An adaptive backup power automatic switching monitoring method, device and medium for a multi-source power grid

Abstract

This invention discloses an adaptive backup and automatic transfer (SALT) monitoring method, device, and medium for multi-source power grids, relating to the field of smart grid industrial data processing technology. By collecting industrial data from smart grid industrial big data, the invention determines the probability of topology changes based on the collected data and triggers real-time adaptive updates to the SALT model. This enables the control strategy of the SALT device to dynamically adapt to changes in power grid operation, improving power supply reliability. By introducing system health as a safety gate for SALT action, it automatically blocks SALT when health is low. Combined with accurate fault type identification, it effectively avoids accidental connection to the faulty bus during bus faults, preventing cascading failures. By prioritizing the selection of the connection path based on path capacity margin, it ensures the safety of load transfer, forming a fully intelligent closed-loop process for adaptive SALT decision-making, improving the timeliness, accuracy, and safety of power grid emergency decision-making.

Description

Technical Field

[0001] This invention relates to the field of smart grid industrial data processing technology, specifically to an adaptive backup and automatic transfer monitoring method, device, and medium for multi-source power grids. Background Technology

[0002] With the rapid development of smart grids, the scale of power grids is expanding and their structure is becoming increasingly complex. Furthermore, the large-scale integration of distributed energy resources and flexible loads makes the grid topology dynamically changing. Ensuring the safe and stable operation of the power grid, and enabling rapid and accurate diagnosis and recovery in the event of faults, is a key technological challenge in the field of power system automation. Existing technologies typically employ a layered and phased approach, which has the following main limitations:

[0003] Current mainstream fault diagnosis systems rely on threshold judgments or simple logical associations of the action signals and alarm information of devices such as relay protection devices and circuit breakers. For more sensitive electrical quantities, such as three-phase current and voltage waveforms, features are typically extracted using methods such as Fourier transform and empirical mode decomposition, combined with traditional machine learning algorithms such as support vector machines and decision trees for classification. However, these methods have significant shortcomings: First, they are weak in effectively fusing and synchronizing multi-source, asynchronous switching and analog signals, and data quality directly affects the diagnostic results; second, traditional signal processing methods are insufficient in extracting non-stationary and nonlinear features from transient current and voltage signals, making it difficult to effectively identify early weak faults or complex faults (such as high-resistance grounding and intermittent arcing faults), thus limiting the accuracy and timeliness of diagnosis.

[0004] The fault diagnosis module only outputs the location and type of the faulty equipment or line, while the recovery decision module generates a control scheme based on a relatively static power grid topology model and a pre-set expert rule base (such as automatic backup power supply logic). This architecture has prominent problems: First, the decision-making lacks a global perspective. The decision-making process does not fully consider the overall health status of the network after the fault, the load capacity (capacity margin) of the remaining lines, and the real-time connectivity of the network topology, which may lead to an infeasible recovery scheme or introduce new overload risks. Second, the system lacks adaptive capabilities. The power grid topology changes dynamically due to adjustments in operating modes, distributed power supply switching, maintenance operations, etc., but the network model on which the decision-making relies is often pre-set offline and difficult to update in real time, leading to a disconnect between the formulated recovery strategy and the actual network structure, and even causing malfunctions.

[0005] Traditional automatic transfer switches (ATS) for backup power sources (such as automatic transfer on backup lines and automatic transfer on bus sections) are mostly triggered based on local electrical quantities (such as voltage loss and current flow) and simple timing logic. Their strategies are rigid and cannot adapt to complex changes in operating conditions. Although recent research has attempted to introduce intelligent algorithms for optimization, these methods are usually still based on simplified or fixed network models, failing to deeply integrate real-time, global power grid topology and multi-dimensional operating states (such as real-time capacity margins of each path and overall system health) into the decision-making logic. Therefore, existing technologies struggle to make accurate judgments after a fault occurs. Summary of the Invention

[0006] The technical problem this invention aims to solve is that traditional automatic transfer switch (ATS) devices lack adaptive sensing capabilities for grid operation status and multi-dimensional comprehensive decision-making basis, leading to erroneous operation and difficulty in adaptation. The goal is to provide an adaptive ATS monitoring method, device, and medium for multi-source power grids. By triggering real-time adaptive updates of the ATS model through topology change probability, the control strategy of the ATS device can dynamically adapt to changes in grid operation mode, improving power supply reliability. By introducing system health as a safety gate for ATS action, it automatically blocks ATS when health is low. Combined with accurate fault type identification, it effectively avoids cascading faults caused by accidental connection to the faulty bus during bus faults. By prioritizing the selection of the connection path based on path capacity margin, it ensures the safety of load transfer, forming a fully intelligent closed-loop adaptive decision-making process for ATS, improving the timeliness, accuracy, and safety of grid emergency decision-making.

