Intelligent power grid relay protection reliability evaluation and promotion control method and system

Abstract

The invention provides an intelligent power grid relay protection reliability evaluation and lifting control method and system, and relates to the technical field of intelligent power grid relay protection, and the method comprises the steps: collecting the operation data of a relay protection device, carrying out the preprocessing and multi-domain feature extraction, and inputting a deep hybrid neural network to obtain a comprehensive feature vector; a multi-stage fault diagnosis model is utilized to identify a fault mode, a collaborative anomaly and a reliability level at a device level, an interval level and a station level respectively, and a fault propagation path and influence are predicted based on a causal reasoning network. And finally, constructing a hierarchical cooperative control strategy, and through multi-objective optimization, distributed game optimization and hierarchical rolling optimization, realizing coordinated optimization of a protection fixed value and improving the reliability of a relay protection device, thereby reducing the fault propagation probability and ensuring safe and stable operation of a power grid.

Description

Technical Field

[0001] The present invention relates to a smart grid relay protection technology, and in particular to a smart grid relay protection reliability assessment and improvement control method and system. Background Art

[0002] Relay protection devices are an important part of smart grids to ensure the safe operation of power systems. With the expansion of the scale of power grids and the increase in the complexity of equipment, traditional relay protection technology faces problems such as inaccurate fault mode identification, difficulty in optimizing protection settings, and difficulty in capturing coordinated anomalies between devices, which affects the reliability and efficiency of relay protection devices.

[0003] At present, the fault diagnosis of relay protection devices mostly relies on static settings and empirical rules, which cannot adapt to the dynamic changes in the operating status of the power system. In addition, traditional methods fail to effectively combine multiple data sources and complex relationships between devices, resulting in insufficient accuracy in fault prediction and protection optimization.

[0004] Therefore, there is an urgent need for a new method that combines big data analysis, artificial intelligence, and optimization control technology to improve the fault diagnosis capability and system reliability of relay protection devices, thereby improving the safety and stability of smart grids. Summary of the invention

[0005] The embodiments of the present invention provide a smart grid relay protection reliability assessment and improvement control method and system, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is a smart grid relay protection reliability assessment and improvement control method, comprising:

[0008] The operation data of the relay protection device is collected, and the data is processed by an adaptive data preprocessing model to divide the data into a steady-state section, a transient section, and a fault section. The data of the steady-state section, the transient section, and the fault section are denoised by using an improved empirical mode decomposition method, and multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix. The enhanced feature matrix is ​​input into a deep hybrid neural network to extract spatial features, and the temporal association is captured by a cyclic branch. The spatial features and the temporal association are combined by an adaptive fusion layer to obtain a comprehensive feature vector;

[0009] The comprehensive feature vector is input into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomalies of adjacent protection devices are analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The failure mode, performance degradation trend, coordinated anomaly and reliability level are input into a causal reasoning network to establish a dynamic causal graph, and the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with a fuzzy cognitive graph, and fault warning information is output;

[0010] A hierarchical collaborative control strategy is constructed according to the fault warning information, and the protection setting optimization problem is converted into a multi-objective optimization model at the local control layer. The multi-objective optimization model is solved by a hybrid optimization algorithm to obtain candidate setting schemes; a distributed game optimization network is established at the global coordination layer based on the Nash equilibrium criterion, and the candidate setting schemes are coordinated and optimized by the potential game method; the setting schemes after coordinated optimization are input into a hierarchical rolling optimization controller, and the hierarchical rolling optimization controller establishes a time domain optimization problem based on model predictive control, realizes closed-loop control through online state estimation and feedback correction, and improves the reliability of the relay protection device.

[0011] The operation data of the relay protection device is collected and processed by an adaptive data preprocessing model to divide it into steady-state segment, transient segment and fault segment. The data of the steady-state segment, transient segment and fault segment are denoised by using an improved empirical mode decomposition method. The multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix including:

[0012] Collecting operation data of the relay protection device, the operation data including a primary device current signal, a primary device voltage signal, a protection device action signal, a communication link status signal, a circuit breaker mechanical characteristic signal, and a secondary circuit insulation impedance signal, and normalizing the operation data to obtain a standardized data sequence;

[0013] A variational modal decomposition model is established for the standardized data sequence. The variational modal decomposition model constructs a variational optimization objective function including a center frequency term and a bandwidth constraint term, and uses an alternating direction multiplier method to iteratively optimize the objective function to obtain an intrinsic mode function group. The mutual correlation coefficient of adjacent intrinsic mode functions is calculated based on the energy distribution characteristics of the intrinsic mode function group. An adaptive segmentation threshold is set according to the mutual correlation coefficient to divide the standardized data sequence into a steady-state segment, a transient segment, and a fault segment.

[0014] The data of the steady-state section, transient section and fault section are subjected to improved empirical mode decomposition denoising processing respectively, and the intrinsic mode components are processed by an adaptive soft threshold function based on data distribution, the parameters of the adaptive soft threshold function are adjusted according to the number of layers and energy density of the intrinsic mode components, and the local mean envelope is used to replace the spline interpolation to eliminate the endpoint effect, so as to obtain the denoised data;

[0015] Performing a wavelet packet transform on the denoised data, calculating the entropy of the wavelet packet decomposition coefficients, and selecting the node coefficient with the largest entropy value as the frequency domain feature; performing a Hilbert-Huang transform on the denoised data, calculating the instantaneous amplitude and instantaneous phase of the obtained analytical signal, and obtaining the instantaneous frequency according to the instantaneous phase to obtain the time-frequency feature;

[0016] The frequency domain feature vector and the time-frequency feature vector are adaptively weighted fused, and high-order statistical features based on kernel density estimation are introduced, including mean, variance, skewness, kurtosis, quantile features and entropy features. The principal component analysis method is used to reduce the dimension to form an enhanced feature matrix.

[0017] A variational modal decomposition model is established for the standardized data sequence. The variational modal decomposition model constructs a variational optimization objective function including a center frequency term and a bandwidth constraint term, and uses an alternating direction multiplier method to iteratively optimize the objective function to obtain an intrinsic mode function group. The mutual correlation coefficient of adjacent intrinsic mode functions is calculated based on the energy distribution characteristics of the intrinsic mode function group. An adaptive segmentation threshold is set according to the mutual correlation coefficient, and the standardized data sequence is divided into a steady-state segment, a transient segment, and a fault segment, including:

[0018] The standardized data sequence is discretely sampled and framed, each frame of data is converted into an analytical signal through Hilbert transform, the analytical signal is frequency modulated to obtain a center frequency sequence, and a reconstruction error matrix and a spectrum constraint matrix are constructed based on the center frequency sequence;

[0019] The reconstruction error matrix and the spectrum constraint matrix are input into a Wiener filter, the frequency response of the Wiener filter is adaptively adjusted according to the bandwidth constraint parameters, and the modal components are output; soft threshold shrinkage processing is applied to the modal components, and the soft threshold parameters are determined by cross-validation; the power spectrum centroid is calculated based on the processed modal components to obtain an updated center frequency sequence; the center frequency sequence is optimized by gradient iteration to obtain an intrinsic mode function group;

[0020] Calculating the instantaneous energy spectrum of each intrinsic mode function in the intrinsic mode function group to obtain an energy density distribution curve, and calculating the energy concentration and energy dispersion based on the energy density distribution curve to obtain an energy distribution characteristic matrix;

[0021] Extracting energy envelopes of adjacent intrinsic mode functions, calculating normalized cross-correlation functions of the energy envelopes, calculating peak values ​​of the normalized cross-correlation functions within a sliding time window, and generating a cross-correlation coefficient sequence;

[0022] Calculate the mean and standard deviation of the mutual correlation coefficient sequence, and set the mean as the reference threshold; multiply the energy concentration and energy dispersion in the energy distribution feature matrix to obtain a threshold adjustment factor; set the product of the reference threshold and the threshold adjustment factor as the adaptive segmentation threshold;

[0023] In the sliding time window, the mutual correlation coefficient sequence is compared with the corresponding adaptive segmentation threshold, and when the mutual correlation coefficients of three consecutive windows are all less than the adaptive segmentation threshold, they are marked as candidate mutation points; the local energy density of the candidate mutation point is calculated, and the final modal mutation point position is determined based on the minimum energy density; the standardized data sequence is divided into a steady-state segment, a transient segment and a fault segment according to the modal mutation point.

[0024] The comprehensive feature vector is input into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomaly of adjacent protection devices is analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level, including:

[0025] Constructing a multi-stage fault diagnosis model, wherein the multi-stage fault diagnosis model includes a device-level diagnosis sub-model, a bay-level diagnosis sub-model, and a station-level diagnosis sub-model;

[0026] The device-level diagnosis sub-model adopts a deep belief network, inputs the comprehensive feature vector into the deep belief network, performs layer-by-layer pre-training through a restricted Boltzmann machine, optimizes the network weights using a contrastive divergence algorithm and a stochastic gradient descent method, introduces label data for network fine-tuning, and obtains a failure mode probability vector and a performance degradation index for a single protection device;

[0027] The interval-level diagnosis sub-model adopts a conditional random field structure, takes the fault mode probability vector and performance degradation index as node features, constructs a spatial potential function based on the primary equipment topological relationship, and constructs a time series potential function based on the operating state correlation. The fault mode probability vector, performance degradation index, spatial potential function and time series potential function are input into the conditional random field structure, and the model parameters are optimized by maximum likelihood estimation, and the coordinated abnormality probability matrix of adjacent devices is output;

[0028] The station-level diagnosis submodel adopts a hierarchical clustering structure, and fuses the fault mode probability vector, performance degradation index and collaborative abnormality probability matrix to construct a state characterization matrix. The state distance between devices is calculated based on the state characterization matrix, and the hierarchical clustering algorithm with adaptive distance measurement is used to group the devices. The optimal number of groups is determined by the silhouette coefficient to obtain the reliability level of the protection device.

