Switch protection strategy determination method and device, equipment and storage medium

Abstract

The invention relates to a switch protection strategy determination method and device, equipment and a storage medium. The method comprises the steps of obtaining current operation state data of a power distribution network; inputting the current operation state data into the target intelligent agent model to obtain a current protection strategy of the power distribution network; wherein the target agent model is obtained by training based on simulation operation state data of the power distribution network and a power grid knowledge graph of the power distribution network, and the power grid knowledge graph comprises a fault type entity, a fault feature entity, a power grid normal operation entity, a reference feature entity and a protection action entity; and controlling a power grid protection device in the power distribution network to execute a corresponding action according to the current protection strategy. By adopting the method, the accuracy and timeliness of power distribution network protection can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system protection, and in particular to a switch protection strategy determination method and device, equipment and a storage medium. BACKGROUND

[0002] With the rapid development of power systems towards intelligentization and distribution, switch devices (such as low-voltage switches) as key control and protection devices at the end of distribution networks directly affect power supply reliability, equipment safety and user experience. In traditional technology, switch protection strategies are designed based on fixed electrical parameter thresholds (such as rated current and voltage mutation value) and rely on manual on-site operation and maintenance.

[0003] However, this traditional method has been difficult to adapt to the development needs of current power grids. On the one hand, the operating state of the power grid has dynamic change characteristics, and the protection strategy with fixed thresholds cannot be flexibly adjusted according to the actual operating state of the power grid, which is prone to protection misoperation or refusal. SUMMARY

[0004] Therefore, it is necessary to provide a switch protection strategy determination method, device, equipment and storage medium to adapt to the dynamic changes of the power grid and improve the accuracy of distribution network protection.

[0005] In a first aspect, the present application provides a switch protection strategy determination method, comprising:

[0006] obtaining current operating state data of a distribution network;

[0007] inputting the current operating state data into a target intelligent agent model to obtain a current protection strategy of the distribution network; wherein the target intelligent agent model is trained based on simulation operating state data of the distribution network and a power grid knowledge graph of the distribution network, and the power grid knowledge graph includes power grid fault type entities, fault feature entities, power grid normal operation entities, benchmark feature entities and protection action entities;

[0008] controlling a power grid protection device in the distribution network to perform a corresponding action according to the current protection strategy.

[0009] In a second aspect, the present application also provides a switch protection strategy determination device, comprising:

[0010] a first obtaining module configured to obtain current operating state data of a distribution network;

[0011] The policy determination module is configured to input the current operation state data into a target agent model to obtain a current protection policy of the power distribution network, wherein the target agent model is trained based on simulation operation state data of the power distribution network and a power grid knowledge graph of the power distribution network, and the power grid knowledge graph comprises a power grid fault type entity, a fault feature entity, a power grid normal operation entity, a benchmark feature entity, and a protection action entity.

[0012] The action execution module is configured to control a power grid protection device in the power distribution network to perform a corresponding action according to the current protection policy.

[0013] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the method embodiments provided in the first aspect when executing the computer program.

[0014] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the steps of the method embodiments provided in the first aspect are implemented when the computer program is executed by a processor.

[0015] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, and the steps of the method embodiments provided in the first aspect are implemented when the computer program is executed by a processor.

[0016] The above switch protection policy determination method, device, equipment and storage medium obtain current operation state data of a power distribution network, input the current operation state data into a target agent model to obtain a current protection policy of the power distribution network, wherein the target agent model is trained based on simulation operation state data of the power distribution network and a power grid knowledge graph of the power distribution network, and the power grid knowledge graph comprises a power grid fault type entity, a fault feature entity, a power grid normal operation entity, a benchmark feature entity, and a protection action entity, and a power grid protection device in the power distribution network is controlled to perform a corresponding action according to the current protection policy. The target agent model is free from the limitation of a fixed threshold, can flexibly adjust decisions according to real-time operation state data, and the power grid knowledge graph provides rich domain knowledge for the target agent model, thereby guaranteeing the accuracy of decisions and making the output current protection policy more accurate. Therefore, the present application can effectively overcome the defects of a traditional protection policy, realize self-adaptation to dynamic changes of a power grid, improve the accuracy of protection decisions, and guarantee the safe and stable operation of a power distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0018] Figure 1 An application environment diagram of a switch protection strategy determination method provided by an embodiment of the present application is provided.

[0019] Figure 2 A flowchart of a switch protection strategy determination method provided by an embodiment of the present application is provided.

