Rapid fault positioning method, system and equipment based on three-section protection and FA cooperative protection, and medium

Abstract

The invention discloses a rapid fault positioning method, system and device based on three-section protection and FA cooperative protection, and a medium, and the method specifically comprises the steps: carrying out the detection and judgment through employing a preset three-section action rule, and obtaining a real-time protection action signal; acquiring action time sequence data of each protection device of the power distribution network in real time, constructing an FA fault feature matrix according to the action time sequence data, the electrical quantity information and the topological structure, analyzing time sequence association of the FA fault feature matrix by using an LSTM neural network, and generating an FA fault probability distribution diagram; based on the real-time protection action signal and the FA fault probability distribution diagram, a fault positioning algorithm is adopted for collaborative analysis, and a fault positioning result is obtained; and dynamically planning a load transfer path of the power distribution network by adopting a reinforcement learning algorithm according to a fault positioning result, and generating an optimal isolation and recovery scheme. The method is suitable for various complex power distribution network structures and operation modes, and can realize rapid and accurate fault positioning and processing under different fault conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution networks, and in particular to a method, system, equipment and medium for quickly locating faults based on three-stage protection and FA collaborative protection. Background Art

[0002] In power systems, rapid and accurate fault location and subsequent isolation and restoration of power supply are key factors in ensuring power supply reliability and power quality. Traditionally, the three-stage protection mechanism, including current quick-break protection, time-limited current quick-break protection and overcurrent protection, as the basic protection method in power systems, has achieved fault detection and response to a certain extent. However, with the increasing complexity of distribution network structures, especially the emergence of a large number of branch and ring network structures, the three-stage protection has gradually revealed its limitations in terms of the accuracy and speed of fault location.

[0003] Specifically, the application of traditional three-stage protection in distribution networks faces many challenges. On the one hand, quick-break protection is often difficult to achieve perfect coordination with the upper-level protection, which easily leads to overlap or omission of the protection range, affecting the accuracy and reliability of the protection. On the other hand, although overcurrent protection can cover a wider range of faults, its long action delay often cannot meet the needs of modern power systems for rapid fault handling.

[0004] At the same time, as an advanced means of monitoring and controlling distribution networks, feeder automation (FA) technology has significantly improved the automation level of distribution networks by real-time monitoring of the operating status of distribution networks and combining intelligent algorithms for fault location, isolation and power restoration. However, a single FA system also has shortcomings in some cases. For example, when the fault signal is complex or there are multiple fault points, the FA system may not be able to determine the fault location in a timely and accurate manner. In addition, the performance of the FA system depends largely on the reliability and speed of the communication network. Once the communication network fails or is delayed, the performance of the FA system will be seriously affected.

[0005] In order to overcome the limitations of traditional three-stage protection and separate FA systems, the existing technology has proposed some improvement schemes. For example, by introducing the current criterion and voltage-assisted locking mechanism, attempts are made to improve the accuracy and reliability of protection. However, these schemes do not fundamentally solve the technical difficulties of adaptive adjustment of protection settings and linkage with FA logic. Especially in complex topological structures such as four-segment and three-connection overhead lines, traditional protection methods are prone to problems such as over-tripping, which further affects the stable operation of the distribution network. Summary of the invention

[0006] The purpose of the present invention is to provide a method, system, device and medium for quickly locating faults based on three-stage protection and FA collaborative protection. By real-time acquisition and preprocessing of electrical quantity information and coordinating three-stage protection and FA technology, faults can be quickly located, thereby improving the fault handling efficiency and reliability of the distribution network. The method is suitable for various complex distribution network structures and operating modes, and can achieve fast and accurate fault location and processing under different fault conditions to solve at least one of the above-mentioned prior art problems.

[0007] In a first aspect, the present invention provides a method for quickly locating faults based on three-stage protection and FA collaborative protection, the method specifically comprising:

[0008] Collecting electrical quantity information of each monitoring point in the distribution network in real time, and obtaining the topological structure and real-time operation mode of the distribution network by preprocessing and extracting features of the electrical quantity information;

[0009] The protection range sensitivity is dynamically calculated according to the topological structure and real-time operation mode. Based on the protection range sensitivity and electrical quantity information, the preset three-stage action rules are used for detection and judgment to obtain real-time protection action signals.

[0010] The action timing data of each protection device of the distribution network is collected in real time, and the FA fault feature matrix is ​​constructed according to the action timing data, electrical quantity information and topological structure. The LSTM neural network is used to analyze the time series correlation of the FA fault feature matrix to generate the FA fault probability distribution diagram;

[0011] Based on the real-time protection action signal and FA fault probability distribution diagram, the fault location algorithm is used for collaborative analysis to obtain the fault location result;

[0012] According to the fault location results, the reinforcement learning algorithm is used to dynamically plan the load transfer path of the distribution network and generate the optimal isolation and restoration plan.

[0013] In a second aspect, the present invention provides a system for quickly locating faults based on three-stage protection and FA collaborative protection, the system specifically comprising:

[0014] The data acquisition and preprocessing module is used to collect the electrical quantity information of each monitoring point in the distribution network in real time, and obtain the topological structure and real-time operation mode of the distribution network by preprocessing and extracting the characteristics of the electrical quantity information;

[0015] The three-stage protection module is used to dynamically calculate the protection range sensitivity according to the topological structure and real-time operation mode. Based on the protection range sensitivity and electrical quantity information, it uses the preset three-stage action rules to perform detection and judgment to obtain real-time protection action signals;

[0016] FA fault analysis module, which is used to collect the action timing data of each protection device of the distribution network in real time, build the FA fault feature matrix according to the action timing data, electrical quantity information and topological structure, and use the LSTM neural network to analyze the time series correlation of the FA fault feature matrix to generate the FA fault probability distribution diagram;

[0017] The collaborative control module is used to perform collaborative analysis based on the real-time protection action signal and the FA fault probability distribution diagram using the fault location algorithm to obtain the fault location result;

[0018] The fault self-healing module is used to dynamically plan the load transfer path of the distribution network based on the fault location results and generate the optimal isolation and recovery plan using the reinforcement learning algorithm.

[0019] In a third aspect, the present invention provides a computer device comprising: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, it implements a method for quickly locating faults based on three-stage protection and FA collaborative protection as described in any one of the above methods.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for quickly locating faults based on three-stage protection and FA collaborative protection as described in any one of the above methods is implemented.

[0021] Compared with the prior art, the present invention has at least one of the following technical effects:

[0022] 1. The present invention can quickly locate faults by real-time acquisition and preprocessing of electrical quantity information and coordinated three-stage protection and FA technology, thereby improving the fault handling efficiency and reliability of the distribution network. It is suitable for various complex distribution network structures and operation modes, and can achieve fast and accurate fault location and processing under different fault conditions.

[0023] 2. The present invention can quickly isolate the fault area based on the accurate fault location results, and perform load transfer and power supply restoration operations, thereby reducing the power outage time and scope and improving power supply reliability.

[0024] 3. The present invention can more accurately locate the fault position and shorten the fault processing time by collecting electrical quantity information in real time and dynamically calculating the sensitivity of the protection range, combining the three-stage action rule and the timing correlation analysis of the FA fault characteristic matrix.

[0025] 4. The present invention can dynamically adjust the protection range and setting value according to the topological structure and real-time operation mode of the distribution network, effectively solving the problem that traditional protection setting values ​​are difficult to adjust adaptively.

[0026] 5. The present invention dynamically plans the load transfer path through the reinforcement learning algorithm and generates the optimal isolation and restoration plan, which can minimize the impact of faults on the operation of the distribution network and improve the power supply reliability and power quality.

[0027] 6. The present invention adopts wavelet threshold denoising and Lagrange interpolation method to pre-process electrical quantity information, dynamically reconstruct the topological graph model, and accurately obtain the real-time operation mode of the distribution network.

[0028] 7. The present invention dynamically calculates the sensitivity of the protection range according to the topological structure and the real-time operation mode, generates an adaptive three-stage protection criterion set, and improves the accuracy and flexibility of protection.