[0029] The failure mode, performance degradation trend, collaborative anomaly and reliability level are input into the causal reasoning network to establish a dynamic causal graph. The fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with the fuzzy cognitive graph. The fault warning information output includes:

[0030] The failure modes, performance degradation trends, coordinated anomalies and reliability levels of relay protection devices are used to construct a node set of a causal reasoning network. The conditional probability table between nodes is calculated based on the conditional independence assumption. The node status is updated online using the recursive Bayesian estimation method. The conditional probability parameters are optimized based on historical observation data, and a dynamic causal graph that reflects the dynamic correlation between nodes is established.

[0031] Based on the dynamic causal graph, the causal support of the node is calculated by forward propagation, the diagnostic support of the node is calculated by backward propagation, the causal support and the diagnostic support are combined by a normalization factor to update the node belief value, the propagation path of the fault in the network is calculated according to the node belief value and conditional probability, and the fault propagation probability is calculated based on the transmission probability between nodes and the intermediate node status;

[0032] The fault propagation path and fault propagation probability are mapped into concept nodes and weight matrices of fuzzy cognitive graphs respectively, a fuzzy relationship matrix including fault source, fault type, impact range, and propagation path is constructed, and the target node state vector is obtained through fuzzy synthesis operation and hyperbolic tangent activation function iterative update;

[0033] The direct influence is calculated based on the weight matrix of the fuzzy cognitive graph and the target node state vector, the indirect influence is calculated by combining the fault propagation probability and the preset attenuation factor, and the direct influence and the indirect influence are combined by an exponential weighting method to obtain the cumulative influence of the fault;

[0034] An early warning decision matrix is ​​constructed based on the probability of fault propagation and the cumulative impact of faults. The fuzzy hierarchical analysis is used to determine the warning level threshold. The warning information is divided into four levels. The warning information including fault source equipment, propagation path, risk level and disposal measures is output in a hierarchical and progressive manner according to the warning level.

[0035] Based on the dynamic causal graph, the causal support of the node is calculated by forward propagation, the diagnostic support of the node is calculated by backward propagation, the causal support and the diagnostic support are combined by a normalization factor to update the node belief value, the propagation path of the fault in the network is calculated according to the node belief value and the conditional probability, and the fault propagation probability is calculated based on the transmission probability between nodes and the intermediate node state, including:

[0036] Obtain the state information of nodes in the dynamic causal graph and construct the conditional probability matrix between nodes;

[0037] Identify the parent node set of the current node in the dynamic causal graph, calculate the information transmission delay time from each node in the parent node set to the current node, determine the timing weight based on the information transmission delay time, combine the timing weight with the physical coupling strength to generate a transmission coefficient, and obtain the causal support of the current node through forward propagation calculation based on the transmission coefficient and the initial belief value of each node in the parent node set;

[0038] Extracting a subnode set of the current node from the dynamic causal graph, calculating the state deviation of each node in the subnode set, constructing a feedback influence function based on the state deviation, multiplying the feedback influence function by the distance attenuation coefficient between nodes to obtain a diagnosis weight, and obtaining the diagnosis support of the current node through back propagation calculation based on the diagnosis weight and the belief value of each node in the subnode set;

[0039] Calculating an adaptive normalization factor based on the system operation status, weighting and combining the causal support and the diagnostic support through the adaptive normalization factor, and updating the belief value of the current node;

[0040] Based on the updated current node belief value and the conditional probability matrix, the fault transmission probability between adjacent nodes is calculated, a directed propagation network is constructed based on the fault transmission probability, and a fault propagation path is extracted from the directed propagation network;

[0041] The real-time state value of each node on the critical fault propagation path is obtained, the real-time state value is converted into a state function, and the cumulative propagation probability on the fault propagation path is calculated according to the state function and the fault transmission probability to obtain the final fault propagation probability.

[0042] A hierarchical collaborative control strategy is constructed according to the fault warning information. The protection setting optimization problem is converted into a multi-objective optimization model at the local control layer. A hybrid optimization algorithm is used to solve the multi-objective optimization model to obtain a candidate setting scheme. A distributed game optimization network is established at the global coordination layer based on the Nash equilibrium criterion. The potential game method is used to coordinate and optimize the candidate setting schemes, including:

[0043] Receiving system fault warning information, and extracting the fault type, fault location, fault current, fault duration and fault phase angle in the fault warning information as feature quantities;

[0044] Calculate the protection reliability index according to the fault type and fault current, calculate the protection sensitivity index according to the fault location and fault current, and calculate the protection selectivity index according to the fault duration and protection coordination time interval;

[0045] Constructing a multi-objective optimization model at the local control layer based on the protection reliability index, the protection sensitivity index and the protection selectivity index;

[0046] An adaptive crossover operator is set based on population density to solve the multi-objective optimization model to obtain a crossover solution, an adaptive mutation operator is set based on fitness value to mutate the crossover solution to obtain a mutated solution, a reference point set is constructed based on the mutated solution, the Euclidean distance from the mutated solution to the reference point set is calculated, and the solution with the smallest Euclidean distance is selected as a candidate fixed value solution;

[0047] At the global coordination layer, the protection unit is set as a game participant, the candidate setting scheme is set as the strategy space of the game participant, and a distributed game optimization network is constructed;

[0048] In the distributed game optimization network, the reliability benefits of the game participants when selecting strategies are calculated based on the protection sensitivity index, the coordination benefits of the game participants and adjacent protection units are calculated based on the protection coordination time interval, and the interference losses suffered by the game participants are calculated based on the electrical distance between adjacent protection units;

[0049] Taking the reliability benefit, coordination benefit and interference loss of the game participants as comprehensive benefits, constructing the utility function of the game participants;

[0050] Calculating the impedance values ​​between the protection units in the distributed game optimization network to obtain the degree of electrical connection, and determining the coupling coefficient between the game participants according to the degree of electrical connection; calculating the action time difference between the protection units to obtain the coordination relationship, and determining the interactive utility between the game participants according to the coordination relationship; constructing a potential function with the utility function, coupling coefficient and interactive utility of the game participants;

[0051] An initial strategy combination of game participants is generated in the distributed game optimization network, the utility function values ​​and potential function values ​​of the game participants under the initial strategy combination are calculated, the initial strategy combination is updated based on the utility function value to obtain an optimized response strategy, a new potential function value under the optimized response strategy is calculated, and when the difference between the new potential function value and the potential function value of the previous iteration is less than a preset threshold, the optimized response strategy is output as a coordinated optimization constant value scheme that meets the Nash equilibrium criterion.

[0052] A second aspect of an embodiment of the present invention provides a smart grid relay protection reliability assessment and improvement control system, comprising:

[0053] The first unit is used to collect the operation data of the relay protection device, and process the data through an adaptive data preprocessing model to divide it into a steady-state segment, a transient segment and a fault segment. The data of the steady-state segment, the transient segment and the fault segment are denoised by using an improved empirical mode decomposition method, and multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix. The enhanced feature matrix is ​​input into a deep hybrid neural network to extract spatial features, and time series associations are captured through cyclic branches. The spatial features and time series associations are combined by an adaptive fusion layer to obtain a comprehensive feature vector;

[0054] The second unit is used to input the comprehensive feature vector into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomalies of adjacent protection devices are analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The failure mode, performance degradation trend, coordinated anomaly and reliability level are input into a causal reasoning network to establish a dynamic causal graph, and the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with a fuzzy cognitive graph, and fault warning information is output;

[0055] The third unit is used to construct a hierarchical collaborative control strategy according to the fault warning information, transform the protection setting optimization problem into a multi-objective optimization model at the local control layer, and use a hybrid optimization algorithm to solve the multi-objective optimization model to obtain a candidate setting scheme; establish a distributed game optimization network based on the Nash equilibrium criterion at the global coordination layer, and use the potential game method to coordinate and optimize the candidate setting schemes; input the coordinated and optimized setting schemes into a hierarchical rolling optimization controller, and the hierarchical rolling optimization controller establishes a time domain optimization problem based on model predictive control, realizes closed-loop control through online state estimation and feedback correction, and completes the reliability improvement of the relay protection device.

[0056] A third aspect of the embodiments of the present invention

[0057] An electronic device is provided, comprising:

[0058] processor;

[0059] a memory for storing processor-executable instructions;

[0060] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0061] A fourth aspect of the embodiments of the present invention is:

[0062] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0063] In this embodiment, by combining multiple signal processing methods and deep learning models, it is possible to more accurately identify the failure mode, performance degradation trend, and coordinated anomalies of adjacent protection devices of a single protection device, and predict the fault propagation path and impact degree based on the causal reasoning network, thereby improving the accuracy of fault warning. The use of multi-objective optimization models and distributed game optimization networks to optimize protection settings can achieve coordinated optimization of settings at the local and global levels, and improve the reliability and adaptability of protection devices. A hierarchical rolling optimization controller is used to achieve online state estimation and feedback correction based on model predictive control, which can dynamically adjust the protection settings according to the system operating status, achieve closed-loop control, and further improve the reliability and stability of the relay protection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flow chart of a method for evaluating and improving the reliability of relay protection in a smart grid according to an embodiment of the present invention;

[0065] Figure 2 It is a structural schematic diagram of a smart grid relay protection reliability assessment and improvement control system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0068] Figure 1 FIG. 1 is a flow chart of a method for evaluating and improving the reliability of relay protection in a smart grid according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0069] S101. Collect the operation data of the relay protection device, and process the data through the adaptive data preprocessing model to divide it into a steady-state section, a transient section and a fault section. The data of the steady-state section, the transient section and the fault section are denoised by using an improved empirical mode decomposition method, and multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix. The enhanced feature matrix is ​​input into a deep hybrid neural network to extract spatial features, and the time series association is captured through a loop branch. The spatial features and the time series association are combined by an adaptive fusion layer to obtain a comprehensive feature vector;

[0070] S102. Input the comprehensive feature vector into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomaly of adjacent protection devices is analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The failure mode, performance degradation trend, coordinated anomaly and reliability level are input into a causal reasoning network, a dynamic causal graph is established, the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with a fuzzy cognitive graph, and fault warning information is output;

[0071] S103. Construct a hierarchical collaborative control strategy based on the fault warning information, transform the protection setting optimization problem into a multi-objective optimization model at the local control layer, and use a hybrid optimization algorithm to solve the multi-objective optimization model to obtain candidate setting schemes; establish a distributed game optimization network based on the Nash equilibrium criterion at the global coordination layer, and use the potential game method to coordinate and optimize the candidate setting schemes; input the coordinated and optimized setting schemes into a hierarchical rolling optimization controller, and the hierarchical rolling optimization controller establishes a time domain optimization problem based on model predictive control, realizes closed-loop control through online state estimation and feedback correction, and completes the reliability improvement of the relay protection device.