[0020] Figure 3 A flowchart of a method for obtaining current operation state data of a power distribution network provided by an embodiment of the present application is provided.

[0021] Figure 4 A flowchart of a method for determining a verified recognition result provided by an embodiment of the present application is provided.

[0022] Figure 5 A flowchart of a fault type recognition method provided by an embodiment of the present application is provided.

[0023] Figure 6 A flowchart of an agent model training method provided by an embodiment of the present application is provided.

[0024] Figure 7 A flowchart of a method for selecting a target action provided by an embodiment of the present application is provided.

[0025] Figure 8 A structural block diagram of a switch protection strategy determination device provided by an embodiment of the present application is provided.

[0026] Figure 9 An internal structure diagram of a computer device provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0027] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0028] It should be noted that the terms "first", "second", etc. used in the present application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "include" and "have" and any variations thereof used in the present application are intended to cover non-exclusive inclusion. The term "a plurality of" used in the present application refers to two or more. The term "and / or" used in the present application refers to one of the options or any combination of the options.

[0029] The switch protection strategy determination method provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 . In the application environment, the terminal 102 communicates with the server 104 through a network. The data storage system can store data required to be processed by the server 104, such as current operating state data of the power distribution network, etc. The data storage system can be integrated on the server 104, or can be placed on a cloud or other network server. The terminal 102 can be a data acquisition terminal, a monitoring terminal, etc. in the field of the power distribution network, used to acquire the current operating state data of the power distribution network and transmit the data to the server 104. The server 104 can be a background server deployed with a target intelligent agent model, used to receive the data transmitted by the terminal 102, determine a protection strategy through the target intelligent agent model, and issue a control instruction to the terminal 102 or directly to a power grid protection device to control the power grid protection device to perform a corresponding protection action. The power grid protection device includes a circuit breaker, a relay, a backup power supply control device, etc., used to execute the protection strategy issued by the server 104.

[0030] In an exemplary embodiment, as shown in Figure 2 , a switch protection strategy determination method is provided. Taking the server 104 in Figure 1 as an example, the method includes the following steps:

[0031] S201, acquiring current operating state data of a power distribution network.

[0032] The so-called power distribution network refers to a power network that accepts electric energy from a power transmission network or a regional power plant and distributes the electric energy to various users through power distribution facilities on site or step by step. It usually includes high-voltage, medium-voltage and low-voltage distribution lines and related equipment such as substations and switch stations.

[0033] The so-called current operating state data refers to a set of various data that can reflect the operating conditions of the power distribution network during real-time operation, and is basic data for judging the operating state of the power distribution network and formulating a protection strategy.

[0034] Optionally, the current operation state data of the power distribution network can be collected in real time by a data collection system (such as sensors, data monitoring terminals, etc.) of the power distribution network, and the data collection frequency can be set according to the operation characteristics of the power distribution network. The collected current operation state data of the power distribution network can include power grid operation electrical basic data, equipment health state data, protection device operation data, and environmental parameters of switching devices, the power grid operation electrical basic data including voltage amplitude and phase angle of each key connection point, line current, output power of power generation equipment, and electrical parameters (such as input and output current, voltage, and power of switching devices) of switching devices; the equipment health state data including temperature of power grid core equipment (such as transformers) and health state parameters (such as contact temperature, cabinet temperature, insulation resistance value, and mechanical operation number of switching devices) of switching devices; the protection device operation data including whether the power grid protection device has executed an action and the specific time of executing the action; and the environmental correlation data including environmental parameters (such as temperature, humidity, and dust concentration of the environment where the switching devices are located) of switching devices.

[0035] As an optional implementation, the voltage amplitude and phase angle of each key connection point, line current, output power of power generation equipment, temperature of transformers, whether the power grid protection device has executed an action, and the specific time of executing the action can be selected as the current operation state data.

[0036] S202, inputting the current operation state data into the target agent model to obtain a current protection strategy of the power distribution network.

[0037] The target agent model is trained based on simulation operation state data of the power distribution network and a power grid knowledge graph, and the power grid knowledge graph includes power grid fault entities, fault feature entities, power grid normal operation entities, benchmark feature entities, and protection action entities.

[0038] The target agent model refers to an artificial intelligence model with autonomous learning and decision-making capabilities, which can output an adaptive protection strategy according to the input current operation state data after being trained by the simulation operation state data of the power distribution network and the power grid knowledge graph.