[0029] 8. The three-stage protection criterion set of the present invention specifies the current threshold, action time and related parameters to ensure the rapidity and reliability of the protection action.

[0030] 9. The present invention uses a two-layer bidirectional LSTM network to analyze the FA fault feature matrix, generate a fault probability distribution map, and improve the timing correlation and accuracy of fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 It is a flowchart of a method for quickly locating faults based on three-stage protection and FA collaborative protection provided by an embodiment of the present invention;

[0033] Figure 2 It is a structural diagram of a system for quickly locating faults based on three-stage protection and FA collaborative protection provided by an embodiment of the present invention;

[0034] Figure 3 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0036] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0037] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0038] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0039] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0040] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0041] In the embodiment of the present application, the execution subject of the process includes a terminal device, which includes but is not limited to: a server, a computer, a smart phone, a tablet computer, and other devices capable of executing the method disclosed in the present application. Figure 1 A schematic diagram of a process of a method for quickly locating a fault based on three-stage protection and FA coordinated protection disclosed in the first embodiment of the present invention is shown, and the details are as follows:

[0042] S101, collecting electrical quantity information of each monitoring point in the distribution network in real time, and obtaining the topological structure and real-time operation mode of the distribution network by preprocessing and extracting features of the electrical quantity information.

[0043] In this embodiment, current transformers and voltage transformers are installed at various key nodes of the distribution network, such as the outlet of the substation, the branch line node, etc., and the collected current and voltage signals are converted into digital signals and transmitted to the data acquisition unit. The data acquisition unit uses a digital filtering algorithm, such as an FIR filter, to filter the collected signals, remove high-frequency noise and interference signals, verify and filter the electrical quantity data, and remove abnormal values ​​and erroneous data. The data is standardized to meet the requirements of subsequent analysis and processing. For example, a hash table index method can be used to improve the prefix matching speed, select a hash function suitable for the measured data of the power system, and improve the matching efficiency of the prefix string in the dictionary, thereby improving the algorithm execution speed.

[0044] Use deep learning models (such as convolutional neural networks, recursive neural networks, etc.) to extract features from preprocessed electrical quantity data. By training the model, it can automatically capture key features in electrical quantity data, such as voltage fluctuations, current changes, power factor, etc. For example, a deep learning model suitable for processing power grid data can be designed and constructed, using data processing methods including convolutional neural networks, recursive neural networks, long short-term memory, autoencoders, and generative adversarial networks to process image data and time series features of power grid equipment and capture complex features and patterns of power grid data.

[0045] Based on the extracted electrical quantity characteristics, combined with the physical connection relationship and operation rules of the distribution network, the topological structure analysis is carried out. By constructing a formal basic model for topological analysis, the topological analysis process is specified and verified to ensure the accuracy of the analysis results. For example, the power system equipment can be selected as the smallest description unit, and the basic electrical topology formed by the physical connection through the equipment state quantity information can be extracted and expressed.

[0046] Combine the topology analysis results with the real-time collected electrical quantity information to analyze the real-time operation mode of the distribution network. Evaluate key indicators such as the distribution network's operating status, load conditions, power factor, and discover potential operating risks and problems in a timely manner.

[0047] In this embodiment, the accuracy and reliability of the data are ensured by real-time acquisition and preprocessing of electrical quantity information, and feature extraction is performed using a deep learning model to improve the efficiency and accuracy of data analysis. Through topological structure analysis, the topological model of the distribution network can be automatically constructed and updated, providing accurate topological information support for the planning, design and operation of the distribution network. Through real-time operation mode analysis, abnormal conditions and potential risks in the distribution network can be discovered in a timely manner, providing timely and accurate operation information for operation and maintenance personnel, and supporting rapid response and decision-making.

[0048] S102, dynamically calculating the protection range sensitivity according to the topological structure and the real-time operation mode, and based on the protection range sensitivity and the electrical quantity information, using the preset three-stage action rule to perform detection and judgment to obtain a real-time protection action signal.

[0049] In this embodiment, the three-stage protection device is designed using a microprocessor and a corresponding protection algorithm. According to the operating parameters of the distribution network and the short-circuit current calculation results, the setting values ​​of the current quick-break protection, the time-limited current quick-break protection and the overcurrent protection are reasonably set. When the collected current exceeds the corresponding setting value, the protection device sends an action signal according to the set time limit.

[0050] Specifically, the sensitivity of each protection area is dynamically calculated based on the topological structure model and real-time operation mode (such as load conditions, power supply status, etc.). The sensitivity is usually expressed as the ratio of the short-circuit current at the end of the protection area to the setting current of the circuit breaker, ensuring that the circuit breaker can be operated in time when a fault occurs.

[0051] The first stage (current quick-break protection): The setting value is set relatively high, and there is no setting time delay. Once the current is detected to exceed this setting value, the protection device will act immediately and quickly cut off the fault circuit. It is mainly used to protect the short-circuit fault near the line, and the action is fast but the protection range is limited.

[0052] The second stage (time-limited current quick-break protection): The setting value is lower than the quick-break protection, and a setting time delay is introduced. This protection will only operate when the line current reaches the setting value and lasts for a period of time. The setting value setting must ensure that it can cover the full length of this line and appropriately extend to the first half of the next-level line. It serves as the main protection of this line and assumes the remote backup protection responsibility of the next-level line.

[0053] The third stage (time-limited overcurrent protection): The setting value is lower than the previous two, and the time delay is longer. It not only ensures that the full length of the line is covered, but also has a longer protection range than the time-limited current quick-break protection. As the backup protection of the line, it also undertakes the remote backup protection task of the next level or even lower level lines.

[0054] Based on the protection range sensitivity and real-time collected electrical quantity information, the distribution network is monitored in real time. The real-time monitored data is matched with the preset three-stage action rules. Once the match is successful, the fault is detected, and the corresponding protection action signal is immediately triggered to control the circuit breaker to cut off the fault circuit.

[0055] In this embodiment, by dynamically calculating the sensitivity of the protection range, it is possible to more accurately determine whether the fault occurs in the protection area to avoid false operation or refusal to operate. The three-stage action rule is adopted, which can respond quickly according to different fault conditions, cut off the fault circuit in time, and reduce the impact of the fault on the power grid. Through real-time monitoring and rapid protection action, it is possible to effectively prevent the expansion and spread of faults, ensure the stable operation of the power grid, improve the power supply reliability of the power grid, reduce the time and scope of power outages, and improve user satisfaction. Dynamically adjusting the protection range and sensitivity according to the topological structure and real-time operation mode can optimize the protection configuration and reduce unnecessary investment in protection equipment. Reduce operation and maintenance costs and improve the economic benefits of the power grid.

[0056] S103, collect the action timing data of each protection device of the distribution network in real time, build an FA fault feature matrix according to the action timing data, electrical quantity information and topological structure, and use the LSTM neural network to analyze the timing correlation of the FA fault feature matrix to generate a FA fault probability distribution map.

[0057] In this embodiment, the monitoring terminal of the FA system collects the voltage and current information of each monitoring point in real time, and transmits the data to the master station system through the communication network. The master station system establishes a fault analysis model based on the topological structure and electrical parameters of the distribution network. When a fault occurs, the master station system determines whether there is a fault by comparing the changes in electrical quantities at different monitoring points, such as voltage sag, current mutation, etc., and preliminarily determines the approximate area of ​​the fault using the fault location algorithm.

[0058] Specifically, in the distribution network, a data acquisition module is configured for each protection device to collect its action timing data in real time, including action time, action type (such as tripping, closing), etc. The action timing data, electrical quantity information and topological structure data of the protection device are integrated to form a data set containing multi-dimensional features. Features related to fault analysis are extracted from the integrated data set, such as the interval of the action timing, the amplitude of the change of the electrical quantity, the connectivity of the topological structure, etc. The extracted features are arranged in time series to construct an FA fault feature matrix. Each row of the matrix represents a feature vector at a time point, and each column represents a feature dimension.