[0072] Among them, the loop branch refers to the introduction of a feedback mechanism in the network structure to capture the temporal correlation in the data. This structure can effectively process data with time series characteristics, such as the operating data of power equipment.

[0073] Among them, the coordinated optimized setting scheme will be input into the hierarchical rolling optimization controller. The hierarchical rolling optimization controller can make timely adjustments during the system operation process through dynamic estimation of the power system state to ensure the adaptability of the protection setting under different operating conditions.

[0074] In an optional implementation, the operation data of the relay protection device is collected, and the data is processed by an adaptive data preprocessing model to divide it into a steady-state segment, a transient segment, and a fault segment. The data of the steady-state segment, the transient segment, and the fault segment are denoised by using an improved empirical mode decomposition method, and the multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix, including:

[0075] Collecting operation data of the relay protection device, the operation data including a primary device current signal, a primary device voltage signal, a protection device action signal, a communication link status signal, a circuit breaker mechanical characteristic signal, and a secondary circuit insulation impedance signal, and normalizing the operation data to obtain a standardized data sequence;

[0076] A variational modal decomposition model is established for the standardized data sequence. The variational modal decomposition model constructs a variational optimization objective function including a center frequency term and a bandwidth constraint term, and uses an alternating direction multiplier method to iteratively optimize the objective function to obtain an intrinsic mode function group. The mutual correlation coefficient of adjacent intrinsic mode functions is calculated based on the energy distribution characteristics of the intrinsic mode function group. An adaptive segmentation threshold is set according to the mutual correlation coefficient to divide the standardized data sequence into a steady-state segment, a transient segment, and a fault segment.

[0077] The data of the steady-state section, transient section and fault section are subjected to improved empirical mode decomposition denoising processing respectively, and the intrinsic mode components are processed by an adaptive soft threshold function based on data distribution, the parameters of the adaptive soft threshold function are adjusted according to the number of layers and energy density of the intrinsic mode components, and the local mean envelope is used to replace the spline interpolation to eliminate the endpoint effect, so as to obtain the denoised data;

[0078] Performing a wavelet packet transform on the denoised data, calculating the entropy of the wavelet packet decomposition coefficients, and selecting the node coefficient with the largest entropy value as the frequency domain feature; performing a Hilbert-Huang transform on the denoised data, calculating the instantaneous amplitude and instantaneous phase of the obtained analytical signal, and obtaining the instantaneous frequency according to the instantaneous phase to obtain the time-frequency feature;

[0079] The frequency domain feature vector and the time-frequency feature vector are adaptively weighted fused, and high-order statistical features based on kernel density estimation are introduced, including mean, variance, skewness, kurtosis, quantile features and entropy features. The principal component analysis method is used to reduce the dimension to form an enhanced feature matrix.

[0080] Exemplarily, the operation data of the relay protection device is first collected, including the current signal of the primary device, the voltage signal of the primary device, the action signal of the protection device, the communication link status signal, the mechanical characteristic signal of the circuit breaker and the insulation impedance signal of the secondary circuit. These signals are acquired by sensors and data acquisition devices and stored in digital form. For example, the current signal can be collected by a current transformer, the voltage signal can be collected by a voltage transformer, the action signal of the protection device can be collected by a digital input module, the communication link status signal can be collected by a network monitoring module, the mechanical characteristic signal of the circuit breaker can be collected by a travel switch and a position sensor, and the insulation impedance signal of the secondary circuit can be collected by an insulation monitoring device. The collected data contains a timestamp for subsequent analysis.

[0081] The collected operation data is normalized. The purpose of normalization is to eliminate the influence of different signal dimensions and magnitudes and convert the data into a unified range. For example, the min-max normalization method can be used to scale the value of each signal to between 0 and 1.

[0082] A variational mode decomposition model is established for the normalized data sequence. The model constructs a variational optimization objective function containing a center frequency term and a bandwidth constraint term, and uses the alternating direction multiplier method to iteratively optimize the objective function to obtain a set of intrinsic mode functions. The center frequency term is used to describe the center frequency of each intrinsic mode function, and the bandwidth constraint term is used to limit the bandwidth of each intrinsic mode function. The alternating direction multiplier method is an iterative optimization algorithm that decomposes a complex optimization problem into multiple simple sub-problems, and solves these sub-problems alternately to finally obtain the global optimal solution.

[0083] The mutual correlation coefficients of adjacent intrinsic mode functions are calculated based on the energy distribution characteristics of the eigenmode function group. The mutual correlation coefficient is used to measure the similarity between two intrinsic mode functions. An adaptive segmentation threshold is set according to the mutual correlation coefficient, and the standardized data sequence is divided into a steady-state segment, a transient segment, and a fault segment. For example, if the mutual correlation coefficient of two adjacent intrinsic mode functions is greater than a preset threshold, the two intrinsic mode functions are considered to belong to the same segment. Assuming that the calculated mutual correlation coefficient sequence is [0.9, 0.8, 0.2, 0.9, 0.7] and the preset threshold is 0.5, the data sequence can be divided into three segments: [0.9, 0.8], [0.2], [0.9, 0.7].

[0084] The data of the steady-state section, transient section and fault section are denoised by improved empirical mode decomposition. The intrinsic mode components are processed by an adaptive soft threshold function based on data distribution, and the parameters of the function are adjusted according to the number of layers and energy density of the intrinsic mode components. The local mean envelope is used to replace the spline interpolation to eliminate the endpoint effect and obtain the denoised data. For example, for an intrinsic mode component, if its energy density is low, it is considered that the component mainly contains noise, and its amplitude is reduced; conversely, if its energy density is high, it is considered that the component mainly contains useful signals, and its amplitude is retained or amplified.

[0085] Perform wavelet packet transform on the denoised data, calculate the entropy of the wavelet packet decomposition coefficients, and select the node coefficient with the largest entropy value as the frequency domain feature. Wavelet packet transform is a signal decomposition method that can decompose a signal into sub-signals of different frequency bands. Entropy is used to measure the complexity of a signal. For example, if the entropy value of a node coefficient is large, it is considered that the node contains more information.

[0086] The denoised data is subjected to Hilbert-Huang transform, and the instantaneous amplitude and instantaneous phase of the analytical signal are calculated. The instantaneous frequency is obtained according to the instantaneous phase to obtain the time-frequency characteristics. Hilbert-Huang transform is a time-frequency analysis method that can decompose the signal into a series of intrinsic mode functions and calculate the instantaneous frequency of each intrinsic mode function.

[0087] The frequency domain feature vector and the time-frequency feature vector are adaptively weighted and fused, and high-order statistical features based on kernel density estimation are introduced, including mean, variance, skewness, kurtosis, quantile features and entropy features. The principal component analysis method is used to reduce the dimension to form an enhanced feature matrix. The purpose of adaptive weighted fusion is to assign different weights according to the importance of different features. The principal component analysis method is a dimensionality reduction method that can project high-dimensional data into a low-dimensional space and retain the main information.

[0088] In this embodiment, by extracting and enhancing multi-domain features of the operating data, the operating status of the power system can be more comprehensively reflected, thereby improving the accuracy of fault identification. By adaptive data preprocessing and denoising, the influence of noise and interference can be effectively suppressed, and the anti-interference ability of the relay protection device can be improved. By reducing the feature dimension and enhancing the construction of the feature matrix, the amount of calculation can be reduced and the operating efficiency of the relay protection device can be improved.

