devices;

[0152] Based on the matrix nodes in the weighted adjacency matrix and the connection relationships between matrix nodes, a node relationship graph is obtained;

[0153] Based on the actual topology of the power grid, determine the confidence level of the connections between nodes in the node relationship diagram;

[0154] Based on the confidence of the connections between nodes in the node relationship graph, the connection relationships between nodes in the node relationship graph are traversed to obtain the connection matrix.

[0155] In one possible implementation, module 402 can also be specifically used for:

[0156] Based on the interconnection matrix, power grid operating parameters, and historical fault data, key topology feature vectors, target feature vectors, and graph structure data are obtained. The key topology feature vectors are extracted from the normalized interconnection matrix; the target feature vectors are obtained through principal component analysis of the interconnection matrix, power grid operating parameters, and historical fault data; and the graph structure data are extracted from a multimodal graph constructed based on the interconnection matrix, power grid operating parameters, and historical fault data.

[0157] The key topological feature vectors, target feature vectors, and graph structure data are input into the generator of the generative adversarial network to generate initial simulated fault data.

[0158] Based on the distribution characteristics of the initial simulated fault data and historical fault data, the initial simulated fault data is optimized and adjusted to obtain the simulated fault data.

[0159] In one possible implementation, the multi-objective optimization module 403 can also be specifically used for:

[0160] Based on the preset sliding time window, determine the data and temporal feature vectors of the expanded training dataset in each sliding time window;

[0161] Based on the temporal feature vectors of the expanded training dataset in each sliding time window, a temporal feature sequence is obtained;

[0162] Pareto hierarchical optimization is performed on the time-series feature sequences to obtain the hierarchically optimized Pareto front.

[0163] Based on the hierarchical optimization of the Pareto front and the engineering constraints, the initial objective constant solution set is obtained.

[0164] In one possible implementation, the compliance verification module 404 can also be specifically used for:

[0165] Perform compliance verification on the initial target set of solutions to determine the first set of solutions representing the compliant state and the second set of solutions representing the non-compliant state at the boundary.

[0166] The second set of fixed values ​​representing the non-compliant state at the boundary is input into the anti-generative network for calibration processing, generating a calibrated set of fixed values ​​representing the compliant state.

[0167] The first set of constant values ​​and the calibration set of constant values ​​are filtered to obtain the target set of constant values.

[0168] In one possible implementation, the compliance verification module 404 can also be specifically used for:

[0169] Based on the operating status of power grid equipment in the power grid, the corresponding adjustment setpoints are determined from the target setpoint solution set;

[0170] The adjustment setting value is subjected to a security verification process, and the adjusted setting value after security verification is sent to the power grid equipment in the power grid;

[0171] After the power grid equipment in the power grid is adjusted according to the adjustment setpoint, the corresponding power grid data is collected through the edge nodes distributed in the power grid;

[0172] Based on the corresponding power grid data and historical power grid data collected by edge nodes distributed in the power grid, a corresponding monitoring report is generated and displayed; among them, the data representing the entire link record of the digital thread during the generation of the monitoring report is stored in the blockchain.

[0173] The low-voltage protection setting processing device in the regional power grid provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0174] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0175] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0176] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0177] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0178] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0179] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0180] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0181] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0182] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0183] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0184] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0185] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0186] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0187] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0188] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0189] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for processing low voltage protection setting in a regional power grid, characterized in that, The method comprises the following steps: According to the connection relationship and characteristic data of the power grid equipment in the power grid at different time points, the dynamic topology characteristic vector of the power grid equipment in the preset time window, the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, and the contact relationship matrix representing the connection closeness between the power grid equipment are determined; wherein the dynamic topology characteristic data is obtained by tensor decomposition analysis of the dynamic topology characteristic vector; The contact relationship matrix, the operating parameters and historical fault data in the power grid are input into a generative adversarial network to obtain simulated fault data output by the generative adversarial network; According to the augmented training data set, the initial target setting solution set is obtained by multi-objective optimization solution; the augmented training data set is obtained according to the simulated fault data and the historical fault data; The initial target setting solution set is subjected to compliance verification to obtain a target setting solution set, which is used to adjust the operating parameters of the power grid equipment according to the operating state of the power grid equipment.