[0059] Build an LSTM neural network model that has the ability to process time series data and can capture the time series associations in the feature matrix. Use historical fault data to train the LSTM model and adjust the model parameters so that it can accurately identify fault features and predict fault probabilities. Input the FA fault feature matrix into the trained LSTM model, and the model will analyze the time series associations in the feature matrix to capture the precursors and patterns of faults. The LSTM model will output the fault probability at each time point, which reflects the possibility of faults occurring at different time points. Visualize the calculated fault probabilities according to the time series to generate a FA fault probability distribution graph. This graph can intuitively reflect the changing trend of the fault probability over time.

[0060] In this embodiment, by collecting the action timing data, electrical quantity information and topological structure information of the protection device in real time, the FA fault feature matrix is ​​constructed, which can more comprehensively reflect the operation status and fault characteristics of the power grid. The LSTM neural network is used to analyze the timing association, which can capture the precursors and laws of the fault occurrence and improve the accuracy of fault prediction. The FA fault probability distribution map is generated, which can intuitively reflect the changing trend of the fault probability over time and provide intuitive fault warning information for the operation and maintenance personnel. The operation and maintenance personnel can formulate more reasonable maintenance plans and emergency measures based on the fault probability distribution map to improve the reliability and safety of the power grid.

[0061] S104, based on the real-time protection action signal and the FA fault probability distribution diagram, a fault location algorithm is used to perform collaborative analysis to obtain a fault location result.

[0062] In this embodiment, the collaborative control unit uses a high-performance computer and data processing algorithm to receive the action signal of the three-stage protection device and the fault analysis result of the FA system. First, according to the action range of the three-stage protection and the fault area determined by the FA system, the area where the fault is impossible to exist is eliminated to reduce the fault range. Then, the fault location algorithm based on the electrical quantity characteristics is used to accurately calculate the reduced fault area and determine the location of the fault point.

[0063] Specifically, an algorithm suitable for fault location in the distribution network is selected, such as a fault location algorithm based on the impedance method or a fault location algorithm based on the traveling wave method. Taking the impedance method as an example, the algorithm uses the voltage and current measured at the time of the fault to calculate the impedance of the fault circuit, and then determines the fault location based on the impedance parameters of the line.

[0064] The real-time protection action signal and the FA fault probability distribution map are integrated. For example, the protection device location information in the protection action signal is matched with the location information in the fault probability distribution map to determine the fault probability corresponding to each protection action.

[0065] Based on the protection action signal, determine the interval where the fault may occur. For example, if a circuit breaker trips, the fault may occur in the line interval protected by the circuit breaker.

[0066] In the initially determined fault interval, the fault probability information provided by the FA fault probability distribution map is used to further narrow the fault location range. The location with the highest fault probability is selected as the final fault location result.

[0067] Compare and verify the fault location results with the actual fault location. The actual fault location information can be obtained through on-site inspections, equipment testing, etc. If the location results match the actual fault location, it means that the fault location algorithm is effective; if not, the algorithm needs to be optimized and adjusted. Output the fault location results in an intuitive way, such as marking the fault location on a map, generating a fault report, etc. At the same time, transmit the results to the relevant operation and maintenance personnel so that they can take maintenance measures in time.

[0068] In this embodiment, by combining the real-time protection action signal and the FA fault probability distribution map, the action information of the protection device and the probability information of the fault occurrence can be fully utilized to more accurately determine the fault location. Compared with the traditional single fault location method, this method takes more factors into consideration and reduces the positioning error. Accurate fault location results can help operation and maintenance personnel quickly find the fault point and reduce the time for troubleshooting. Operation and maintenance personnel can go directly to the fault site for maintenance based on the positioning results, which improves the efficiency of fault repair, shortens the power outage time, and reduces the power outage loss. Timely and accurate fault location helps to quickly restore the normal operation of the power grid, reduce the impact of faults on the power grid, improve the reliability and stability of the power grid, and ensure the power demand of users.

[0069] S105, based on the fault location result, a reinforcement learning algorithm is used to dynamically plan the load transfer path of the distribution network to generate an optimal isolation and restoration plan.

[0070] In this embodiment, the collaborative control unit issues control instructions based on the fault location results to control the corresponding switchgear actions. For example, when it is determined that the fault point is located on a branch line, the circuit breaker of the branch line is controlled to trip and isolate the fault area. At the same time, the FA system automatically selects the appropriate contact switch for closing operation based on the topological structure and load conditions of the distribution network, transfers the load in the non-fault area to other power sources, and restores power supply. Equipped with a backup battery power supply to ensure that the local protection action can be completed independently in the event of a power outage.

[0071] Specifically, the topological structure, load distribution, switch status, etc. of the distribution network are used as the state space of reinforcement learning. Each state represents the operating state of the distribution network at a certain moment. The switch operations (such as closing and tripping) in the distribution network are used as the action space of reinforcement learning. Each action represents a change in the topological structure of the distribution network. A reward function is designed to encourage the algorithm to find the optimal load transfer path. The reward function can include multiple indicators such as load recovery amount, line loss reduction, and number of switch operations. For example, when the load recovery amount increases, the line loss decreases, or the number of switch operations decreases, a positive reward is given; otherwise, a negative reward is given.

[0072] Use deep reinforcement learning algorithms (such as DQN, DDPG, etc.) to train the load transfer problem of the distribution network. During the training process, the algorithm learns how to find the optimal load transfer path by constantly trying different combinations of switch operations. Based on the trained reinforcement learning model, the load transfer path of the distribution network is dynamically planned according to the real-time collected electrical quantity information and fault location results. The algorithm will select the optimal action (i.e., switch operation) according to the current state (i.e., the operating state of the distribution network) to achieve rapid load recovery and minimize line loss. Based on the fault location results and the load transfer path generated by the reinforcement learning algorithm, formulate the optimal isolation plan. The isolation plan includes operations such as cutting off the fault line and isolating the fault area to prevent the fault from expanding and spreading. After isolating the fault area, the load in the non-fault area is transferred to other reliable power supply lines according to the load transfer path. The recovery plan should include operations such as closing operations and load distribution to ensure that the load in the non-fault area can be restored as soon as possible.

[0073] In this embodiment, the reinforcement learning algorithm can automatically learn the operating rules and load characteristics of the distribution network, find the optimal load transfer path, improve the efficiency and accuracy of load transfer, and reduce the risk of manual intervention and misoperation. By dynamically planning the load transfer path and generating the optimal isolation and recovery plan, the power supply to the non-fault area can be quickly restored after a fault occurs, reducing the time and scope of power outages and improving the power supply reliability of the power grid. The reinforcement learning algorithm will take line loss factors into consideration when planning the load transfer path, and try to select a path with less line loss for transfer, which reduces the line loss rate of the power grid and improves the economic benefits of the power grid. At the same time, the automated load transfer process reduces the workload of operation and maintenance personnel and reduces operation and maintenance costs.

[0074] In some embodiments, in the above step S101, the obtaining of the topological structure and real-time operation mode of the distribution network by preprocessing and extracting features of the electrical quantity information specifically includes:

[0075] The electrical quantity information is decomposed and reconstructed by using a wavelet threshold denoising algorithm, the electrical quantity information is data compensated by using a Lagrange interpolation method, and the electrical quantity information of each monitoring point after decomposition, reconstruction and data compensation is time-aligned to obtain standard electrical quantity information;

[0076] Constructing a voltage and current video feature matrix based on the standard electrical quantity information, calculating node associations on the voltage and current video feature matrix, and generating a multidimensional feature vector set that characterizes the spatiotemporal correlation of electrical quantities;

[0077] Using the node association threshold and switch state quantity in the multi-dimensional feature vector set, dynamically generating and updating the adjacency matrix, and reconstructing the topological graph model of the distribution network;

[0078] Based on the topology model, through distributed power flow calculation and overload index analysis, the operation status mapping table and abnormal warning signal of the distribution network are output to obtain the real-time operation mode of the distribution network.