[0089] In an optional implementation, a variational modal decomposition model is established for the standardized data sequence, the variational modal decomposition model constructs a variational optimization objective function including a center frequency term and a bandwidth constraint term, and uses an alternating direction multiplier method to iteratively optimize the objective function to obtain an intrinsic mode function group, calculates the mutual correlation coefficient of adjacent intrinsic mode functions based on the energy distribution characteristics of the intrinsic mode function group, sets an adaptive segmentation threshold according to the mutual correlation coefficient, and divides the standardized data sequence into a steady-state segment, a transient segment, and a fault segment, including:

[0090] The standardized data sequence is discretely sampled and framed, each frame of data is converted into an analytical signal through Hilbert transform, the analytical signal is frequency modulated to obtain a center frequency sequence, and a reconstruction error matrix and a spectrum constraint matrix are constructed based on the center frequency sequence;

[0091] The reconstruction error matrix and the spectrum constraint matrix are input into a Wiener filter, the frequency response of the Wiener filter is adaptively adjusted according to the bandwidth constraint parameters, and the modal components are output; soft threshold shrinkage processing is applied to the modal components, and the soft threshold parameters are determined by cross-validation; the power spectrum centroid is calculated based on the processed modal components to obtain an updated center frequency sequence; the center frequency sequence is optimized by gradient iteration to obtain an intrinsic mode function group;

[0092] Calculating the instantaneous energy spectrum of each intrinsic mode function in the intrinsic mode function group to obtain an energy density distribution curve, and calculating the energy concentration and energy dispersion based on the energy density distribution curve to obtain an energy distribution characteristic matrix;

[0093] Extracting energy envelopes of adjacent intrinsic mode functions, calculating normalized cross-correlation functions of the energy envelopes, calculating peak values ​​of the normalized cross-correlation functions within a sliding time window, and generating a cross-correlation coefficient sequence;

[0094] Calculate the mean and standard deviation of the mutual correlation coefficient sequence, and set the mean as the reference threshold; multiply the energy concentration and energy dispersion in the energy distribution feature matrix to obtain a threshold adjustment factor; set the product of the reference threshold and the threshold adjustment factor as the adaptive segmentation threshold;

[0095] In the sliding time window, the mutual correlation coefficient sequence is compared with the corresponding adaptive segmentation threshold, and when the mutual correlation coefficients of three consecutive windows are all less than the adaptive segmentation threshold, they are marked as candidate mutation points; the local energy density of the candidate mutation point is calculated, and the final modal mutation point position is determined based on the minimum energy density; the standardized data sequence is divided into a steady-state segment, a transient segment and a fault segment according to the modal mutation point.

[0096] Exemplarily, first, the standardized data sequence is discretely sampled and framed. For example, for a signal with a sampling frequency of 1kHz, 100 sampling points can be used as a frame, and the frame shift is 50 sampling points. Then, each frame of data is converted into an analytical signal through a Hilbert transform. The analytical signal contains the original signal and its Hilbert transform, which can be used to extract the instantaneous amplitude and instantaneous frequency of the signal. The analytical signal is frequency modulated to obtain a center frequency sequence. The center frequency sequence reflects the change of the signal frequency over time. A reconstruction error matrix and a spectrum constraint matrix are constructed based on the center frequency sequence. The reconstruction error matrix measures the difference between the reconstructed signal and the original signal, and the spectrum constraint matrix is ​​used to limit the bandwidth of the modal component.

[0097] Next, the reconstruction error matrix and the spectrum constraint matrix are input into the Wiener filter. The frequency response of the Wiener filter is adaptively adjusted according to the bandwidth constraint parameter, and the modal component is output. The bandwidth constraint parameter controls the frequency range of the modal component. For example, the bandwidth constraint parameter can be set to 0.1, indicating that the bandwidth of the modal component does not exceed 0.1 times the sampling frequency. Soft threshold shrinkage processing is applied to the modal component, and the soft threshold parameter is determined by cross-validation. Soft threshold shrinkage can remove noise and interference in the modal component. For example, 5-fold cross-validation can be used to select the best soft threshold parameter. Based on the processed modal component, the power spectrum centroid is calculated to obtain an updated center frequency sequence. The power spectrum centroid reflects the main frequency components of the modal component. The center frequency sequence is optimized by gradient iteration to obtain the intrinsic mode function group. Gradient iteration can gradually adjust the center frequency sequence to make it close to the optimal solution.

[0098] Then, the instantaneous energy spectrum of each intrinsic mode function in the intrinsic mode function group is calculated to obtain the energy density distribution curve. The instantaneous energy spectrum reflects the distribution of signal energy at different frequencies. Based on the energy density distribution curve, the energy concentration and energy dispersion are calculated to obtain the energy distribution characteristic matrix. The energy concentration indicates the concentration degree of energy near the main frequency component, and the energy dispersion indicates the dispersion degree of energy at different frequencies. For example, the variance of the energy density distribution curve can be calculated as an indicator of energy dispersion.

[0099] Extract the energy envelope of adjacent intrinsic mode functions and calculate the normalized cross-correlation function of the energy envelope. The energy envelope reflects the change of signal energy over time. Calculate the peak value of the normalized cross-correlation function within the sliding time window to generate a cross-correlation coefficient sequence. The sliding time window can be set to 10 frames, i.e. 1 second. The cross-correlation coefficient sequence reflects the correlation between adjacent intrinsic mode functions.

[0100] Calculate the mean and standard deviation of the mutual correlation coefficient sequence, and set the mean as the reference threshold. At the same time, multiply the energy concentration and energy dispersion in the energy distribution feature matrix to obtain the threshold adjustment factor. Set the product of the reference threshold and the threshold adjustment factor as the adaptive segmentation threshold. The adaptive segmentation threshold can be dynamically adjusted according to the energy distribution characteristics of the signal.

[0101] In the sliding time window, the mutual correlation coefficient sequence is compared with the corresponding adaptive segmentation threshold. When the mutual correlation coefficients of three consecutive windows are all less than the adaptive segmentation threshold, they are marked as candidate mutation points. The local energy density of the candidate mutation point is calculated, and the final modal mutation point position is determined based on the minimum energy density. The standardized data sequence is divided into steady-state segment, transient segment and fault segment according to the modal mutation point.

[0102] Assume that a vibration signal containing a fault is analyzed. Before the fault occurs, the energy of the signal is mainly concentrated in the low-frequency band, and the cross-correlation coefficient is high. After the fault occurs, the energy distribution of the signal changes, the high-frequency component increases, and the cross-correlation coefficient decreases. By comparing the cross-correlation coefficient with the adaptive segmentation threshold, the time point of the fault can be accurately identified, and the signal can be divided into a steady-state segment, a transient segment, and a fault segment.

[0103] In this embodiment, by introducing the center frequency term and the bandwidth constraint term, different modal components in the signal can be extracted more accurately and the modal aliasing phenomenon can be reduced. By adaptively adjusting the segmentation threshold based on the energy distribution characteristics, the characteristics of different signals can be better adapted and the accuracy of state division can be improved. By using the alternating direction multiplier method for iterative optimization, the computational complexity can be effectively reduced and the operation efficiency of the algorithm can be improved.

[0104] In an optional implementation, the comprehensive feature vector is input into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomaly of adjacent protection devices is analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The model includes:

[0105] Constructing a multi-stage fault diagnosis model, wherein the multi-stage fault diagnosis model includes a device-level diagnosis sub-model, a bay-level diagnosis sub-model, and a station-level diagnosis sub-model;

[0106] The device-level diagnosis sub-model adopts a deep belief network, inputs the comprehensive feature vector into the deep belief network, performs layer-by-layer pre-training through a restricted Boltzmann machine, optimizes the network weights using a contrastive divergence algorithm and a stochastic gradient descent method, introduces label data for network fine-tuning, and obtains a failure mode probability vector and a performance degradation index for a single protection device;

[0107] The interval-level diagnosis sub-model adopts a conditional random field structure, takes the fault mode probability vector and performance degradation index as node features, constructs a spatial potential function based on the primary equipment topological relationship, and constructs a time series potential function based on the operating state correlation. The fault mode probability vector, performance degradation index, spatial potential function and time series potential function are input into the conditional random field structure, and the model parameters are optimized by maximum likelihood estimation, and the coordinated abnormality probability matrix of adjacent devices is output;

[0108] The station-level diagnosis submodel adopts a hierarchical clustering structure, and fuses the fault mode probability vector, performance degradation index and collaborative abnormality probability matrix to construct a state characterization matrix. The state distance between devices is calculated based on the state characterization matrix, and the hierarchical clustering algorithm with adaptive distance measurement is used to group the devices. The optimal number of groups is determined by the silhouette coefficient to obtain the reliability level of the protection device.

[0109] Exemplarily, first, it is necessary to obtain a comprehensive feature vector of the protection device. This can be achieved in a variety of ways, for example, collecting physical quantities such as current, voltage, and temperature of the protection device, and combining historical data such as the operating time and number of actions of the protection device to extract features reflecting the state of the protection device. For example, the current value, voltage value, and ambient temperature of a certain protection device can be collected, and combined with the operating time and number of actions of the protection device to form a six-dimensional feature vector: [10A, 220V, 25°C, 3600h, 100 times, 0 times]. Among them, the last digit "0 times" means that the protection device did not malfunction during the statistical period.

[0110] Next, a multi-stage fault diagnosis model is constructed, which includes three sub-models: device level, bay level and station level.

[0111] The device-level diagnosis sub-model adopts a deep belief network structure. The obtained comprehensive feature vector is input into a pre-trained deep belief network. The network is optimized by pre-training and fine-tuning layer by layer, and finally outputs the failure mode probability vector and performance degradation index of a single protection device. For example, for the input feature vector [10A, 220V, 25℃, 3600h, 100 times, 0 times], the failure mode probability vector output by the device-level diagnosis sub-model may be [0.1, 0.05, 0.8, 0.05], representing the probabilities of the four failure modes of normal, aging, refusal to operate and misoperation respectively. The performance degradation index can be 0.2, indicating that the performance of the device has declined to a certain extent.

[0112] The interval-level diagnosis submodel adopts a conditional random field structure. The fault mode probability vector and performance degradation index output by the device-level diagnosis submodel are used as node features, and the spatial potential function and time series potential function are constructed in combination with the topological relationship and operating status correlation of the primary equipment. For example, if two protection devices are geographically close, their spatial potential function values ​​are higher. If the operating states of the two protection devices are highly correlated in time, their time series potential function values ​​are higher. By inputting this information into the conditional random field model, the collaborative abnormality probability matrix of adjacent devices can be obtained. For example, for two adjacent protection devices, their collaborative abnormality probability matrix may be [[0.9, 0.1], [0.1, 0.9]], indicating that the probability of the two devices being abnormal at the same time is high.