[0039] The simulation operation state data refers to operation data generated in a simulation environment by simulating different operation scenarios (including normal operation scenarios and various fault operation scenarios) of the power distribution network, which is used for model training to enable the model to have decision-making capabilities for multiple scenarios.

[0040] The power grid knowledge graph refers to a semantic network that represents knowledge in the power grid field in a structured form, including core entities such as power grid fault entities, fault feature entities, power grid normal operation entities, benchmark feature entities, and protection action entities, and the association relationships between the entities, which provides knowledge support for fault identification and strategy matching.

[0041] The so-called power grid fault type entity refers to an entity concept for defining various types of faults that can occur in the power distribution network, such as line short-circuit faults, transformer faults, and power generation unit faults.

[0042] The so-called fault feature entity refers to an entity concept associated with the power grid fault type entity, used to describe specific feature conditions exhibited by various types of faults, such as current surge features and voltage drop features corresponding to short-circuit faults.

[0043] The so-called power grid normal operation entity refers to an entity concept for representing the normal working state of the power distribution network.

[0044] The so-called reference feature entity refers to an entity concept associated with the power grid normal operation entity, used to describe the reference feature conditions that various parameters should satisfy when the power distribution network is in normal operation.

[0045] The so-called protection action entity refers to an entity concept for defining protection operations that can be taken for various types of power grid states (normal or fault), such as circuit breaker tripping, reclosing, overcurrent protection action current value adjustment, etc.

[0046] Optionally, the collected current operating state data is preprocessed, such as data cleaning, standardization, etc., to ensure that the data format meets the input requirements of the target agent model. The preprocessed current operating state data is converted into a state vector, which is input into the target agent model that has been trained. The target agent model has learned the corresponding relationship between state features and protection strategies in different operating scenarios through a large amount of power distribution network simulation operating state data during the training process, and can accurately identify the operating state (normal or fault, fault type, etc.) of the current power distribution network and output the appropriate current protection strategy, combined with the association relationship of each entity in the power grid knowledge graph.

[0047] As an optional implementation, the current protection strategy can be a combination of multiple actions, including switching operations of switching devices (such as circuit breaker tripping, reclosing operation, and fuse blowing), setting value adjustment of relay protection (such as action current and action time setting of overcurrent protection), and input of backup power or backup device, etc.

[0048] S203, according to the current protection strategy, control the power grid protection device in the power distribution network to perform corresponding actions.

[0049] Among them, the so-called power grid protection device refers to various types of devices in the power distribution network for executing protection strategies, including switching devices (including circuit breakers, disconnectors, load switches, fuses, etc.), relays, and backup power control devices, etc.

[0050] Optionally, according to the current protection strategy output by the target intelligent agent model, corresponding control instructions are generated and sent to the corresponding power grid protection device through the control bus or wireless communication network of the power distribution network to control it to perform corresponding protection actions, such as controlling the circuit breaker to trip to isolate the fault line, adjusting the action current value of the overcurrent protection to adapt to the current load change, etc.

[0051] As an optional implementation, the processes of S201-S203 are applicable to both the determination of the protection strategy in the normal operation state of the power distribution network and the determination of the protection strategy in the fault state of the power distribution network.

[0052] In the above switch protection strategy processing method, the current operation state data of the power distribution network is obtained; the current operation state data is input into the target intelligent agent model to obtain the current protection strategy of the power distribution network; wherein the target intelligent agent model is trained based on the simulation operation state data of the power distribution network and the power grid knowledge graph of the power distribution network, and the power grid knowledge graph includes power grid fault type entities, fault feature entities, power grid normal operation entities, benchmark feature entities and protection action entities; according to the current protection strategy, the power grid protection device in the power distribution network is controlled to perform corresponding actions. The target intelligent agent model is free from the limitation of fixed threshold, and can flexibly adjust the decision according to the real-time operation state data. At the same time, the power grid knowledge graph provides rich domain knowledge for the target intelligent agent model, ensuring the accuracy of the decision, so that the output current protection strategy is more accurate. And the process of determining the current protection strategy by the target intelligent agent does not require human on-site operation and decision-making, greatly improving the response speed of the protection strategy. Therefore, this method can effectively overcome the defects of the traditional protection strategy, realize the self-adaptation to the dynamic changes of the power grid, improve the accuracy and timeliness of the protection decision, and ensure the safe and stable operation of the power distribution network.