[0007] This invention is achieved through the following technical solution:

[0008] The first aspect of this invention provides an adaptive backup and automatic transfer monitoring method for a multi-source power grid, comprising the following specific steps:

[0009] Acquire smart grid operation data, which includes at least charging status signals, output action signals, and abnormal alarm signals;

[0010] The smart grid operation data is preprocessed to obtain preprocessed synchronization data frames;

[0011] Three-phase current and three-phase voltage sequences are extracted from the preprocessed synchronization data frames, and fault category labels are identified based on the three-phase current and three-phase voltage sequences.

[0012] A power grid graph structure is constructed using each power device in the power grid as a node and the transmission line as an edge, and the fault category label is used as the initial feature vector of the node.

[0013] The power grid graph structure is input into a graph encoder to extract the global graph embedding;

[0014] The global graph is embedded into multiple decoders for parallel decoding, and the system health, topology change probability, accurate fault type and optimal power supply path are output respectively.

[0015] The path capacity margin is obtained based on system health, topology change probability, precise fault type, and optimal power supply path;

[0016] Based on system health, topology change probability, and path capacity margin, adaptive backup and automatic transfer action judgment and optimal backup power supply connection path decision are made.

[0017] Furthermore, the preprocessing of smart grid operation data specifically includes:

[0018] The running data is subjected to time alignment processing of multi-source heterogeneous data to generate synchronized data with a unified time base;

[0019] The synchronized data is subjected to data purification processing, which includes at least outlier detection and repair, and missing value imputation.

[0020] The purified charging status signal, the output action signal, and the abnormal alarm signal are fused together according to a unified time tag to generate a preprocessed synchronous data frame.

[0021] Furthermore, the step of extracting three-phase current and three-phase voltage sequences from the preprocessed synchronization data frames, and identifying fault category labels based on the three-phase current and three-phase voltage sequences, specifically includes:

[0022] Extract the three-phase current sequence and the three-phase voltage sequence from the preprocessed synchronization data frame;

[0023] Wavelet packet decomposition is performed on the three-phase current sequence and the three-phase voltage sequence to extract the energy distribution in each frequency band, calculate the energy entropy of each frequency band, and construct an initial high-dimensional fault feature vector.

[0024] Kernel principal component analysis is used to reduce the dimensionality of the initial high-dimensional fault feature vector to obtain a low-dimensional sensitive feature set after dimensionality reduction.

[0025] The low-dimensional sensitive feature set is input into a pre-trained support vector machine classifier, which outputs fault category labels.

[0026] Furthermore, after constructing the power grid diagram structure, the process also includes updating the power grid diagram structure:

[0027] Acquire real-time operation data of the smart grid, wherein the real-time operation data includes at least charging status signals, output action signals, abnormal alarm signals and measurement section data;

[0028] Based on the exit action signal and the abnormal alarm signal, the connection status of the edges in the power grid diagram structure is updated to generate a dynamic topology model, which reflects the actual physical connection relationship of the power grid at the current moment.

[0029] Based on the measured cross-sectional data, update each node and each edge in the power grid diagram structure;

[0030] The dynamic topology graph model is associated and fused with the refreshed node feature values ​​and edge feature values ​​to generate a real-time power grid graph structure with time-stamped correction labels.

[0031] Furthermore, the step of inputting the power grid graph structure into the graph encoder to extract the global graph embedding includes:

[0032] Construct an adjacency matrix based on the physical connection relationships between nodes in the power grid diagram structure;

[0033] Construct a node feature matrix based on the initial feature vectors of the nodes in the power grid diagram structure;

[0034] The adjacency matrix and the node feature matrix are input into the graph encoder. The graph encoder uses a graph convolutional network to perform message passing and updating on the node feature matrix through multi-layer graph convolution operations to generate a node-level embedding matrix.