[0079] In this embodiment, the wavelet threshold denoising algorithm is used to decompose and reconstruct the electrical quantity information, and the decomposed and reconstructed electrical quantity information is compensated by the Lagrange interpolation method to obtain the preliminary processed electrical quantity information. The preliminary processed electrical quantity information of each monitoring point is time aligned to obtain the standard electrical quantity information. The voltage and current video feature matrix is ​​constructed according to the standard electrical quantity information, and the numerical distribution characteristics of each element in the matrix are calculated to obtain the feature matrix distribution data. The node association degree is calculated for the feature matrix distribution data to generate a correlation degree set representing the relationship between nodes. Multidimensional features are extracted through the correlation degree set to generate a multidimensional feature vector set representing the spatiotemporal correlation of electrical quantities.

[0080] Through node association and switch status, the correlation data is obtained, the adjacency matrix is ​​dynamically updated, and the reconstructed matrix data is obtained. The reconstructed matrix data is used to generate the distribution network topology model and determine the network structure. For the topology model, distributed power flow calculation is performed to obtain power flow distribution data. Through the power flow distribution data, the overload index is calculated and the overload area is determined. If the overload index exceeds the threshold setting, an abnormal warning signal is generated and the warning area is determined. The operating status data is obtained, and the operating status mapping table is generated in combination with the abnormal warning signal to obtain the real-time operating mode. The change trend is extracted from the real-time operating mode, the threshold setting and the matrix update strategy are updated, and the optimized operating parameters are obtained.

[0081] For example, assuming that the voltage signal of a distribution network monitoring point is affected by high-frequency noise, the signal can be decomposed into different frequency components by wavelet transform, the noise part can be filtered out by setting a threshold, and then a smooth signal can be reconstructed. This method can effectively improve the data quality and provide a reliable basis for subsequent analysis. In one possible implementation, if the original voltage signal amplitude is 220V, it fluctuates to 215-225V after superimposing noise, and can be restored to a stable value close to 220V after wavelet denoising, significantly reducing the error. The Lagrange interpolation method is used to compensate the data of electrical quantity information, mainly to solve the problem of missing data or uneven sampling. Specifically, if the current data at a certain moment is missing due to equipment failure at a certain monitoring point, the Lagrange interpolation can be used to estimate the value at t2 to be about 11A based on the known data points before and after the moment, such as 10A at t1 and 12A at t3. This compensation method is simple and efficient, and can ensure the continuity of data. Preferably, combined with the actual business scenario, if there are many missing points, time window smoothing can also be introduced to further improve the compensation accuracy. In a possible implementation, the electrical quantity information after decomposition, reconstruction and data compensation is time-aligned, which can be achieved through timestamp calibration. For example, due to different sampling frequencies, the data timestamps of multiple monitoring points in the distribution network may be misaligned, such as the data at point A is 10:00:00 and that at point B is 10:00:02. By aligning to the unified time axis 10:00:00, standard electrical quantity information is generated. This alignment lays the foundation for subsequent matrix construction and avoids the influence of time deviation on the analysis results. When constructing the voltage and current video feature matrix based on the standard electrical quantity information, the voltage and current changes over time can be regarded as "video frames". Assuming that a distribution network has 3 nodes, each node records the voltage and current data at 5 moments, a 3×10 matrix can be constructed, in which the rows represent the nodes and the columns represent the voltage and current sequences. The node association degree of the matrix is ​​calculated, for example, the spatiotemporal correlation of the electrical quantities between nodes is analyzed by the correlation coefficient, and a multidimensional feature vector set is generated. This method intuitively reflects the dynamic characteristics of the power grid and helps to discover abnormal patterns. When the node association threshold and switch state quantity in the multidimensional feature vector set are used to dynamically generate and update the adjacency matrix, it can be understood as reshaping the topological relationship according to the electrical connection and switch state. For example, if the association degree of a node is lower than the threshold of 0.8 and the switch is disconnected, the corresponding element of the adjacency matrix is ​​set to 0, otherwise it is set to 1. This dynamic update can reflect the topological changes of the distribution network in real time, such as line disconnection or grid-connected operation, and improve the adaptability of the model. When performing distributed power flow calculation and overload index analysis based on the topological graph model, the operating status can be judged by the power flow direction of each node. In one possible implementation, assuming that the power flow of a certain line is 500kW, the rated capacity is 600kW, and the overload index is 83%, which is lower than the warning threshold of 90%, then the normal state mapping table is output. If the power flow of another line reaches 650kW and the overload index exceeds the standard, an abnormal warning signal is triggered. This analysis method can quickly locate the problem area and ensure the safe operation of the power grid.It should be noted that the above method forms a complete technical chain through denoising, compensation, alignment, feature extraction, topology reconstruction and state analysis. Each step improves data accuracy and analysis reliability, and ultimately achieves accurate monitoring and optimized management of the real-time operation mode of the distribution network.

[0082] In some embodiments, in the above step S102, the protection range sensitivity is dynamically calculated according to the topological structure and the real-time operation mode, and based on the protection range sensitivity and the electrical quantity information, a preset three-stage action rule is used for detection and judgment to obtain a real-time protection action signal, specifically including:

[0083] Based on the topology and the real-time operation mode, a directional sensitivity coefficient is set for each branch of the topology, wherein the sensitivity coefficient is used to characterize the response strength of the branch to a downstream fault;

[0084] Constructing a sensitivity matrix according to the directional sensitivity coefficients, performing a Hadamard product operation on the sensitivity matrix and the adjacency matrix, eliminating invalid branches, and obtaining an optimized sensitivity matrix, wherein the sensitivity matrix is ​​used to determine the protection range sensitivity of each branch;

[0085] Based on the sensitivity matrix and the real-time load current in the electrical quantity information, dynamically generate a three-stage protection criterion set with adaptive threshold and time delay characteristics;

[0086] By verifying the topological connectivity of the three-stage protection criterion set, a real-time protection action signal for the fault branch is generated.

[0087] In this embodiment, the directional sensitivity coefficient of each branch is obtained through the topological structure and the operation mode, and the coefficient represents the response strength. The validity of the branch direction is judged by a preset threshold value to obtain an initial sensitivity set. A sensitivity matrix is ​​constructed from the initial sensitivity set. According to the correspondence between the branch direction and the topological structure, the distribution characteristics of the response strength are determined by assigning values ​​to the matrix elements. The adjacency matrix is ​​obtained, and the sensitivity matrix and the adjacency matrix are processed by the Hadamard product operation, and invalid branches are eliminated to obtain the optimized first matrix. The protection range is analyzed by the first matrix. If the response strength of a branch is lower than the preset threshold, it is determined to be an invalid range, and the optimization matrix is ​​updated to the second matrix. The second matrix is ​​classified by the K-means clustering algorithm, and the protection range boundary of each branch is determined according to the classification result to obtain the range allocation matrix. The priority sequence of the fault response is calculated by the range allocation matrix, and the optimization matrix is ​​adjusted according to the priority sequence to obtain the final sensitivity distribution. The protection range sensitivity of each branch is extracted from the final sensitivity distribution, and the linear regression algorithm is used to predict the changing trend of the fault response to obtain the dynamic adjustment strategy of the branch protection.

[0088] The current information is calculated by the sensitivity matrix and the real-time load, and the adaptive threshold and delay characteristics are dynamically generated. According to the adaptive threshold and delay characteristics, the criterion set of the three-stage protection is constructed. By integrating the criterion set, the data structure required for the topological connectivity verification is obtained. If the topological connectivity verification passes, the location of the fault branch is determined. According to the location of the fault branch, the corresponding protection action instruction is generated. Through the protection action instruction, the real-time signal for the fault branch is obtained. The real-time signal is compared with the current information to determine whether the system operation status is normal.