2. The method of claim 1, wherein, According to the connection relationship and characteristic data of the power grid equipment in the power grid at different time points, the dynamic topology characteristic vector of the power grid equipment in the preset time window, the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, and the contact relationship matrix representing the connection closeness between the power grid equipment are determined; wherein the dynamic topology characteristic data is obtained by tensor decomposition analysis of the dynamic topology characteristic vector; According to the connection relationship and characteristic data of the power grid equipment in the power grid at different time points, the static topology characteristic vector of the power grid equipment at different time points is determined; the characteristic data includes electrical quantity data and non-electrical quantity data of the power grid equipment; The static topology characteristic vectors of the power grid equipment at different time points are subjected to aggregation processing to determine the dynamic topology characteristic vector of the power grid equipment in the preset time window; The dynamic topology characteristic vector of the power grid equipment in the preset time window is subjected to tensor decomposition analysis to obtain the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector; According to the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, the contact relationship matrix representing the connection closeness between the power grid equipment is obtained.

3. The method of claim 2, wherein, The dynamic topology characteristic vector of the power grid equipment in the preset time window is subjected to tensor decomposition analysis to obtain the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, which comprises: The dynamic topology characteristic vector of the power grid equipment in the preset time window is subjected to tensor decomposition to obtain decoupling characteristics; According to dynamic time warping, the decoupling characteristics are subjected to alignment processing to obtain dynamic topology characteristic data.

4. The method of claim 2, wherein, According to the dynamic topology characteristic data corresponding to the dynamic topology characteristic vector, the contact relationship matrix representing the connection closeness between the power grid equipment is obtained, which comprises: According to the power grid equipment in the power grid and the equipment connection relationship between the power grid equipment, an adjacency matrix is generated; According to the electrical quantity characteristics and non-electrical quantity characteristics in the dynamic topology characteristic data, the connection weight between the power grid equipment is determined; According to the connection weight between the power grid equipment and the adjacency matrix, a weighted adjacency matrix is generated; According to the matrix nodes in the weighted adjacency matrix and the matrix node connection relationship between the matrix nodes, a node relationship graph is obtained; According to the actual topological structure of the power grid, the confidence of the relationship node connection between the relationship nodes in the node relationship graph is determined; According to the confidence of the relationship node connection between the relationship nodes in the node relationship graph, the relationship node connection relationship between the relationship nodes in the node relationship graph is traversed, and a tie relationship matrix is obtained.

5. The method of claim 1, wherein, The method further comprises: According to the tie relationship matrix, the operating parameters and the historical fault data in the power grid, a key topological feature vector, a target feature vector and graph structure data are obtained; wherein the key topological feature vector is extracted from the tie relationship matrix after normalization processing; the target feature vector is obtained by principal component analysis on the tie relationship matrix, the operating parameters and the historical fault data in the power grid; and the graph structure data is extracted from a multi-modal graph constructed according to the tie relationship matrix, the operating parameters and the historical fault data in the power grid; The key topological feature vector, the target feature vector and the graph structure data are input into the generator of the generative adversarial network to generate initial simulated fault data; According to the distribution characteristics of the initial simulated fault data and the historical fault data, the initial simulated fault data is optimized and adjusted to obtain simulated fault data.

6. The method of claim 1, wherein, The method further comprises: According to the preset sliding time window, the data and the time sequence feature vector of the extended training data set in each sliding time window are determined; According to the time sequence feature vector of the extended training data set in each sliding time window, a time sequence feature sequence is obtained; The time sequence feature sequence is subjected to Pareto hierarchical optimization to obtain a hierarchical optimized Pareto frontier; According to the hierarchical optimized Pareto frontier and the engineering constraint condition, an initial target constant value solution set is obtained.

7. The method of claim 1, wherein, The method further comprises: The initial target constant value solution set is subjected to compliance verification to obtain a target constant value solution set, comprising: The initial target constant value solution set is subjected to compliance verification to determine a first constant value solution set representing a compliance state and a second constant value solution set representing a boundary non-compliance state in the initial target constant value solution set; The second constant value solution set representing the boundary non-compliance state is input into the anti-generation network for calibration processing to generate a calibrated constant value solution set representing a compliance state; 8. The method according to any one of claims 1 to 7, characterized in that, The first constant value solution set and the calibrated constant value solution set are subjected to screening processing to obtain a target constant value solution set. The method further comprises: According to the operating state of the power grid equipment in the power grid, an adjustment constant value corresponding to the target constant value solution set is determined; The adjustment constant value is subjected to safety verification processing, and the safety verification processed adjustment constant value is sent to the power grid equipment in the power grid; After power grid devices in the power grid are adjusted according to the adjustment set value, corresponding power grid data is collected by edge nodes distributed in the power grid; According to the corresponding power grid data collected by the edge nodes distributed in the power grid and the historical data of the power grid, a corresponding monitoring report is generated and displayed; wherein the data representing the digital thread full-link record in the process of generating the monitoring report are stored in the block chain.

9. An electronic device, comprising: Comprise: Memory, processor; The memory stores computer execution instructions; The processor executes the computer execution instructions stored in the memory, so that the processor executes the method of any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to realize the method of any one of claims 1-8.

Images