[0035] The node-level embedding matrix is ​​subjected to global pooling to obtain the global graph embedding.

[0036] Furthermore, the parallel decoding includes:

[0037] The global graph is embedded, copied, and input into the health regression decoder, topology change binary classification decoder, fault type multi-classification decoder, and path planning decoder respectively.

[0038] The health regression decoder performs multi-layer fully connected mapping on the global graph embedding and outputs the system health, which characterizes the overall operating status of the power grid.

[0039] The topology change binary classifier performs multi-layer fully connected mapping on the global graph embedding and outputs the probability of a change in the power grid topology.

[0040] The fault type multi-class decoder performs multi-layer fully connected mapping on the global graph embedding to output the precise fault type;

[0041] The path planning decoder, based on the global graph embedding and power grid topology connection relationship, uses a strategy network or heuristic search algorithm to output the optimal power supply path that satisfies line capacity constraints and voltage drop constraints.

[0042] Furthermore, the path capacity margin is obtained based on system health, topology change probability, precise fault type, and optimal power supply path, including:

[0043] Determine whether to trigger graph structure re-update based on the probability of topology change: if the probability of topology change is greater than a preset threshold, return to the step of real-time update of the power grid graph structure; otherwise, perform capacity margin calculation based on the current power grid graph structure.

[0044] A fault impact coefficient is determined based on the precise fault type, and the fault impact coefficient is used to characterize the degree of impact of different fault types on the allowable transmission capacity of the equipment.

[0045] A health correction coefficient is determined based on the system health status, and the health correction coefficient is positively correlated with the system health status.

[0046] Extract several nodes and edges contained in the optimal power supply path, and obtain the equipment parameters of the several nodes and edges from the real-time power grid graph structure. The equipment parameters include at least the rated capacity and the current real-time load.

[0047] Calculate the path capacity margin based on the fault impact factor, health correction factor, rated capacity, and current real-time load;

[0048] Iterate through all devices and lines on the optimal power supply path, and take the minimum path capacity margin as the path capacity margin of that path.

[0049] Furthermore, the adaptive backup power supply action judgment and optimal backup power supply connection path decision based on system health, topology change probability, and path capacity margin includes:

[0050] The automatic backup action is triggered based on the aforementioned topology change probability.

[0051] If the probability of topology change is greater than the first preset threshold, it is determined that the power grid topology has changed, the state of the automatic transfer switch model is changed, and the automatic transfer switch control model is analyzed and updated in real time through the power grid automatic topology search algorithm so that the automatic transfer switch model can adapt to the current power grid operation mode.

[0052] If the probability of topology change is not greater than the first preset threshold, then the current standby self-starting model state is maintained;

[0053] When changing the status of the backup automatic transfer model, the feasibility of the backup automatic transfer action is determined based on the system health status:

[0054] If the system health is lower than the second preset threshold, the power grid as a whole is determined to be in a risky operating state, a backup automatic transfer blocking signal is generated, and the backup power supply operation is prohibited.

[0055] If the system health is not lower than the second preset threshold, then it is determined that the backup automatic switching action is allowed;

[0056] When it is determined that the automatic backup power supply action is permitted, the backup power supply selection and connection path decision are made based on the aforementioned path capacity margin:

[0057] Extract the path capacity margin corresponding to each candidate backup power source from the optimal power supply path and its alternative paths;

[0058] The candidate path with the largest path capacity margin is selected as the optimal backup power supply path.

[0059] If the path capacity margin of the optimal backup power supply path is greater than the third preset threshold, then the backup power supply operation on that path is executed.

[0060] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement an adaptive backup and automatic transfer monitoring method for a multi-source power grid.

[0061] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an adaptive backup and automatic transfer monitoring method for a multi-source power grid.

[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0063] By preprocessing the key state signals and performing deep feature extraction on the three-phase current and three-phase voltage sequences, subtle patterns in transient and non-stationary fault signals can be effectively captured, thereby enabling earlier and more accurate classification and identification of complex fault types and reducing the failure rate and misjudgment rate of faults.