[0113] The station-level diagnosis submodel adopts a hierarchical clustering structure. The fault mode probability vector output by the device-level diagnosis submodel, the performance degradation index and the collaborative abnormality probability matrix output by the interval-level diagnosis submodel are fused to construct a state characterization matrix. The state distance between devices is calculated based on the matrix, and the devices are grouped using a hierarchical clustering algorithm with an adaptive distance metric. For example, the state characterization matrix can reflect the state similarity between different protection devices. Through the hierarchical clustering algorithm, protection devices with similar states can be divided into the same group. Finally, the optimal number of groups is determined by the silhouette coefficient to obtain the reliability level of the protection device. For example, the protection device can be divided into three reliability levels: high, medium, and low, corresponding to different degrees of reliability.

[0114] In this embodiment, the information at the device level, bay level and station level is comprehensively considered, and the operating status of the protection device can be analyzed more comprehensively, thereby improving the accuracy of fault diagnosis. By analyzing the performance degradation trend of the protection device, early warning of faults can be achieved to avoid power outages caused by protection device failures. By classifying the reliability level of the protection device, the equipment maintenance strategy can be optimized, the maintenance efficiency can be improved, and the maintenance cost can be reduced.

[0115] In an optional implementation, the failure mode, performance degradation trend, collaborative anomaly and reliability level are input into a causal reasoning network to establish a dynamic causal graph, and the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with the fuzzy cognitive graph. The output of fault warning information includes:

[0116] The failure modes, performance degradation trends, coordinated anomalies and reliability levels of relay protection devices are used to construct a node set of a causal reasoning network. The conditional probability table between nodes is calculated based on the conditional independence assumption. The node status is updated online using the recursive Bayesian estimation method. The conditional probability parameters are optimized based on historical observation data, and a dynamic causal graph that reflects the dynamic correlation between nodes is established.

[0117] Based on the dynamic causal graph, the causal support of the node is calculated by forward propagation, the diagnostic support of the node is calculated by backward propagation, the causal support and the diagnostic support are combined by a normalization factor to update the node belief value, the propagation path of the fault in the network is calculated according to the node belief value and conditional probability, and the fault propagation probability is calculated based on the transmission probability between nodes and the intermediate node status;

[0118] The fault propagation path and fault propagation probability are mapped into concept nodes and weight matrices of fuzzy cognitive graphs respectively, a fuzzy relationship matrix including fault source, fault type, impact range, and propagation path is constructed, and the target node state vector is obtained through fuzzy synthesis operation and hyperbolic tangent activation function iterative update;

[0119] The direct influence is calculated based on the weight matrix of the fuzzy cognitive graph and the target node state vector, the indirect influence is calculated by combining the fault propagation probability and the preset attenuation factor, and the direct influence and the indirect influence are combined by an exponential weighting method to obtain the cumulative influence of the fault;

[0120] An early warning decision matrix is ​​constructed based on the probability of fault propagation and the cumulative impact of faults. The fuzzy hierarchical analysis is used to determine the warning level threshold. The warning information is divided into four levels. The warning information including fault source equipment, propagation path, risk level and disposal measures is output in a hierarchical and progressive manner according to the warning level.

[0121] Exemplarily, first, a causal reasoning network is constructed. Various failure modes of relay protection devices, such as refusal to operate, malfunction, slow operation, etc., as well as performance degradation trends, such as contact wear, insulation aging, etc., as well as coordinated abnormalities, such as abnormalities in multiple devices at the same time, and reliability levels, are used as nodes of the causal reasoning network. Then, based on the conditional independence assumption, the conditional probability between nodes is calculated. For example, the occurrence of contact wear will increase the probability of slow operation. The recursive Bayesian estimation method is used to update the node status online, and the conditional probability parameters are optimized based on historical observation data, such as regular inspection records and operation data. Finally, a dynamic causal graph reflecting the dynamic association relationship between nodes is established. For example, a simplified causal graph can include a causal chain in which "contact wear" leads to "increased contact resistance", which in turn leads to "slow operation" and "heating", and finally leads to "device failure".

[0122] Next, the fault propagation path and probability are predicted based on the dynamic causal graph. Forward propagation is used to calculate the causal support of the node, for example, the support of "contact wear" for "slow action". Backward propagation is used to calculate the diagnostic support of the node, for example, the support of "slow action" for "contact wear". The causal support and diagnostic support are combined through the normalization factor to update the node belief value, indicating the confidence that the node is in a certain state. The propagation path of the fault in the network is calculated based on the node belief value and conditional probability. For example, if the belief value of "contact wear" is high, it is believed that the probability of "slow action" will also be high, thus forming a propagation path. Then, the fault propagation probability is calculated based on the inter-node transmission probability and the intermediate node state. For example, if the transmission probability from "contact wear" to "slow action" is 0.8 and the transmission probability from "slow action" to "device failure" is 0.6, then the transmission probability of "contact wear" leading to "device failure" is 0.8*0.6=0.48.

[0123] Then, the fault propagation path and fault propagation probability are mapped to the fuzzy cognitive graph. The fault source, fault type, impact range, propagation path, etc. are used as concept nodes of the fuzzy cognitive graph, and the fault propagation probability is used as the weight matrix. For example, if the propagation probability of "contact wear" leading to "device failure" is 0.48, the weight of the edge connecting the two concept nodes is set to 0.48. Construct a fuzzy relationship matrix containing the fault source, fault type, impact range, and propagation path, and obtain the target node state vector through fuzzy synthesis operation and hyperbolic tangent activation function iterative update. For example, if "device failure" affects "line tripping", the state vector of "line tripping" will be updated as the state vector of "device failure" changes.

[0124] Next, evaluate the impact of the fault. Calculate the direct impact based on the weight matrix of the fuzzy cognitive graph and the target node state vector. For example, the direct impact of "device failure" on "line tripping". Calculate the indirect impact by combining the fault propagation probability with the preset attenuation factor. For example, the indirect impact of "contact wear" on "line tripping" through "device failure" will decay as the propagation path length increases. Use the exponential weighting method to combine the direct impact and indirect impact to get the cumulative impact of the fault.

[0125] Finally, the fault warning information is output. The warning decision matrix is ​​constructed based on the fault propagation probability and the cumulative impact of the fault. The warning level threshold is determined by fuzzy hierarchical analysis, and the warning information is divided into four levels, such as: slight, general, serious, and urgent. According to the warning level, the warning information including the fault source equipment, propagation path, risk level, and disposal measures is output in a hierarchical and progressive manner. For example, if the propagation probability of "contact wear" leading to "device failure" is very high, and the cumulative impact is also very high, then the emergency warning information is output to prompt the immediate replacement of the device. Assuming that the propagation probability from "contact wear" to "device failure" is 0.8, and the direct impact of "device failure" on "line tripping" is 0.9, the warning information may be: Fault source equipment: Line 1 protection device; Propagation path: Contact wear->Device failure->Line tripping; Risk level: Emergency; Disposal measures: Immediately replace Line 1 protection device.

[0126] In this embodiment, by combining the dynamic causal graph and the fuzzy cognitive graph, the fault propagation path and impact degree can be predicted more accurately, thereby improving the accuracy of the early warning and avoiding false alarms and missed alarms. Through fuzzy hierarchical analysis and early warning decision matrix, the fault risks can be graded, and corresponding disposal measures can be taken according to different levels to improve the effectiveness and practicality of the early warning. Through effective early warning of relay protection device failures, potential faults can be discovered and handled in a timely manner to avoid the expansion and spread of faults, thereby improving the reliability and safety of power system operation.

[0127] In an optional implementation, based on the dynamic causal graph, the causal support of the node is calculated by forward propagation, the diagnostic support of the node is calculated by backward propagation, the causal support and the diagnostic support are combined by a normalization factor to update the node belief value, the propagation path of the fault in the network is calculated according to the node belief value and the conditional probability, and the fault propagation probability is calculated based on the transmission probability between nodes and the intermediate node state, including:

[0128] Obtain the state information of nodes in the dynamic causal graph and construct the conditional probability matrix between nodes;

[0129] Identify the parent node set of the current node in the dynamic causal graph, calculate the information transmission delay time from each node in the parent node set to the current node, determine the timing weight based on the information transmission delay time, combine the timing weight with the physical coupling strength to generate a transmission coefficient, and obtain the causal support of the current node through forward propagation calculation based on the transmission coefficient and the initial belief value of each node in the parent node set;

[0130] Extracting a subnode set of the current node from the dynamic causal graph, calculating the state deviation of each node in the subnode set, constructing a feedback influence function based on the state deviation, multiplying the feedback influence function by the distance attenuation coefficient between nodes to obtain a diagnosis weight, and obtaining the diagnosis support of the current node through back propagation calculation based on the diagnosis weight and the belief value of each node in the subnode set;

[0131] Calculating an adaptive normalization factor based on the system operation status, weighting and combining the causal support and the diagnostic support through the adaptive normalization factor, and updating the belief value of the current node;

[0132] Based on the updated current node belief value and the conditional probability matrix, the fault transmission probability between adjacent nodes is calculated, a directed propagation network is constructed based on the fault transmission probability, and a fault propagation path is extracted from the directed propagation network;

[0133] The real-time state value of each node on the critical fault propagation path is obtained, the real-time state value is converted into a state function, and the cumulative propagation probability on the fault propagation path is calculated according to the state function and the fault transmission probability to obtain the final fault propagation probability.