[0089] For example, when setting a directional sensitivity coefficient for each branch of the topological structure in the distribution network, it can be understood as an indicator to measure the impact of the branch on downstream faults. Assuming that a branch connects nodes A and B, if a short circuit fault occurs in the downstream node B, the current flow direction and amplitude of the branch will change significantly. The directional sensitivity coefficient can be determined by analyzing historical fault data and branch location. For example, a branch close to the load center may be assigned a higher coefficient such as 0.9, while a remote branch coefficient is lower such as 0.5. This design can more accurately reflect the response characteristics of the branch to faults. In one possible implementation, when constructing a sensitivity matrix based on the directional sensitivity coefficient, the distribution network can be simplified to a system of three nodes. Assume that the branch sensitivity coefficient from node 1 to node 2 is 0.8, and that from node 2 to node 3 is 0.6, and the matrix is ​​initially constructed in the form of 3×3. By performing a Hadamard product operation with the adjacency matrix, invalid branches are eliminated. For example, if a branch has no electrical connection due to a disconnected switch, its corresponding element is set to 0. The optimized sensitivity matrix clearly shows the sensitivity of the protection range of each branch. For example, the branch protection range from node 1 to node 2 is more inclined to respond quickly to downstream large current faults. Specifically, when a three-stage protection criterion set is generated based on the sensitivity matrix and the real-time load current, it can be dynamically adjusted in combination with the actual operation data. Assuming that the normal load current of a branch is 100A, the sensitivity matrix shows that its coefficient is 0.7, and the current suddenly increases to 500A under a short-circuit fault. The criterion set can be set as follows: the first section is instantaneous at 400A, the second section is delayed by 0.5 seconds and is operated at 300A, and the third section is delayed by 1 second and is operated at 200A. This adaptive threshold and delay characteristic can flexibly respond to different fault scenarios and avoid false or missed actions. Preferably, when the three-stage protection criterion set is integrated and the topological connectivity is verified, the accuracy of the protection action can be ensured by checking the connectivity status of the faulty branch with the upstream and downstream nodes. For example, if a node downstream of a branch has lost power due to line disconnection, the judgment set will remove the protection action signal for it after verification, and only generate real-time protection instructions for branches that still have electrical connections. This method can effectively reduce redundant operations and improve system response efficiency. In a possible implementation, when generating a real-time protection action signal for a faulty branch, the specific location of the fault can be further combined. Assuming that a ground fault occurs in the downstream branch of node 3 in a distribution network, the sensitivity matrix shows that its coefficient is 0.6, and the real-time current jumps to 450A, exceeding the threshold of the first judgment section. The system then generates a protection action signal to cut off the power supply of the branch. This fast-response signal generation method can isolate the fault area in time and ensure the normal operation of other branches. It should be noted that the introduction of the directional sensitivity coefficient makes the protection range more targeted. For example, if the directionality is not set and only the current amplitude is relied on for judgment, the upstream branch may malfunction. By optimizing the sensitivity matrix, the fault impact range can be clearly distinguished and unnecessary power outage areas can be reduced. This technical means significantly improves the accuracy and reliability of distribution network protection.It is understandable that the adaptive characteristics of the three-stage protection criterion set provide flexibility for different operating conditions. During peak load periods, the threshold can be appropriately adjusted upward to tolerate normal fluctuations; while during trough periods, the threshold is lowered to quickly capture anomalies. This dynamic adjustment capability ensures that the protection strategy is highly matched with the operating status of the power grid, providing strong support for real-time monitoring.

[0090] Furthermore, the three-stage protection criterion set specifically includes:

[0091]

[0092]

[0093]

[0094]

[0095] in, Indicates the action current threshold of stage I, Indicates the action current threshold of stage II, Indicates the action current threshold of stage III, It represents the reliability factor used to ensure that the operating current threshold of stage I is higher than the maximum short-circuit current. Indicates the coordination coefficient with the I-stage action current threshold, Indicates the overload protection factor, represents the sensitivity coefficient, Indicates the maximum short-circuit current of the three-phase at the end of the branch. represents the time step constant, represents the sensitivity and time adjustment factor, Indicates the first stage of action time, Indicates the action time of the second stage, Indicates the action time of the third stage, Indicates the real-time load current of the branch. represents the time margin, Indicates the effective value of the current measured by the protection device in real time. represents the rate of change of current, Indicates the duration of current exceeding the threshold.

[0096] In this embodiment, For the first stage (instantaneous quick break), For Section II (limited time quick cut), This is Section III (Time-limited Quick Cut).

[0097] The first stage action current threshold is used for current quick-break protection and is set relatively high to ensure that it can act instantly and quickly cut off the fault current when a serious short-circuit fault occurs in the vicinity of the line. Its value is usually set according to the maximum short-circuit current that may occur at the exit of the next line under the maximum operating mode of the system, and multiplied by the reliability coefficient to ensure selectivity.

[0098] The action current threshold of Section II is used for time-limited current quick-break protection. Its value is lower than Section I but higher than Section III. It is designed to protect the entire length of the line and serve as a backup for the current quick-break protection. Its setting needs to be coordinated with the current quick-break protection of the next section of the line, and a coordination coefficient is usually introduced.

[0099] The third stage action current threshold is used for time-limited overcurrent protection. It has the lowest setting value and is adjusted according to the maximum load current to be avoided and multiplied by the overload protection coefficient to ensure that there will be no false operation during normal operation.

[0100] The reliability coefficient is used to ensure that the operating current threshold of stage I is higher than the maximum short-circuit current, and that it can operate reliably even in the maximum operating mode of the system, thereby improving the reliability of protection.

[0101] The coordination coefficient is used to coordinate the operating current thresholds between adjacent protections to ensure that only the protection closest to the fault point is activated in the event of a fault, thereby achieving protection selectivity.

[0102] The overload protection factor is used in time-limited overcurrent protection to ensure that the operating current threshold is higher than the maximum load current to prevent false operation during normal load fluctuations.

[0103] The sensitivity coefficient is used to measure the sensitivity of the protection to faults and is usually related to the ratio of the minimum short-circuit current to the operating current threshold, ensuring that the protection can operate reliably even in the event of minor faults.

[0104] The maximum three-phase short-circuit current at the end of the branch is used to set the operating current threshold, especially in current quick-break protection, to ensure that the protection range covers important parts of the line.

[0105] The time step constant is used to set the action time difference between adjacent protections, ensuring that only the protection closest to the fault point is actuated first in case of a fault, thus achieving time selectivity.

[0106] The sensitivity and time adjustment factor is used to adjust the action time according to the sensitivity requirements to ensure timely action even at the edge of the protection range or in the event of a minor fault.

[0107] Action time , and They correspond to the action time of Section I, Section II and Section III protection respectively, ensuring that they act in the set sequence and time interval in case of a fault.

[0108] The branch real-time load current is used to monitor the line load in real time and is compared with the action current threshold to determine whether action is required.

[0109] The time margin is used to add a certain margin to the action time to cope with factors such as measurement errors and system fluctuations to ensure the reliability of protection.

[0110] The effective value of the current measured by the protection device in real time is used for real-time comparison with the action current threshold to determine whether protection needs to be started.

[0111] The current change rate is used in some protections to assist in determining the type or severity of a fault and to improve the sensitivity and speed of the protection.

[0112] The current exceeding threshold duration is used in the time-limited overcurrent protection. When the current exceeds the action current threshold and the duration reaches the set value, the protection will be activated to prevent false operation caused by instantaneous overcurrent.

[0113] In this embodiment, by reasonably setting the action current threshold, coordination coefficient and time step constant, it is ensured that only the protection closest to the fault point is actuated in the event of a fault, thereby avoiding over-tripping and expanding the scope of power outage.

[0114] Enhance the speed of protection: The current quick-break protection can act instantly, quickly cut off serious short-circuit faults, and reduce equipment damage and power outage time. By setting parameters such as sensitivity coefficient and monitoring current change rate, it is ensured that the protection can also act reliably in the case of minor faults, thereby improving the safety and reliability of the power grid. The time-limited current quick-break protection and time-limited overcurrent protection can protect the entire length of the line, ensuring timely and effective protection when a fault occurs at any location. By dynamically adjusting the action current threshold and time parameters, the protection can adapt to changes in the system operation mode, improving the adaptability and flexibility of the protection.