[0064] A graph structure is introduced to model the physical power grid, and a graph encoder is used to extract a global graph embedding containing node and connection relationships. This embedding simultaneously decodes four key state indicators: system health, topology change probability, fault type, and optimal power supply path. This achieves deep fusion and unified representation from local signals to global topology, and from current state to future change probability, providing a panoramic, multi-dimensional state profile for decision-making.

[0065] By updating the power grid diagram structure in real time, the system can dynamically track changes in the power grid operation mode and synchronously read the path capacity margin when making decisions, ensuring the timeliness of the decision-making basis.

[0066] By constructing a power grid graph structure and applying graph neural networks to achieve multi-dimensional state feature extraction and parallel decoding, the data preprocessing, fault diagnosis and recovery decision-making processes are effectively integrated. This enables the effective fusion and synchronization of multi-source heterogeneous data, improves the accuracy and real-time performance of fault diagnosis, supports dynamic updates of the power grid topology, and optimizes the adaptability and reliability of automatic backup power transfer decisions. Attached Figure Description

[0067] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0068] Figure 1 This is a flowchart of the monitoring method in an embodiment of the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0070] As one possible implementation method, such as Figure 1 As shown, this embodiment provides an adaptive backup and automatic transfer monitoring method for multi-source power grids. This embodiment targets intelligent large-scale smart grid power supply systems such as energy storage systems, distributed control of 750 kV and above AC transmission, large-scale power grid security and defense systems, and intelligent dispatching systems. First, scattered messages and telemetry records from substations, feeders, and distribution terminals are collected to obtain smart grid operation data, including charging status signals, output action signals, and abnormal alarm signals. This smart grid operation data is preprocessed to obtain preprocessed synchronization data frames. Three-phase current sequences and three-phase voltage sequences are extracted from the preprocessed synchronization data frames. Based on the three-phase current sequences... The system identifies fault category labels based on three-phase voltage sequences; constructs a power grid graph structure with each power device in the grid as a node and transmission lines as edges, and uses the fault category labels as the initial feature vectors of the nodes; inputs the power grid graph structure into a graph encoder to extract global graph embedding; inputs the global graph embedding into multiple decoders in parallel for parallel decoding, outputting system health, topology change probability, accurate fault type, and optimal power supply path respectively; obtains path capacity margin based on system health, topology change probability, accurate fault type, and optimal power supply path; and performs adaptive backup power supply action judgment and optimal backup power supply path decision based on system health, topology change probability, and path capacity margin.

[0071] The specific implementation process is as follows:

[0072] S1. The system monitors in real time and acquires smart grid operation data. Based on the operation data, it accurately extracts core signals reflecting the grid status, including key status information such as charging status signals, output action signals, and abnormal alarm signals.

[0073] In this embodiment, smart grid operation data refers to various real-time or near-real-time data collected from the smart grid, used to reflect the grid's operating status, equipment status, and event information. Charging status signals refer to signals reflecting the charging and discharging status, power, and capacity of loads such as energy storage devices in the grid. Output action signals refer to action commands or status change signals issued by relay protection devices or circuit breakers when they detect faults or anomalies, used to indicate the location or equipment where the fault occurred. Abnormal alarm signals refer to various abnormal states or potential risks alerted by the monitoring system or the equipment itself during grid operation, such as overload alarms and voltage limit exceedance alarms.

[0074] S2. Preprocess the smart grid operation data to obtain preprocessed synchronization data frames;

[0075] A unified time reference system provided by a high-precision network time protocol or a global clock synchronization server is introduced as the system-level master clock source. Subsequently, the original data streams (including charging voltage / current / temperature signals sampled continuously at the millisecond level, output opening and closing action change signals recorded in an event-triggered manner, and randomly generated abnormal alarm messages) are timestamped and standardized.

[0076] Outlier detection and repair: Instead of simply removing detected outlier data, the system combines the physical constraints of the effective neighboring values ​​before and after the data with those of other relevant parameters at the same time. It uses weighted moving average or local polynomial fitting based on least squares to perform smooth correction, ensuring the continuity and physical consistency of the data.

[0077] Missing value imputation: For data gaps caused by network packet loss or sampling pauses, differentiated strategies are adopted based on the missing value pattern. For short-term random missing values, linear interpolation or cubic Hermite interpolation is used for imputation; for long-term systematic missing values, dynamic time warping (DTW) similarity matching algorithm based on the pattern of historical data from the same period is introduced, or a multiple linear regression model is used to predict missing values ​​with other intact parameters at the same timestamp as input, thereby restoring the true operating state of the system to the greatest extent possible.