[0134] Exemplarily, first, the state information of each node in the dynamic causal graph is obtained, such as the operating state and performance indicators of the node, and the conditional probability matrix between the nodes is constructed based on this state information. For example, a system containing nodes A, B, and C, A is the parent node of B, and B is the parent node of C. If A fails, the probability of B failing is 0.8, the probability of C failing under normal circumstances of B is 0.1, and the probability of C failing under the condition of B failure is 0.9. These probability values ​​constitute the conditional probability matrix.

[0135] Next, identify the parent node set of the current node in the dynamic causal graph. For each current node, find all parent nodes that directly affect it. For example, the parent node of node B is A. Then, calculate the information transmission delay time from each node in the parent node set to the current node. For example, the information transmission delay time from A to B is 2 time units. Determine the timing weight based on the information transmission delay time. The shorter the delay time, the greater the weight. Assuming that the weight is inversely proportional to the delay time, the timing weight from A to B is 1 / 2. Combine the timing weight with the physical coupling strength to generate the transfer coefficient. The physical coupling strength indicates the tightness of the connection between nodes. For example, the connection strength between A and B is 0.9. Assuming that the transfer coefficient is equal to the timing weight multiplied by the physical coupling strength, the transfer coefficient from A to B is (1 / 2)*0.9=0.45. Finally, according to the transfer coefficient and the initial belief value of each node in the parent node set (for example, the initial belief value of A is 0.2, indicating that the probability of A failing is 0.2), the causal support of the current node is calculated by forward propagation. For example, the causal support of B is 0.45*0.2=0.09.

[0136] Then, extract the child node set of the current node from the dynamic causal graph. For example, the child node of node B is C. Calculate the state deviation of each node in the child node set. The state deviation represents the difference between the state of the child node and its expected state. For example, the expected state value of C is 10, and the actual state value is 8, then the state deviation is 2. Construct a feedback influence function based on the state deviation. The larger the deviation, the greater the influence. For example, assuming that the feedback influence function is equal to the square root of the state deviation, the feedback influence function of C is √2≈1.41. Multiply the feedback influence function by the distance attenuation coefficient between nodes to obtain the diagnostic weight. The distance attenuation coefficient represents the weakening effect of the distance between nodes on the influence. The farther the distance, the smaller the influence. For example, assuming that the distance attenuation coefficient between B and C is 0.8, the diagnostic weight of C to B is 1.41*0.8=1.13. Finally, according to the diagnostic weight and the belief value of each node in the child node set (for example, the belief value of C is 0.3), the diagnostic support of the current node is calculated by backpropagation. For example, the diagnostic support of B is 1.13*0.3=0.34.

[0137] Next, the adaptive normalization factor is calculated based on the system operation status. For example, the higher the system load, the smaller the normalization factor. The causal support and the diagnostic support are weighted and combined by the adaptive normalization factor to update the belief value of the current node. For example, assuming the normalization factor is 0.5, the belief value of B is updated to 0.5*0.09+0.5*0.34=0.22.

[0138] Based on the updated current node belief value and conditional probability matrix, the fault transmission probability between adjacent nodes is calculated. For example, if the belief value of B is 0.22, the fault transmission probability from B to C is 0.22*0.9+(1-0.22)*0.1=0.28. A directed propagation network is constructed based on the fault transmission probability, and the fault propagation path is extracted from the directed propagation network. For example, the propagation path from A to C is A->B->C.

[0139] Finally, obtain the real-time state value of each node on the critical path of fault propagation and convert the real-time state value into a state function. For example, the real-time state value of node B is 8, which is converted into a state function f(B) = 8. Calculate the cumulative propagation probability on the fault propagation path based on the state function and the fault transfer probability to obtain the final fault propagation probability. For example, the fault propagation probability from A to C is the transfer probability from A to B multiplied by the transfer probability from B to C. Assuming that the transfer probability from A to B is 0.45*0.2=0.09, the fault propagation probability from A to C is 0.09*0.28=0.025.

[0140] In this embodiment, causal relationships, diagnostic information, timing characteristics, and system operating status are comprehensively considered, and the fault propagation path and probability can be predicted more accurately, thereby improving the accuracy of fault prediction. By accurately predicting the fault propagation path, preventive measures can be taken in advance to block the propagation of faults and avoid the occurrence of larger-scale faults, thereby improving the reliability and stability of the system. This method can help identify the key paths and nodes of fault propagation, thereby optimizing resource allocation in a targeted manner, such as strengthening the monitoring and maintenance of key nodes and improving resource utilization efficiency.

[0141] In an optional implementation, a hierarchical collaborative control strategy is constructed according to the fault warning information, the protection setting optimization problem is converted into a multi-objective optimization model at the local control layer, and a hybrid optimization algorithm is used to solve the multi-objective optimization model to obtain a candidate setting scheme; a distributed game optimization network is established at the global coordination layer based on the Nash equilibrium criterion, and the candidate setting schemes are coordinated and optimized using a potential game method, including:

[0142] Receiving system fault warning information, and extracting the fault type, fault location, fault current, fault duration and fault phase angle in the fault warning information as feature quantities;

[0143] Calculate the protection reliability index according to the fault type and fault current, calculate the protection sensitivity index according to the fault location and fault current, and calculate the protection selectivity index according to the fault duration and protection coordination time interval;

[0144] Constructing a multi-objective optimization model at the local control layer based on the protection reliability index, the protection sensitivity index and the protection selectivity index;

[0145] An adaptive crossover operator is set based on population density to solve the multi-objective optimization model to obtain a crossover solution, an adaptive mutation operator is set based on fitness value to mutate the crossover solution to obtain a mutated solution, a reference point set is constructed based on the mutated solution, the Euclidean distance from the mutated solution to the reference point set is calculated, and the solution with the smallest Euclidean distance is selected as a candidate fixed value solution;

[0146] At the global coordination layer, the protection unit is set as a game participant, the candidate setting scheme is set as the strategy space of the game participant, and a distributed game optimization network is constructed;

[0147] In the distributed game optimization network, the reliability benefits of the game participants when selecting strategies are calculated based on the protection sensitivity index, the coordination benefits of the game participants and adjacent protection units are calculated based on the protection coordination time interval, and the interference losses suffered by the game participants are calculated based on the electrical distance between adjacent protection units;

[0148] Taking the reliability benefit, coordination benefit and interference loss of the game participants as comprehensive benefits, constructing the utility function of the game participants;

[0149] Calculating the impedance values ​​between the protection units in the distributed game optimization network to obtain the degree of electrical connection, and determining the coupling coefficient between the game participants according to the degree of electrical connection; calculating the action time difference between the protection units to obtain the coordination relationship, and determining the interactive utility between the game participants according to the coordination relationship; constructing a potential function with the utility function, coupling coefficient and interactive utility of the game participants;

[0150] An initial strategy combination of game participants is generated in the distributed game optimization network, the utility function values ​​and potential function values ​​of the game participants under the initial strategy combination are calculated, the initial strategy combination is updated based on the utility function value to obtain an optimized response strategy, a new potential function value under the optimized response strategy is calculated, and when the difference between the new potential function value and the potential function value of the previous iteration is less than a preset threshold, the optimized response strategy is output as a coordinated optimization constant value scheme that meets the Nash equilibrium criterion.

[0151] For example, first, receive fault warning information from the system monitoring device. This information includes fault type (short circuit, ground fault), fault location (specific line or transformer number), fault current size (current value in amperes), fault duration (time length in milliseconds) and fault phase angle (angle value in degrees). This information will serve as the basis for subsequent analysis and calculation.

[0152] Next, these characteristic quantities are extracted and processed. The protection reliability index is calculated according to the fault type and the magnitude of the fault current. For example, the corresponding reliability index value is obtained by looking up a pre-established correspondence table between the fault type and the reliability index. The protection sensitivity index is calculated according to the fault location and the magnitude of the fault current. For example, the sensitivity is quantified by calculating the ratio of the fault current to the protection starting current. The protection selectivity index is calculated according to the fault duration and the protection coordination time interval (the difference between the action times of two adjacent protection devices). For example, the selectivity is evaluated by comparing the fault duration and the protection coordination time interval.

[0153] At the local control level, a multi-objective optimization model is constructed, with the goal of optimizing protection reliability, sensitivity and selectivity at the same time. These three indicators are used as objective functions, and corresponding constraints are set. The action time of the protection device must be within a certain range.

[0154] In order to solve this multi-objective optimization model, a hybrid optimization algorithm is adopted. First, an adaptive crossover operator is used to generate new candidate solutions. Specifically, the crossover probability is dynamically adjusted according to the population density of the current solution to improve the search efficiency. Then, the crossover solution is mutated using an adaptive mutation operator to generate more diverse solutions. The mutation probability is dynamically adjusted according to the fitness value of the current solution to balance exploration and utilization. Finally, a reference point set is constructed based on the mutated solution, the Euclidean distance from each mutated solution to the reference point set is calculated, and the solution with the smallest distance is selected as the candidate fixed value scheme. The reference point is set to an ideal combination of reliability, sensitivity, and selectivity.

[0155] In the global coordination layer, each protection unit is regarded as a game participant, and the candidate fixed value schemes obtained in the local control layer are used as the strategy space of the game participants to build a distributed game optimization network.

[0156] In this network, each game participant calculates its benefits according to the strategy it chooses. The reliability benefit is calculated based on the protection sensitivity index, and the higher the value, the higher the reliability; the coordination benefit with adjacent protection units is calculated based on the protection coordination time interval, and the shorter the time interval, the better the coordination; the interference loss is calculated based on the electrical distance (line length) between adjacent protection units, and the closer the distance, the greater the interference. Taking these three factors into consideration, the utility function of each game participant is constructed, for example, adding the reliability benefit and the coordination benefit, and then subtracting the interference loss.