[0053] On the basis of the above embodiments, in an exemplary embodiment, the acquisition of the current operation state data of the power distribution network in S201 is further refined. Optionally, as shown in Figure 3 may include the following steps:

[0054] S301, obtaining current device data of the switch device.

[0055] The current device data includes electrical parameters, health state parameters and environmental parameters. The switch device refers to a device used for controlling, protecting and isolating electrical equipment in the power distribution network, including circuit breakers, disconnectors, load switches, fuses and other components, to ensure the safe operation of the power system.

[0056] Optionally, the current device data collected for the switching device in the power distribution network can include electrical parameters (such as input and output current, voltage, power, etc. of the switching device), health status parameters (such as device operating temperature, insulation performance index, mechanical wear degree, etc.), and environmental parameters (such as temperature, humidity, dust concentration, etc. of the environment where the device is located).

[0057] As an optional implementation, when the switching device is a low-voltage switch cabinet, a plurality of high-precision sensors of current, voltage, temperature, and humidity can be arranged in the low-voltage switch cabinet, and the data collected by these sensors can be used as the current device data.

[0058] As another optional implementation, before the server 104 receives the raw data (i.e. the current device data), the raw data can be preliminarily screened, cleaned, and converted in format by the edge server, so as to reduce the data transmission amount, improve the processing efficiency, and use the machine learning algorithm to identify the abnormal data in real time, thereby providing a basis for subsequent analysis.

[0059] The preliminary screening, cleaning, and format conversion of the raw data are performed as follows:

[0060] (1) Data receiving and preliminary transmission: the data stream received from the sensor device is transmitted efficiently by using a message queue, and the received data stream is preliminarily preprocessed to ensure the accuracy and integrity of the data.

[0061] (2) Distributed data cleaning task allocation: the preprocessed data stream is sent to a distributed processing architecture, and the data cleaning tasks are intelligently allocated to corresponding processing nodes according to the characteristics and requirements of the data stream, so as to improve the processing efficiency and flexibility.

[0062] The distributed processing architecture includes a distributed computing framework, a node manager, a distributed file system, and a load balancing algorithm.

[0063] The process of using the distributed processing structure to process the allocated data cleaning tasks specifically includes: managing the processing nodes according to the distributed computing framework; monitoring the state of the processing nodes, allocating tasks, and adjusting the nodes according to the node manager; splitting the data cleaning tasks into a plurality of subtasks, allocating each subtask to a corresponding processing node, and storing the data through the distributed file system; balancing the load according to the processing nodes through the load balancing algorithm; and performing real-time task allocation according to the processing capacity and current load of the processing nodes.

[0064] (3) Real-time anomaly detection and processing: On the processing node, load the anomaly detection model to monitor the incoming data in real time, automatically identify and process the detected abnormal data, and correct and replace the abnormal data according to the type of abnormal data. For abnormal data that cannot be corrected, mark it for subsequent processing or analysis.

[0065] Optionally, the anomaly detection model is a machine learning model, and the specific process of using the machine learning model to identify and mark abnormal data in real time is as follows:

[0066] Collect raw data in real time through sensor devices.

[0067] Preliminary screening and cleaning of raw data to remove redundant, irrelevant or incorrect data, standardization or normalization of data to improve the performance of machine learning algorithms, and extraction of useful features such as statistical features (mean, standard deviation, etc.), time domain features, frequency domain features, etc. for subsequent anomaly detection.

[0068] Select appropriate machine learning algorithms, including unsupervised learning algorithms, supervised learning algorithms and semi-supervised learning algorithms. Unsupervised learning algorithms are suitable for scenarios without labeled data, commonly used algorithms include Isolation Forest, K-means clustering, density-based clustering, etc. These algorithms identify abnormal data by finding the internal structure and pattern of data. Supervised learning algorithms are suitable for scenarios with labeled data, commonly used algorithms include Support Vector Machine (SVM), Random Forest, Neural Network, etc. These algorithms identify abnormal data by learning labeled data (normal data and abnormal data). Semi-supervised learning algorithms combine the characteristics of unsupervised learning and supervised learning, using unlabeled data to discover the structure of the data, and using labeled data to optimize these structures.

[0069] Use the above machine learning algorithms to train the machine learning model using labeled data (if available) or unlabeled data (for unsupervised learning), adjust the hyperparameters of the model to obtain the best performance. Adjust the parameters of the model or select other algorithms for optimization according to the evaluation results, use cross-validation, grid search, etc. to find the optimal model parameters.