[0078] Multidimensional fusion and synchronous data frame generation of key signals after purification:

[0079] The purified charging status signals, output action signals, and abnormal alarm signals now possess unified time stamps and reliable data quality. Based on this, the final data fusion operation is performed: using the standard time stamp as the primary key, the above three types of signals are precisely matched according to their timestamps and combined into a structured data record using a feature concatenation method. Each record constitutes a synchronous data frame.

[0080] S3. Extract the three-phase current sequence and three-phase voltage sequence from the preprocessed synchronization data frame, and identify the fault category label based on the three-phase current sequence and three-phase voltage sequence;

[0081] Three-phase current and three-phase voltage sequences are extracted from the preprocessed synchronous data frames to form six-channel waveform data and a 6×N original signal matrix. The decomposition level is set to L, and wavelet packet decomposition is performed on the three-phase current and three-phase voltage sequences. Wavelet packet decomposition is also performed on each of the six channels, dividing the signal into multiple frequency band subspaces. Single-branch reconstruction is performed on each node (frequency band) of the Lth layer to obtain time-domain components. Energy distributions under each frequency band are extracted based on the time-domain components, and the energy entropy of each frequency band is calculated. The energy entropies calculated for each of the six channels are concatenated sequentially to construct an initial high-dimensional fault feature vector. Kernel principal component analysis is used to perform nonlinear dimensionality reduction on the initial high-dimensional fault feature vector. Principal components are selected according to the cumulative contribution rate to obtain a low-dimensional sensitive feature set. The low-dimensional sensitive feature set is input into a pre-trained support vector machine classifier, which outputs fault category labels.

[0082] The nonlinear dimensionality reduction of the initial high-dimensional fault feature vector using kernel principal component analysis includes:

[0083] A radial basis function (RBF) is selected to implicitly map the original multidimensional features to a high-dimensional reproducing kernel Hilbert space (RKHS).

[0084] The kernel matrix K is centered to eliminate mean bias. The eigenvalues ​​λ and eigenvectors of the centered kernel matrix are solved. The variance contribution rate of each principal component is calculated, and the components are sorted in descending order. The first d principal components are selected to obtain a low-dimensional sensitive feature set, and the dimension is reduced to d.

[0085] By extracting deep time-frequency domain features of the signal through wavelet packet energy entropy, combining kernel principal component analysis to eliminate nonlinear redundancy, and then classifying and outputting fault categories through support vector machine, accurate identification and classification of equipment electrical faults are achieved.

[0086] S4. Construct the power grid diagram structure and update it in real time;

[0087] Based on the physical topology of the power grid and the fault category labels, a power grid graph structure is constructed. The power grid graph structure includes nodes and edges. Nodes represent electrical equipment in the power grid, and edges represent electrical connection lines between equipment. Nodes and edges each have initial feature values.

[0088] After constructing the power grid graph structure, real-time operational data of the smart grid is acquired. This real-time operational data includes at least charging status signals, output action signals, abnormal alarm signals, and measurement section data. Based on the output action signals and abnormal alarm signals, the connection status of the edges in the power grid graph structure is updated to generate a dynamic topology graph model. The dynamic topology graph model reflects the actual physical connection relationship of the power grid at the current moment. Based on the measurement section data, the feature values ​​of each node and edge in the power grid graph structure are refreshed. When there is no measurement data for a certain edge due to a switch being disconnected, virtual zero power flow is used to fill the gap, keeping the dimension of the feature matrix constant. The dynamic topology graph model is correlated and fused with the refreshed node and edge feature values ​​to generate a real-time power grid graph structure with time-stamped correction labels. The time-stamped labels ensure that the data processed in all subsequent graph encoding and decoding processes comes from the same time section, solving the problem of steady-state / transient aliasing in the power system.

[0089] Through dynamic topology updates and measurement data refreshes, the power grid diagram structure can reflect switch changes, fault alarms, and power flow changes in real time, ensuring that the diagram structure upon which subsequent diagram coding and decision-making rely is always consistent with the actual operating status on site.