[0115] In some embodiments, in the above step S103, the FA fault feature matrix is ​​constructed according to the action timing data, electrical quantity information and topological structure, and the timing correlation of the FA fault feature matrix is ​​analyzed by using the LSTM neural network to generate the FA fault probability distribution diagram, which specifically includes:

[0116] Align and tensor-join the action timing data, electrical quantity information and topological structure to form the FA fault feature matrix;

[0117] Construct a two-layer bidirectional LSTM network, input the FA fault feature matrix into the two-layer bidirectional LSTM network for time series correlation analysis, and obtain the fault probability value of each coordinate of the distribution network;

[0118] The fault probability values ​​of each coordinate of the distribution network are output in the form of a heat map to generate a FA fault probability distribution map.

[0119] In this embodiment, the original data is obtained through the action sequence, electrical quantity signal and topological structure, and the consistency of the three is processed by the data alignment method to obtain the aligned data set. The aligned data set is fused by the tensor splicing technology to generate the FA fault feature matrix. A two-layer bidirectional LSTM network is constructed, and the FA fault feature matrix is ​​input into the network for training to obtain the timing association model. The FA fault feature matrix is ​​analyzed by the timing association model to obtain the fault probability value of the distribution network seat. For the fault probability value, if it exceeds the preset threshold, it is judged that the corresponding distribution network seat has a fault risk, and a fault coordinate set is obtained. According to the fault coordinate set, the coordinate analysis method is used to determine the fault area distribution and obtain the regional fault distribution result. The regional fault distribution result is used as input, and a graphical representation of the fault probability is generated by the heat map algorithm to obtain a preliminary distribution map. If there is an abnormal area in the preliminary distribution map, the graphical parameters are adjusted by comparing the probability values ​​to obtain an optimized distribution map. The key analysis results are extracted for the optimized distribution map to determine the FA fault probability distribution characteristics. The final FA fault probability distribution map is obtained, and the result in the form of a heat map is output.

[0120] For example, when the action time series data, electrical quantity information and topological structure are aligned and tensor spliced, it can be understood as a multi-dimensional data fusion process. The action time series data may include the time series of switch tripping, the electrical quantity information covers the real-time values ​​of current and voltage, and the topological structure reflects the node connection relationship of the distribution network. Assume that a distribution network has 3 nodes, and the branch from node 1 to node 2 undergoes a switch action at 10:00, and the current suddenly increases from 100A to 300A. The topology shows that it is a unidirectional power supply path. After alignment, the data is organized into a three-dimensional tensor, for example, the time dimension is 10 seconds sampling, the electrical quantity dimension includes current and voltage, and the topological dimension marks the branch state. This splicing method can fully preserve the spatiotemporal characteristics of the data and provide a comprehensive basis for subsequent analysis. In one possible implementation, when constructing a two-layer bidirectional LSTM network, it can be regarded as a tool to capture temporal dependencies. The two-layer design enhances the model's ability to extract complex patterns, and the two-way considers both past and future contextual information. Specifically, assume that the input FA fault feature matrix contains the current data of a branch for 5 consecutive minutes, which are 100A, 120A, 150A, 300A, and 200A respectively. The two-layer bidirectional LSTM network will analyze the trend changes of this data. For example, a sudden increase from 150A to 300A may indicate a fault, while a subsequent drop to 200A may indicate fault relief. Through multi-directional time series association analysis, the network can generate a fault probability value for each node, such as 0.85 for node 2 and 0.25 for node 3. This method makes full use of the dynamic characteristics of the data. It should be noted that when the fault probability value is output through a heat map, the heat map can intuitively reflect the fault distribution of the distribution network. For example, the probability of 0.85 for node 2 may be displayed as a red high-risk area, and 0.25 for node 3 may be displayed as a green low-risk area. In one possible implementation, assuming that the distribution network covers 10 nodes, the heat map shows that the probability of nodes close to the load center is generally higher than 0.7, while the probability of edge nodes is lower than 0.4. This visualization method makes it easy for operators to quickly locate high-risk areas and improve decision-making efficiency. Preferably, when generating the FA fault probability distribution map, it can be further refined in combination with the actual scenario. Assuming that the insulation of a branch has decreased due to aging, and the current data has fluctuated many times in the past hour, the LSTM network analysis gives a failure probability of 0.9. In the thermal map, the branch is marked as dark red, indicating that it needs priority maintenance. It is understandable that this distribution map not only reflects the current status, but also provides a basis for preventive maintenance. For example, if the probability of a certain area is high for many consecutive days, equipment replacement can be arranged in advance to reduce the occurrence of failures. In one possible implementation, for time series correlation analysis, its reliability can be verified from different aspects. Assume that the current of a node suddenly increases to 250A when the load is low in the early morning, and no abnormalities are observed during the peak in the daytime.The two-layer bidirectional LSTM analyzes the previous and next time series and determines that the sudden increase in the valley period is a sign of fault with a probability of 0.8, while the probability of normal fluctuation in the peak period is only 0.2. This multi-period comparison enhances the accuracy of the analysis. If the topology shows that there is a branch downstream of the node, the network will further confirm the fault location by combining the branch current distribution. This multi-dimensional support ensures the credibility of the probability value and provides strong technical support for distribution network management.

[0121] In some embodiments, in the above step S104, the method of using a fault location algorithm to perform collaborative analysis based on the real-time protection action signal and the FA fault probability distribution diagram to obtain a fault location result specifically includes:

[0122] Based on the real-time protection action signal and FA fault probability distribution diagram, an improved matrix algorithm is used to analyze and generate a candidate set of fault areas.

[0123] Based on the fault area candidate set, transient current features are extracted using wavelet packet transform to construct a decision tree classifier;

[0124] Based on the fault area candidate set, a Petri net is used to perform temporal reasoning on the protection action logic to obtain a protection action logic rule base;

[0125] A collaborative analysis is performed based on the decision tree classifier and the protection action logic rule base to generate a fault location result.

[0126] In this embodiment, features are extracted from the fault probability value, processed by a matrix algorithm, and a preliminary analysis result is generated. If the preliminary analysis result exceeds a preset threshold, the matrix algorithm is optimized by an improved method to obtain an adjusted result. According to the adjusted result and the distribution diagram, the fault analysis technology is used to determine the candidate range of the region. The candidate range of the region is processed by the candidate generation technology to obtain a candidate set of fault regions. The candidate set of fault regions is obtained, and the candidate set is grouped by a clustering algorithm to determine the final fault region. Through the candidate set of fault regions, the transient current signal is decomposed by wavelet packet transform to obtain a feature set. According to the protection action data, the Petri net is used to model the timing reasoning to obtain a logic rule base. For the feature set, a decision tree classifier is constructed to generate a preliminary classification result. Through the logic rule base, the preliminary classification result is constraint verified to obtain the adjusted classification data. If the adjusted classification data matches the fault region, the fault location result is determined; if not, the feature set is updated by collaborative analysis. The decision tree classifier is re-run using the updated feature set to obtain an optimized classification result. The optimized classification result is fused with the logic rule base to determine the final fault location result.