[0157] In order to simulate the mutual influence between protection units, the impedance value between protection units is calculated to determine the coupling coefficient between game participants. The smaller the impedance value, the higher the degree of coupling. At the same time, the action time difference between protection units is calculated to determine the interaction utility between game participants. The smaller the action time difference, the stronger the interaction. The utility function, coupling coefficient and interaction utility are combined to construct the potential function.

[0158] In the distributed game optimization network, the initial strategy combination of the game participants is first randomly generated. Then, the utility function value and potential function value of each game participant under the current strategy combination are calculated. Next, each game participant updates its strategy according to its utility function value and selects the strategy that can maximize its utility. Finally, the new potential function value is calculated. The above process is repeated until the difference between the new potential function value and the potential function value of the previous iteration is less than the preset threshold. At this time, the strategy combination obtained is the coordinated optimization value scheme that meets the Nash equilibrium criterion.

[0159] In this embodiment, through the multi-objective optimization model and Nash equilibrium criterion, the protection reliability, sensitivity and selectivity indicators are comprehensively considered, which can effectively avoid the misoperation or refusal of the protection device, improve the reliability of the protection system, and ensure the safe and stable operation of the power system. The use of adaptive crossover and mutation operators can dynamically adjust the optimization parameters according to the system operation status, improve the search efficiency and adaptability of the algorithm, and better cope with various complex fault scenarios. Based on the fault warning information, the coordinated optimization setting scheme is automatically generated, which reduces manual intervention and debugging work, simplifies the protection setting process, and improves work efficiency.

[0160] Figure 2 FIG. 1 is a schematic diagram of a smart grid relay protection reliability evaluation and improvement control system according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0161] The first unit is used to collect the operation data of the relay protection device, and process the data through an adaptive data preprocessing model to divide it into a steady-state segment, a transient segment and a fault segment. The data of the steady-state segment, the transient segment and the fault segment are denoised by using an improved empirical mode decomposition method, and multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix. The enhanced feature matrix is ​​input into a deep hybrid neural network to extract spatial features, and time series associations are captured through cyclic branches. The spatial features and time series associations are combined by an adaptive fusion layer to obtain a comprehensive feature vector;

[0162] The second unit is used to input the comprehensive feature vector into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomalies of adjacent protection devices are analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The failure mode, performance degradation trend, coordinated anomaly and reliability level are input into a causal reasoning network to establish a dynamic causal graph, and the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with a fuzzy cognitive graph, and fault warning information is output;

[0163] The third unit is used to construct a hierarchical collaborative control strategy according to the fault warning information, transform the protection setting optimization problem into a multi-objective optimization model at the local control layer, and use a hybrid optimization algorithm to solve the multi-objective optimization model to obtain a candidate setting scheme; establish a distributed game optimization network based on the Nash equilibrium criterion at the global coordination layer, and use the potential game method to coordinate and optimize the candidate setting schemes; input the coordinated and optimized setting schemes into a hierarchical rolling optimization controller, and the hierarchical rolling optimization controller establishes a time domain optimization problem based on model predictive control, realizes closed-loop control through online state estimation and feedback correction, and completes the reliability improvement of the relay protection device.

[0164] According to a third aspect of the embodiments of the present invention,

[0165] An electronic device is provided, comprising:

[0166] processor;

[0167] a memory for storing processor-executable instructions;

[0168] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0169] A fourth aspect of the embodiments of the present invention is:

[0170] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0171] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart grid relay protection reliability assessment and improvement control method, characterized in that: include: The operation data of the relay protection device is collected, and the data is processed by an adaptive data preprocessing model to divide the data into a steady-state section, a transient section, and a fault section. The data of the steady-state section, the transient section, and the fault section are denoised by using an improved empirical mode decomposition method, and multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix. The enhanced feature matrix is ​​input into a deep hybrid neural network to extract spatial features, and the temporal association is captured by a cyclic branch. The spatial features and the temporal association are combined by an adaptive fusion layer to obtain a comprehensive feature vector; The comprehensive feature vector is input into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomalies of adjacent protection devices are analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The failure mode, performance degradation trend, coordinated anomaly and reliability level are input into a causal reasoning network to establish a dynamic causal graph, and the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with a fuzzy cognitive graph, and fault warning information is output; A hierarchical collaborative control strategy is constructed according to the fault warning information, and the protection setting optimization problem is converted into a multi-objective optimization model at the local control layer. The multi-objective optimization model is solved by a hybrid optimization algorithm to obtain candidate setting schemes; a distributed game optimization network is established at the global coordination layer based on the Nash equilibrium criterion, and the candidate setting schemes are coordinated and optimized by the potential game method; the setting schemes after coordinated optimization are input into a hierarchical rolling optimization controller, and the hierarchical rolling optimization controller establishes a time domain optimization problem based on model predictive control, realizes closed-loop control through online state estimation and feedback correction, and improves the reliability of the relay protection device.

2. The method according to claim 1, characterized in that The operation data of the relay protection device is collected and processed by an adaptive data preprocessing model to divide it into steady-state segment, transient segment and fault segment. The data of the steady-state segment, transient segment and fault segment are denoised by using an improved empirical mode decomposition method. The multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix including: Collecting operation data of the relay protection device, the operation data including a primary device current signal, a primary device voltage signal, a protection device action signal, a communication link status signal, a circuit breaker mechanical characteristic signal, and a secondary circuit insulation impedance signal, and normalizing the operation data to obtain a standardized data sequence; A variational modal decomposition model is established for the standardized data sequence. The variational modal decomposition model constructs a variational optimization objective function including a center frequency term and a bandwidth constraint term, and uses an alternating direction multiplier method to iteratively optimize the objective function to obtain an intrinsic mode function group. The mutual correlation coefficient of adjacent intrinsic mode functions is calculated based on the energy distribution characteristics of the intrinsic mode function group. An adaptive segmentation threshold is set according to the mutual correlation coefficient to divide the standardized data sequence into a steady-state segment, a transient segment, and a fault segment. The data of the steady-state section, transient section and fault section are subjected to improved empirical mode decomposition denoising processing respectively, and the intrinsic mode components are processed by an adaptive soft threshold function based on data distribution, the parameters of the adaptive soft threshold function are adjusted according to the number of layers and energy density of the intrinsic mode components, and the local mean envelope is used to replace the spline interpolation to eliminate the endpoint effect, so as to obtain the denoised data; Performing a wavelet packet transform on the denoised data, calculating the entropy of the wavelet packet decomposition coefficients, and selecting the node coefficient with the largest entropy value as the frequency domain feature; performing a Hilbert-Huang transform on the denoised data, calculating the instantaneous amplitude and instantaneous phase of the obtained analytical signal, and obtaining the instantaneous frequency according to the instantaneous phase to obtain the time-frequency feature; The frequency domain feature vector and the time-frequency feature vector are adaptively weighted fused, and high-order statistical features based on kernel density estimation are introduced, including mean, variance, skewness, kurtosis, quantile features and entropy features. The principal component analysis method is used to reduce the dimension to form an enhanced feature matrix.

3. The method according to claim 2, characterized in that A variational modal decomposition model is established for the standardized data sequence. The variational modal decomposition model constructs a variational optimization objective function including a center frequency term and a bandwidth constraint term, and uses an alternating direction multiplier method to iteratively optimize the objective function to obtain an intrinsic mode function group. The mutual correlation coefficient of adjacent intrinsic mode functions is calculated based on the energy distribution characteristics of the intrinsic mode function group. An adaptive segmentation threshold is set according to the mutual correlation coefficient, and the standardized data sequence is divided into a steady-state segment, a transient segment, and a fault segment, including: The standardized data sequence is discretely sampled and framed, each frame of data is converted into an analytical signal through Hilbert transform, the analytical signal is frequency modulated to obtain a center frequency sequence, and a reconstruction error matrix and a spectrum constraint matrix are constructed based on the center frequency sequence; The reconstruction error matrix and the spectrum constraint matrix are input into a Wiener filter, the frequency response of the Wiener filter is adaptively adjusted according to the bandwidth constraint parameters, and the modal components are output; soft threshold shrinkage processing is applied to the modal components, and the soft threshold parameters are determined by cross-validation; the power spectrum centroid is calculated based on the processed modal components to obtain an updated center frequency sequence; the center frequency sequence is optimized by gradient iteration to obtain an intrinsic mode function group; Calculating the instantaneous energy spectrum of each intrinsic mode function in the intrinsic mode function group to obtain an energy density distribution curve, and calculating the energy concentration and energy dispersion based on the energy density distribution curve to obtain an energy distribution characteristic matrix; Extracting energy envelopes of adjacent intrinsic mode functions, calculating normalized cross-correlation functions of the energy envelopes, calculating peak values ​​of the normalized cross-correlation functions within a sliding time window, and generating a cross-correlation coefficient sequence; Calculate the mean and standard deviation of the mutual correlation coefficient sequence, and set the mean as the reference threshold; multiply the energy concentration and energy dispersion in the energy distribution feature matrix to obtain a threshold adjustment factor; set the product of the reference threshold and the threshold adjustment factor as the adaptive segmentation threshold; In the sliding time window, the mutual correlation coefficient sequence is compared with the corresponding adaptive segmentation threshold, and when the mutual correlation coefficients of three consecutive windows are all less than the adaptive segmentation threshold, they are marked as candidate mutation points; the local energy density of the candidate mutation point is calculated, and the final modal mutation point position is determined based on the minimum energy density; the standardized data sequence is divided into a steady-state segment, a transient segment and a fault segment according to the modal mutation point.