[0090] Step S5: Input the power grid graph structure into the graph encoder to extract the global graph embedding;

[0091] An adjacency matrix A is constructed based on the physical connections between nodes in the power grid graph structure. A node feature matrix X is constructed based on the initial feature vectors of the nodes in the power grid graph structure. The adjacency matrix A and the node feature matrix X are input into a graph encoder. The graph encoder employs a graph convolutional network, which aggregates the features of all its neighboring nodes for each node through multi-layer graph convolution operations (weighted summation) to achieve message passing and updating, generating a node-level embedding matrix. Global pooling is then applied to the node-level embedding matrix to obtain the global graph embedding. By using a graph convolutional network to perform message passing and aggregation on the power grid graph structure, the global dependencies and inter-node coupling information in the power grid topology are captured, generating a global graph embedding vector with overall awareness.

[0092] Step S6: Perform parallel decoding on the global graph embedding and output multi-dimensional diagnostic information;

[0093] The global graph embedding is simultaneously copied and input into the health regression decoder, topology change binary classification decoder, fault type multi-classification decoder, and path planning decoder. The health regression decoder performs multi-layer fully connected mapping on the global graph embedding and outputs the system health, which characterizes the overall operating status of the power grid. The topology change binary classification decoder performs multi-layer fully connected mapping on the global graph embedding and outputs the probability of changes in the power grid topology. The fault type multi-classification decoder performs multi-layer fully connected mapping on the global graph embedding and outputs the precise fault type. Based on the global graph embedding and the power grid topology connection relationship, the path planning decoder uses a policy network or heuristic search algorithm to output the optimal power supply path that satisfies the line capacity constraint and voltage drop constraint.

[0094] By embedding a single global graph into parallel multi-task decoding, four diagnostic results—health status, topology change probability, fault type, and optimal path—are output simultaneously, achieving information sharing and computational reuse, and improving the real-time performance and efficiency of multi-dimensional situational awareness.

[0095] Step S7: Calculate the path capacity margin based on system health, topology change probability, precise fault type, and optimal power supply path;

[0096] The system determines whether to trigger a graph structure re-update based on the topology change probability: if the topology change probability is greater than a preset threshold, it returns to the step of executing the real-time update of the power grid graph structure; otherwise, it calculates the capacity margin based on the current power grid graph structure. It determines the fault impact coefficient based on the precise fault type, which characterizes the degree of impact of different fault types on the allowable transmission capacity of the equipment. It determines the health correction coefficient based on the system health, which is positively correlated with the system health. It extracts several nodes and edges contained in the optimal power supply path and obtains the equipment parameters of several nodes and edges from the real-time power grid graph structure. The equipment parameters include at least the rated capacity and the current real-time load. It calculates the path capacity margin based on the fault impact coefficient, health correction coefficient, rated capacity, and current real-time load. It traverses all equipment and lines on the optimal power supply path and takes the minimum path capacity margin as the path capacity margin for that path.

[0097] By introducing a fault impact coefficient and a health correction coefficient to dynamically reduce the rated capacity of the equipment, and combining this with the current real-time load to calculate the path capacity margin, the actual constraints of fault type and system health level on power supply capacity are fully considered, ensuring the accuracy of margin assessment and engineering authenticity.

[0098] Step S8: Based on system health, topology change probability and path capacity margin, make adaptive judgments on automatic backup power transfer and determine the optimal backup power supply path.

[0099] The triggering of automatic transfer switching (ATS) is determined based on the probability of topology change: if the probability of topology change is greater than a first preset threshold, it is determined that the grid topology has changed, the ATS model state is changed, and the ATS control model is updated in real time through the automatic grid topology search algorithm to adapt the ATS model to the current grid operation mode; if the probability of topology change is not greater than the first preset threshold, the current ATS model state is maintained. When changing the ATS model state, the feasibility of the ATS action is determined based on the system health: if the system health is lower than a second preset threshold, it is determined that the entire grid is in a risky operating state, an ATS blocking signal is generated, and the execution of backup power supply connection is prohibited; if the system health is not lower than the second preset threshold, the ATS action is allowed.