[0127] Exemplarily, based on the real-time protection action signal and the FA fault probability distribution map, the improved matrix algorithm can be used to generate a candidate set of fault areas. The protection action signal may include the timestamp of the switch tripping, while the FA fault probability distribution map provides the fault probability of each node. Assuming that a distribution network has 5 nodes, node 3 has a protection action at 10:05, and the probability distribution map shows that its probability is 0.9, and the adjacent node 4 is 0.6. The improved matrix algorithm organizes this information into a matrix, combines the signal and probability, and screens out high-risk areas, such as nodes 3 and 4 to form a candidate set. This method can quickly narrow the scope of the fault. In a possible implementation, when using wavelet packet transform to extract transient current features, it can be regarded as a tool for decomposing current signals. Assuming that the current of node 3 suddenly changes from 150A to 350A at the moment of tripping, the wavelet packet transform will decompose this change into high-frequency and low-frequency components, and the high-frequency part reflects the mutation characteristics. By analyzing these features, a decision tree classifier is constructed. For example, when the high-frequency component exceeds a certain threshold, the classifier determines it as a fault signal. This method can effectively distinguish between normal fluctuations and fault transients. Specifically, when using Petri nets to perform sequential reasoning on the protection action logic based on the candidate set of fault areas, it can be understood as a dynamic modeling process. Assuming that after node 3 trips, the relay of node 4 acts at 10:06. The Petri net deduces whether the action sequence conforms to the preset logic through state transition, such as "after node 3 trips, node 4 should respond with a delay". Finally, a rule base is generated, such as "when node 3 fails, node 4 needs to be isolated". This provides a logical basis for subsequent analysis. It should be noted that when the decision tree classifier and the protection action logic rule base are collaboratively analyzed, the two can complement each other. Assume that the decision tree judges that node 3 is the fault source based on the current characteristics with a probability of 0.95, and the rule base verifies that the action of its downstream node 4 is logical. After collaborative analysis, node 3 is confirmed to be the fault point. This combination improves the reliability of positioning. Preferably, in a possible implementation, assuming that the distribution network covers 8 nodes, the current of node 5 suddenly increases to 400A due to a short circuit, and after the wavelet packet transform extracts the features, the decision tree gives a fault probability of 0.88. Petri net reasoning shows that after node 5 tripped, node 6 was not isolated in time, and the rule base prompted an abnormality. After collaborative analysis, the fault was located as node 5, and it was recommended to check the protection device of node 6. This method not only locates the fault, but also finds potential problems. It is understandable that this analysis process is logically progressive from candidate set generation to final positioning. For example, the matrix algorithm first defines the scope, the wavelet packet transform and decision tree focus on the features, the Petri net verifies the logic, and finally outputs the results collaboratively. Assuming that a branch has abnormal current due to lightning strike, the candidate set contains 3 nodes, and node 2 is locked after feature extraction. Logical reasoning confirms that its protection action is reasonable, and finally locates node 2 as the fault point. This multi-step collaboration ensures the credibility of the results.In one possible implementation, consider a node where the current suddenly increases to 300A in the early morning and returns to normal during the day. The matrix algorithm lists it in the candidate set, the wavelet packet transform extracts the mutation features, the decision tree determines it as a fault, and the Petri net combines the upstream and downstream action verification logic to finally locate the node. This multi-faceted analysis enhances accuracy and provides operators with clear fault location information.

[0128] In some embodiments, in the above step S105, the load transfer path of the distribution network is dynamically planned based on the fault location result using a reinforcement learning algorithm to generate an optimal isolation and restoration plan, specifically including:

[0129] The fault location result is used as the state space, the switch operation instruction set is used as the action space, and minimizing the switch cost and load loss is used as the reward function;

[0130] Based on the state space, action space and reward function, the DQN reinforcement learning algorithm is used to make action decisions and output the optimal action sequence;

[0131] According to the optimal action sequence, the switch operation sequence table and the corresponding load transfer path are parsed to generate the optimal isolation and restoration plan.

[0132] In this embodiment, a state space is constructed through the fault location results to obtain all possible state sets. A set of switch operation instructions is generated according to the state space to form an action space and determine the action range. The reward function is calculated by minimizing the switch cost and load loss to obtain the action evaluation basis. Based on the state space, action space and reward function, the DQN algorithm is applied to make action decisions and output the optimal action sequence. The switch operation sequence table is parsed through the optimal action sequence to determine the load transfer path. The optimal isolation plan is generated according to the load transfer path to determine the fault isolation steps. The recovery plan is adjusted through the isolation plan and the action sequence to obtain the load recovery path.

[0133] Exemplarily, using the fault location result as a state space can be understood as taking the identified fault points and their related state information in the distribution network as the basis for decision making. For example, a distribution network has 6 nodes, and node 4 is located as a fault point. The state space contains the fault state of node 4, the switch state of upstream and downstream nodes, and the load distribution. This state space provides a dynamic global view for subsequent decisions. In a possible implementation, the set of switch operation instructions is used as an action space, which is equivalent to defining all possible operation combinations. Specifically, assuming that the distribution network has 10 switches, the action space includes the "on" or "off" instructions of each switch, such as "switch S3 is disconnected" or "switch S7 is closed". The set of these actions provides the algorithm with an operating range to ensure that all possible isolation and restoration paths are covered. It should be noted that the core of minimizing switch cost and load loss as a reward function is to balance the operating cost and power supply reliability. For example, the cost of a switch operation is 50 units, and the load loss is 100 units per kilowatt-hour. The reward function tends to choose a solution with fewer operations and more restored loads. Assuming that disconnecting S4 can isolate the fault but lose 200 kWh of load, while disconnecting S5 only loses 50 kWh, the algorithm will give priority to the latter. This design makes the decision closer to actual needs. Specifically, when using the DQN reinforcement learning algorithm for action decision-making, it can be regarded as an intelligent trial and error process. In a possible implementation, assume that node 4 fails and the load loss is 300 kWh in the initial state. DQN simulates and tries many times, such as disconnecting S2 first and then closing S6, and finds that the load is restored to 80% and the loss is reduced to 60 kWh. The algorithm is continuously optimized and finally outputs the optimal action sequence. This method gradually approaches the optimal solution through learning. Preferably, when parsing the switch operation sequence table and the load transfer path according to the optimal action sequence, it is equivalent to converting the decision into an executable plan. For example, the DQN output sequence is "disconnect S4, close S8", and the corresponding sequence table is: the first step is to disconnect S4 to isolate the fault, and the second step is to close S8 to transfer the load to the backup line. The load transfer path may be from node 1 to node 6. This analysis makes the solution intuitive and operational. It can be understood that the process of generating the optimal isolation and recovery plan is progressive, from state to action to result. For example, node 5 fails due to a short circuit, and the state space shows that its downstream load is 150 kWh. In the action space, an attempt is made to disconnect S5 and close S9. The reward function evaluation cost is 70 units, and the load loss is reduced to 20 kWh. The final solution is "disconnect S5 and close S9", and the load is transferred to another branch. This method ensures the efficiency of fault isolation and the timeliness of power supply restoration. In one possible implementation, suppose a node fails in the early morning, and the state space records a load loss of 100 kWh. Through learning, DQN decides to disconnect S3 first and then close S7. The sequence table shows that the two-step operation completes isolation and recovery. The load transfer path turns from node 2 to node 8.This solution not only quickly isolates the fault, but also maximizes load recovery and provides reliable support for operators.

[0134] Reference Figure 2 An embodiment of the present invention provides a system 2 for quickly locating faults based on three-stage protection and FA collaborative protection, and the system 2 specifically includes:

[0135] The data acquisition and preprocessing module 201 is used to collect the electrical quantity information of each monitoring point in the distribution network in real time, and obtain the topological structure and real-time operation mode of the distribution network by preprocessing and extracting features of the electrical quantity information;

[0136] The three-stage protection module 202 is used to dynamically calculate the protection range sensitivity according to the topological structure and the real-time operation mode, and based on the protection range sensitivity and electrical quantity information, use the preset three-stage action rules to perform detection and judgment to obtain a real-time protection action signal;

[0137] FA fault analysis module 203, used to collect the action time series data of each protection device of the distribution network in real time, build the FA fault feature matrix according to the action time series data, electrical quantity information and topological structure, and use LSTM neural network to analyze the time series correlation of the FA fault feature matrix to generate the FA fault probability distribution diagram;

[0138] The collaborative control module 204 is used to perform collaborative analysis using a fault location algorithm based on the real-time protection action signal and the FA fault probability distribution diagram to obtain a fault location result;

[0139] The fault self-healing module 205 is used to dynamically plan the load transfer path of the distribution network based on the fault location result by using a reinforcement learning algorithm to generate an optimal isolation and restoration plan.