4. The method according to claim 1, characterized in that The comprehensive feature vector is input into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomaly of adjacent protection devices is analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level, including: Constructing a multi-stage fault diagnosis model, wherein the multi-stage fault diagnosis model includes a device-level diagnosis sub-model, a bay-level diagnosis sub-model, and a station-level diagnosis sub-model; The device-level diagnosis sub-model adopts a deep belief network, inputs the comprehensive feature vector into the deep belief network, performs layer-by-layer pre-training through a restricted Boltzmann machine, optimizes the network weights using a contrastive divergence algorithm and a stochastic gradient descent method, introduces label data for network fine-tuning, and obtains a failure mode probability vector and a performance degradation index for a single protection device; The interval-level diagnosis sub-model adopts a conditional random field structure, takes the fault mode probability vector and performance degradation index as node features, constructs a spatial potential function based on the primary equipment topological relationship, and constructs a time series potential function based on the operating state correlation. The fault mode probability vector, performance degradation index, spatial potential function and time series potential function are input into the conditional random field structure, and the model parameters are optimized by maximum likelihood estimation, and the coordinated abnormality probability matrix of adjacent devices is output; The station-level diagnosis submodel adopts a hierarchical clustering structure, and fuses the fault mode probability vector, performance degradation index and collaborative abnormality probability matrix to construct a state characterization matrix. The state distance between devices is calculated based on the state characterization matrix, and the hierarchical clustering algorithm with adaptive distance measurement is used to group the devices. The optimal number of groups is determined by the silhouette coefficient to obtain the reliability level of the protection device.

5. The method according to claim 1, characterized in that The failure mode, performance degradation trend, collaborative anomaly and reliability level are input into the causal reasoning network to establish a dynamic causal graph. The fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with the fuzzy cognitive graph. The fault warning information output includes: The failure modes, performance degradation trends, coordinated anomalies and reliability levels of relay protection devices are used to construct a node set of a causal reasoning network. The conditional probability table between nodes is calculated based on the conditional independence assumption. The node status is updated online using the recursive Bayesian estimation method. The conditional probability parameters are optimized based on historical observation data, and a dynamic causal graph that reflects the dynamic correlation between nodes is established. Based on the dynamic causal graph, the causal support of the node is calculated by forward propagation, the diagnostic support of the node is calculated by backward propagation, the causal support and the diagnostic support are combined by a normalization factor to update the node belief value, the propagation path of the fault in the network is calculated according to the node belief value and conditional probability, and the fault propagation probability is calculated based on the transmission probability between nodes and the intermediate node status; The fault propagation path and fault propagation probability are mapped into concept nodes and weight matrices of fuzzy cognitive graphs respectively, a fuzzy relationship matrix including fault source, fault type, impact range, and propagation path is constructed, and the target node state vector is obtained through fuzzy synthesis operation and hyperbolic tangent activation function iterative update; The direct influence is calculated based on the weight matrix of the fuzzy cognitive graph and the target node state vector, the indirect influence is calculated by combining the fault propagation probability and the preset attenuation factor, and the direct influence and the indirect influence are combined by an exponential weighting method to obtain the cumulative influence of the fault; An early warning decision matrix is ​​constructed based on the probability of fault propagation and the cumulative impact of faults. The fuzzy hierarchical analysis is used to determine the warning level threshold. The warning information is divided into four levels. The warning information including fault source equipment, propagation path, risk level and disposal measures is output in a hierarchical and progressive manner according to the warning level.

6. The method according to claim 5, characterized in that Based on the dynamic causal graph, the causal support of the node is calculated by forward propagation, the diagnostic support of the node is calculated by backward propagation, the causal support and the diagnostic support are combined by a normalization factor to update the node belief value, the propagation path of the fault in the network is calculated according to the node belief value and the conditional probability, and the fault propagation probability is calculated based on the transmission probability between nodes and the intermediate node state, including: Obtain the state information of nodes in the dynamic causal graph and construct the conditional probability matrix between nodes; Identify the parent node set of the current node in the dynamic causal graph, calculate the information transmission delay time from each node in the parent node set to the current node, determine the timing weight based on the information transmission delay time, combine the timing weight with the physical coupling strength to generate a transmission coefficient, and obtain the causal support of the current node through forward propagation calculation based on the transmission coefficient and the initial belief value of each node in the parent node set; Extracting a subnode set of the current node from the dynamic causal graph, calculating the state deviation of each node in the subnode set, constructing a feedback influence function based on the state deviation, multiplying the feedback influence function by the distance attenuation coefficient between nodes to obtain a diagnosis weight, and obtaining the diagnosis support of the current node through back propagation calculation based on the diagnosis weight and the belief value of each node in the subnode set; Calculating an adaptive normalization factor based on the system operation status, weighting and combining the causal support and the diagnostic support through the adaptive normalization factor, and updating the belief value of the current node; Based on the updated current node belief value and the conditional probability matrix, the fault transmission probability between adjacent nodes is calculated, a directed propagation network is constructed based on the fault transmission probability, and a fault propagation path is extracted from the directed propagation network; The real-time state value of each node on the critical fault propagation path is obtained, the real-time state value is converted into a state function, and the cumulative propagation probability on the fault propagation path is calculated according to the state function and the fault transmission probability to obtain the final fault propagation probability.

7. The method according to claim 1, characterized in that Constructing a hierarchical collaborative control strategy based on the fault warning information, converting the protection setting optimization problem into a multi-objective optimization model at the local control layer, and using a hybrid optimization algorithm to solve the multi-objective optimization model to obtain a candidate setting solution; In the global coordination layer, a distributed game optimization network is established based on the Nash equilibrium criterion, and the potential game method is used to coordinate and optimize the candidate fixed value schemes, including: Receiving system fault warning information, and extracting the fault type, fault location, fault current, fault duration and fault phase angle in the fault warning information as feature quantities; Calculate the protection reliability index according to the fault type and fault current, calculate the protection sensitivity index according to the fault location and fault current, and calculate the protection selectivity index according to the fault duration and protection coordination time interval; Constructing a multi-objective optimization model at the local control layer based on the protection reliability index, the protection sensitivity index and the protection selectivity index; An adaptive crossover operator is set based on population density to solve the multi-objective optimization model to obtain a crossover solution, an adaptive mutation operator is set based on fitness value to mutate the crossover solution to obtain a mutated solution, a reference point set is constructed based on the mutated solution, the Euclidean distance from the mutated solution to the reference point set is calculated, and the solution with the smallest Euclidean distance is selected as a candidate fixed value solution; At the global coordination layer, the protection unit is set as a game participant, the candidate setting scheme is set as the strategy space of the game participant, and a distributed game optimization network is constructed; In the distributed game optimization network, the reliability benefits of the game participants when selecting strategies are calculated based on the protection sensitivity index, the coordination benefits of the game participants and adjacent protection units are calculated based on the protection coordination time interval, and the interference losses suffered by the game participants are calculated based on the electrical distance between adjacent protection units; Taking the reliability benefit, coordination benefit and interference loss of the game participants as comprehensive benefits, the utility function of the game participants is constructed; Calculating the impedance values ​​between the protection units in the distributed game optimization network to obtain the degree of electrical connection, and determining the coupling coefficient between the game participants according to the degree of electrical connection; calculating the action time difference between the protection units to obtain the coordination relationship, and determining the interactive utility between the game participants according to the coordination relationship; constructing a potential function with the utility function, coupling coefficient and interactive utility of the game participants; An initial strategy combination of game participants is generated in the distributed game optimization network, the utility function values ​​and potential function values ​​of the game participants under the initial strategy combination are calculated, the initial strategy combination is updated based on the utility function value to obtain an optimized response strategy, a new potential function value under the optimized response strategy is calculated, and when the difference between the new potential function value and the potential function value of the previous iteration is less than a preset threshold, the optimized response strategy is output as a coordinated optimization constant value scheme that meets the Nash equilibrium criterion.

8. A smart grid relay protection reliability assessment and improvement control system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to collect the operation data of the relay protection device, and process the data through an adaptive data preprocessing model to divide it into a steady-state segment, a transient segment and a fault segment. The data of the steady-state segment, the transient segment and the fault segment are denoised by using an improved empirical mode decomposition method, and multi-domain features are extracted by combining wavelet packet transform and Hilbert-Huang transform to generate an enhanced feature matrix. The enhanced feature matrix is ​​input into a deep hybrid neural network to extract spatial features, and time series associations are captured through cyclic branches. The spatial features and time series associations are combined by an adaptive fusion layer to obtain a comprehensive feature vector; The second unit is used to input the comprehensive feature vector into a pre-built multi-stage fault diagnosis model, wherein the failure mode and performance degradation trend of a single protection device are identified based on a deep belief network at the device level, the coordinated anomalies of adjacent protection devices are analyzed based on an improved conditional random field at the interval level, and the reliability level of the protection device group is divided based on a hierarchical clustering method at the station level. The failure mode, performance degradation trend, coordinated anomaly and reliability level are input into a causal reasoning network to establish a dynamic causal graph, and the fault propagation path and propagation probability are predicted based on the dynamic causal graph, and the fault impact is evaluated in combination with a fuzzy cognitive graph, and fault warning information is output; The third unit is used to construct a hierarchical collaborative control strategy according to the fault warning information, transform the protection setting optimization problem into a multi-objective optimization model at the local control layer, and use a hybrid optimization algorithm to solve the multi-objective optimization model to obtain a candidate setting scheme; establish a distributed game optimization network based on the Nash equilibrium criterion at the global coordination layer, and use the potential game method to coordinate and optimize the candidate setting schemes; input the coordinated and optimized setting schemes into a hierarchical rolling optimization controller, and the hierarchical rolling optimization controller establishes a time domain optimization problem based on model predictive control, realizes closed-loop control through online state estimation and feedback correction, and completes the reliability improvement of the relay protection device.

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

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

Images