[0100] When it is determined that the automatic backup power supply action is allowed, the backup power supply selection and connection path decision are made according to the path capacity margin: extract the path capacity margin corresponding to each candidate backup power supply from the optimal power supply path and its alternative paths; select the candidate path with the largest path capacity margin as the optimal backup power supply connection path; if the path capacity margin of the optimal backup power supply connection path is greater than the third preset threshold, then the backup power supply connection operation on that path is executed.

[0101] By triggering the adaptive update of the backup automatic transfer model through topology change probability, and combining it with system health to achieve interlocking protection, and then selecting the optimal connection path based on path capacity margin, a three-level progressive decision-making mechanism of trigger judgment, feasibility verification, and path selection is formed. This effectively avoids the risk of over-level tripping and load power loss caused by the fixed model or blind switching of traditional backup automatic transfer, and achieves synergistic optimization of safety and power supply reliability.

[0102] As one possible implementation, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements an adaptive backup and automatic transfer monitoring method for a multi-source power grid.

[0103] As one possible implementation, this embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements an adaptive backup and automatic transfer monitoring method for a multi-source power grid.

[0104] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive backup power automatic switching monitoring method for a multi-source power grid, characterized in that, The specific steps include the following: Acquire smart grid operation data, which includes at least charging status signals, output action signals, and abnormal alarm signals; The smart grid operation data is preprocessed to obtain preprocessed synchronization data frames; Three-phase current and three-phase voltage sequences are extracted from the preprocessed synchronization data frames, and fault category labels are identified based on the three-phase current and three-phase voltage sequences. A power grid graph structure is constructed using each power device in the power grid as a node and the transmission line as an edge, and the fault category label is used as the initial feature vector of the node. The power grid graph structure is input into a graph encoder to extract the global graph embedding; The global graph is embedded into multiple decoders for parallel decoding, and the system health, topology change probability, accurate fault type and optimal power supply path are output respectively. The path capacity margin is obtained based on system health, topology change probability, precise fault type, and optimal power supply path; Based on system health, topology change probability, and path capacity margin, adaptive backup and automatic transfer action judgment and optimal backup power supply connection path decision are made.

2. The adaptive backup power automatic throw-in monitoring method of a multi-source power grid according to claim 1, characterized in that, The preprocessing of smart grid operation data specifically includes: The running data is subjected to time alignment processing of multi-source heterogeneous data to generate synchronized data with a unified time base; The synchronized data is subjected to data purification processing, which includes at least outlier detection and repair, and missing value imputation. The purified charging status signal, the output action signal, and the abnormal alarm signal are fused together according to a unified time tag to generate a preprocessed synchronous data frame.

3. The method for monitoring the adaptive backup power automatic switching of the multi-source power grid according to claim 1, characterized in that, The step of extracting three-phase current and three-phase voltage sequences from the preprocessed synchronization data frames, and identifying fault category labels based on the three-phase current and three-phase voltage sequences, specifically includes: Extract the three-phase current sequence and the three-phase voltage sequence from the preprocessed synchronization data frame; Wavelet packet decomposition is performed on the three-phase current sequence and the three-phase voltage sequence to extract the energy distribution in each frequency band, calculate the energy entropy of each frequency band, and construct an initial high-dimensional fault feature vector. Kernel principal component analysis is used to reduce the dimensionality of the initial high-dimensional fault feature vector to obtain a low-dimensional sensitive feature set after dimensionality reduction. The low-dimensional sensitive feature set is input into a pre-trained support vector machine classifier, which outputs fault category labels.

4. The method for monitoring adaptive backup automatic switching of a multi-source power grid according to claim 1, characterized in that, After constructing the power grid diagram structure, the process also includes updating the power grid diagram structure: Acquire real-time operation data of the smart grid, wherein the real-time operation data includes at least charging status signals, output action signals, abnormal alarm signals and measurement section data; Based on the exit action signal and the abnormal alarm signal, the connection status of the edges in the power grid diagram structure is updated to generate a dynamic topology model, which reflects the actual physical connection relationship of the power grid at the current moment. Based on the measured cross-sectional data, update each node and each edge in the power grid diagram structure; The dynamic topology graph model is associated and fused with the refreshed node feature values ​​and edge feature values ​​to generate a real-time power grid graph structure with time-stamped correction labels.