[0140] It is understandable that if Figure 1 The contents of the embodiment of the method for quickly locating faults based on three-stage protection and FA coordinated protection shown in the figure are applicable to the embodiment of the system for quickly locating faults based on three-stage protection and FA coordinated protection. The functions specifically implemented by the embodiment of the system for quickly locating faults based on three-stage protection and FA coordinated protection are the same as those in the embodiment of the method for quickly locating faults based on three-stage protection and FA coordinated protection. Figure 1 The method for quickly locating faults based on three-stage protection and FA coordinated protection is the same as the embodiment shown in FIG. Figure 1 The beneficial effects achieved by the embodiment of the method for quickly locating faults based on three-stage protection and FA coordinated protection shown are also the same.

[0141] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0142] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0143] Reference Figure 3 The embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, a method for quickly locating faults based on three-stage protection and FA collaborative protection as described in any one of the above methods is implemented.

[0144] The computer device 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will appreciate that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0145] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0146] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the computer device 3. Further, the memory 302 may also include both an internal storage unit and an external storage device of the computer device 3. The memory 302 is used to store an operating system, an application program, a boot loader, data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is to be output.

[0147] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for quickly locating faults based on three-stage protection and FA collaborative protection as described in any one of the above methods is implemented.

[0148] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, USB flash drive, mobile hard disk, disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0149] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0150] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0151] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are only schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0152] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. A method for quickly locating faults based on three-stage protection and FA collaborative protection, characterized in that: The method specifically comprises: Collecting electrical quantity information of each monitoring point in the distribution network in real time, and obtaining the topological structure and real-time operation mode of the distribution network by preprocessing and extracting features of the electrical quantity information; The protection range sensitivity is dynamically calculated according to the topological structure and real-time operation mode. Based on the protection range sensitivity and electrical quantity information, the preset three-stage action rules are used for detection and judgment to obtain real-time protection action signals. The action timing data of each protection device of the distribution network is collected in real time, and the FA fault feature matrix is ​​constructed according to the action timing data, electrical quantity information and topological structure. The LSTM neural network is used to analyze the time series correlation of the FA fault feature matrix to generate the FA fault probability distribution diagram; Based on the real-time protection action signal and FA fault probability distribution diagram, the fault location algorithm is used for collaborative analysis to obtain the fault location result; According to the fault location results, the reinforcement learning algorithm is used to dynamically plan the load transfer path of the distribution network and generate the optimal isolation and restoration plan.

2. The method according to claim 1, characterized in that The method of obtaining the topological structure and real-time operation mode of the distribution network by preprocessing and extracting features of the electrical quantity information specifically includes: The electrical quantity information is decomposed and reconstructed by using a wavelet threshold denoising algorithm, the electrical quantity information is data compensated by using a Lagrange interpolation method, and the electrical quantity information of each monitoring point after decomposition, reconstruction and data compensation is time-aligned to obtain standard electrical quantity information; Constructing a voltage and current video feature matrix based on the standard electrical quantity information, calculating node associations on the voltage and current video feature matrix, and generating a multidimensional feature vector set that characterizes the spatiotemporal correlation of electrical quantities; Using the node association threshold and switch state quantity in the multi-dimensional feature vector set, dynamically generating and updating the adjacency matrix, and reconstructing the topological graph model of the distribution network; Based on the topology model, through distributed power flow calculation and overload index analysis, the operation status mapping table and abnormal warning signal of the distribution network are output to obtain the real-time operation mode of the distribution network.

3. The method according to claim 2, characterized in that The protection range sensitivity is dynamically calculated according to the topological structure and the real-time operation mode, and based on the protection range sensitivity and the electrical quantity information, a preset three-stage action rule is used for detection and judgment to obtain a real-time protection action signal, specifically including: Based on the topology and the real-time operation mode, a directional sensitivity coefficient is set for each branch of the topology, wherein the sensitivity coefficient is used to characterize the response strength of the branch to a downstream fault; Constructing a sensitivity matrix according to the directional sensitivity coefficients, performing a Hadamard product operation on the sensitivity matrix and the adjacency matrix, eliminating invalid branches, and obtaining an optimized sensitivity matrix, wherein the sensitivity matrix is ​​used to determine the protection range sensitivity of each branch; Based on the sensitivity matrix and the real-time load current in the electrical quantity information, dynamically generate a three-stage protection criterion set with adaptive threshold and time delay characteristics; By verifying the topological connectivity of the three-stage protection criterion set, a real-time protection action signal for the fault branch is generated.

4. The method according to claim 3, characterized in that The three-stage protection criterion set specifically includes:

5. The method according to claim 1, characterized in that The method of constructing a FA fault feature matrix according to the action timing data, electrical quantity information and topological structure, and using an LSTM neural network to analyze the timing correlation of the FA fault feature matrix to generate a FA fault probability distribution diagram specifically includes: Align and tensor-join the action timing data, electrical quantity information and topological structure to form the FA fault feature matrix; Construct a two-layer bidirectional LSTM network, input the FA fault feature matrix into the two-layer bidirectional LSTM network for time series correlation analysis, and obtain the fault probability value of each coordinate of the distribution network; The fault probability values ​​of each coordinate of the distribution network are output in the form of a heat map to generate a FA fault probability distribution map.

6. The method according to claim 1, characterized in that The method uses a fault location algorithm to perform collaborative analysis based on the real-time protection action signal and the FA fault probability distribution diagram to obtain a fault location result, specifically including: Based on the real-time protection action signal and FA fault probability distribution diagram, an improved matrix algorithm is used to analyze and generate a candidate set of fault areas. Based on the fault area candidate set, transient current features are extracted using wavelet packet transform to construct a decision tree classifier; Based on the fault area candidate set, a Petri net is used to perform temporal reasoning on the protection action logic to obtain a protection action logic rule base; A collaborative analysis is performed based on the decision tree classifier and the protection action logic rule base to generate a fault location result.

7. The method according to claim 2, characterized in that According to the fault location result, the load transfer path of the distribution network is dynamically planned by using a reinforcement learning algorithm to generate an optimal isolation and restoration plan, which specifically includes: The fault location result is used as the state space, the switch operation instruction set is used as the action space, and minimizing the switch cost and load loss is used as the reward function; Based on the state space, action space and reward function, the DQN reinforcement learning algorithm is used to make action decisions and output the optimal action sequence; According to the optimal action sequence, the switch operation sequence table and the corresponding load transfer path are parsed to generate the optimal isolation and restoration plan.

8. A system for quickly locating faults based on three-stage protection and FA collaborative protection, characterized in that: The system specifically comprises: The data acquisition and preprocessing module is used to collect the electrical quantity information of each monitoring point in the distribution network in real time, and obtain the topological structure and real-time operation mode of the distribution network by preprocessing and extracting the characteristics of the electrical quantity information; The three-stage protection module is used to dynamically calculate the protection range sensitivity according to the topological structure and real-time operation mode. Based on the protection range sensitivity and electrical quantity information, it uses the preset three-stage action rules to perform detection and judgment to obtain real-time protection action signals; FA fault analysis module is used to collect the action timing data of each protection device of the distribution network in real time, build the FA fault feature matrix according to the action timing data, electrical quantity information and topological structure, and use the LSTM neural network to analyze the time series correlation of the FA fault feature matrix to generate the FA fault probability distribution diagram; The collaborative control module is used to perform collaborative analysis based on the real-time protection action signal and the FA fault probability distribution diagram using the fault location algorithm to obtain the fault location result; The fault self-healing module is used to dynamically plan the load transfer path of the distribution network based on the fault location results and generate the optimal isolation and recovery plan using the reinforcement learning algorithm.

9. A computer device, characterized in that: include: A memory and a processor and a computer program stored in the memory, when the computer program is executed on the processor, implements the method for quickly locating faults based on three-stage protection and FA collaborative protection as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the method for quickly locating faults based on three-stage protection and FA coordinated protection as described in any one of claims 1 to 7 is implemented.

Images