5. The adaptive BRT monitoring method of multi-source power grid according to claim 1 or 4, characterized in that, The step of inputting the power grid graph structure into the graph encoder to extract the global graph embedding includes: Construct an adjacency matrix based on the physical connection relationships between nodes in the power grid diagram structure; Construct a node feature matrix based on the initial feature vectors of the nodes in the power grid diagram structure; The adjacency matrix and the node feature matrix are input into the graph encoder. The graph encoder uses a graph convolutional network to perform message passing and updating on the node feature matrix through multi-layer graph convolution operations to generate a node-level embedding matrix. The node-level embedding matrix is ​​subjected to global pooling to obtain the global graph embedding.

6. The adaptive backup power automatic throw-in monitoring method of a multi-source power grid according to claim 5, characterized in that, The parallel decoding includes: The global graph is embedded, copied, and input into the health regression decoder, topology change binary classification decoder, fault type multi-classification decoder, and path planning decoder respectively. The health regression decoder performs multi-layer fully connected mapping on the global graph embedding and outputs the system health, which characterizes the overall operating status of the power grid. The topology change binary classifier performs multi-layer fully connected mapping on the global graph embedding and outputs the probability of a change in the power grid topology. The fault type multi-class decoder performs multi-layer fully connected mapping on the global graph embedding to output the precise fault type; The path planning decoder, based on the global graph embedding and power grid topology connection relationship, uses a strategy network or heuristic search algorithm to output the optimal power supply path that satisfies line capacity constraints and voltage drop constraints.

7. The adaptive backup power automatic switching monitoring method of a multi-source power grid according to claim 6, characterized in that, The path capacity margin is obtained based on system health, topology change probability, precise fault type, and optimal power supply path, including: Determine whether to trigger graph structure re-update based on the probability of topology change: if the probability of topology change is greater than a preset threshold, return to the step of real-time update of the power grid graph structure; otherwise, perform capacity margin calculation based on the current power grid graph structure. A fault impact coefficient is determined based on the precise fault type, and the fault impact coefficient is used to characterize the degree of impact of different fault types on the allowable transmission capacity of the equipment. A health correction coefficient is determined based on the system health status, and the health correction coefficient is positively correlated with the system health status. Extract several nodes and edges contained in the optimal power supply path, and obtain the equipment parameters of the several nodes and edges from the real-time power grid graph structure. The equipment parameters include at least the rated capacity and the current real-time load. Calculate the path capacity margin based on the fault impact factor, health correction factor, rated capacity, and current real-time load; Iterate through all devices and lines on the optimal power supply path, and take the minimum path capacity margin as the path capacity margin of that path.

8. The method for monitoring adaptive backup automatic switching of a multi-source power grid according to claim 1, characterized in that, The adaptive backup power supply action judgment and optimal backup power supply connection path decision based on system health, topology change probability, and path capacity margin includes: The automatic backup action is triggered based on the aforementioned topology change probability. If the probability of topology change is greater than the first preset threshold, it is determined that the power grid topology has changed, the state of the automatic transfer switch model is changed, and the automatic transfer switch control model is analyzed and updated in real time through the power grid automatic topology search algorithm so that the automatic transfer switch model can adapt to the current power grid operation mode. If the probability of topology change is not greater than the first preset threshold, then the current standby self-starting model state is maintained; When changing the status of the backup automatic transfer model, the feasibility of the backup automatic transfer action is determined based on the system health status: If the system health is lower than the second preset threshold, the power grid as a whole is determined to be in a risky operating state, a backup automatic transfer blocking signal is generated, and the backup power supply operation is prohibited. If the system health is not lower than the second preset threshold, then it is determined that the backup automatic switching action is allowed; When it is determined that the automatic backup power supply action is permitted, the backup power supply selection and connection path decision are made based on the aforementioned path capacity margin: Extract the path capacity margin corresponding to each candidate backup power source from the optimal power supply path and its alternative paths; The candidate path with the largest path capacity margin is selected as the optimal backup power supply path. If the path capacity margin of the optimal backup power supply path is greater than the third preset threshold, then the backup power supply operation on that path is executed.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the adaptive backup and automatic transfer monitoring method for multi-source power grids as described in any one of claims 1 to 8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When executed by the processor, the program implements the adaptive backup and automatic transfer monitoring method for multi-source power grids as described in any one of claims 1 to 8.

Images