[0070] Real-time anomaly detection and labeling: input the real-time collected data into the trained machine learning model, which calculates the anomaly score or probability based on the input data, and judges whether the data is abnormal data according to the set threshold. Mark the identified abnormal data and output it to the designated storage location or send it to the cloud for further analysis. You can set up a real-time monitoring system to discover new abnormal data in a timely manner and take appropriate measures.

[0071] Model evaluation and update: Use the validation set or test set to evaluate the performance of the model, evaluation indicators can include accuracy, recall, F1 value, etc. According to the feedback of practical application and new data, the model is continuously optimized, and the model can be retrained regularly to update the parameters of the model to adapt to the changes of data.

[0072] For example, when a random forest model is selected as an anomaly detection model, the process of real-time monitoring of the incoming data specifically includes:

[0073] Collect a large amount of historical data, carefully analyze these historical data, mark out abnormal data and normal data, and ensure the accuracy of the marking to provide a reliable data basis for subsequent model training.

[0074] Use the labeled abnormal data and normal data to select a random forest algorithm for model training. The random forest algorithm improves the accuracy and stability of the model by building multiple decision trees and combining their output results. During the training process, the parameters of the model are continuously adjusted to optimize its performance. Finally, an anomaly detection model that can accurately identify abnormal data is obtained.

[0075] Deploy the trained anomaly detection model to each processing node to ensure that the anomaly detection model is correctly loaded and run on each processing node to perform real-time anomaly detection on incoming data. In this way, distributed processing and real-time monitoring of data can be achieved, improving response speed and accuracy.

[0076] (4) Incremental data cleaning: Based on the detection results of the data, incremental cleaning is performed on the newly added and changed data. This method can ensure the continuous updating and accuracy of the data, while reducing the workload of repeated processing.

[0077] (5) Streaming data processing and management: Based on the results of data cleaning, use streaming processing technology to efficiently manage data streams, including real-time analysis, storage, distribution, etc. to meet the needs of different application scenarios.

[0078] As an optional implementation, LoRa (Long Range, wireless communication technology) or NB-IoT (Narrowband Internet of Things, low-power wide-area network technology) low-power wide-area network technology can be used to achieve efficient and low-cost transmission between edge servers and servers 104, real-time synchronization, storage and backup of data, and ensure data security and accessibility. According to network conditions and data priority, dynamically adjust the data transmission path to optimize resource utilization.

[0079] Optionally, according to network conditions and data priority, dynamically adjust the data transmission path, specifically including:

[0080] Advanced routing algorithms and protocols are used to dynamically select the optimal transmission path based on real-time network status. Key indicators such as network latency, bandwidth utilization, and packet loss rate are monitored to assess the performance of each path in real-time.

[0081] Data is assigned different priorities based on factors such as importance and urgency. High-priority data (such as real-time video streams and critical business data) is transmitted first to ensure the performance and user experience of critical applications.

[0082] Edge computing nodes or network devices continuously monitor network conditions, including bandwidth, latency, and jitter, and collect and analyze network data in real-time to understand the current state and trends of the network.

[0083] Based on the monitored network conditions and data priorities, intelligent routing algorithms are used to calculate the optimal transmission path, taking into account factors such as bandwidth, latency, packet loss rate, and data priority requirements.

[0084] When the current path is no longer optimal (due to network congestion, device failure, etc.), the system will automatically switch to an alternative path. The switching process should be fast and smooth to ensure the continuity and stability of data transmission.

[0085] During data transmission, the system continuously monitors and optimizes the performance of the transmission path by adjusting parameters such as packet size and transmission rate to further improve data transmission efficiency and reliability.

[0086] S302, input the current device data into the fault identification model to obtain the fault identification result of the power distribution network.

[0087] The fault identification model is a pre-trained artificial intelligence model (such as a neural network model or a support vector machine model) that can identify whether the power distribution network has a fault based on the input switch device current device data and output the fault identification result. The fault identification result is the preliminary judgment result of the power distribution network operation state, including two states: existence of fault and non-existence of fault.

[0088] Optionally, the current device data is input into a pre-trained fault identification model that has been trained with a large amount of historical fault data and normal operation data. The model can output the fault identification result of the power distribution network based on the characteristics of the input data, i.e., existence of fault or non-existence of fault.

[0089] S303, based on the power grid knowledge graph, the fault identification result is verified to obtain the verified identification result.

[0090] The post-verification recognition result refers to a final recognition result obtained by verifying the fault recognition result output by the power grid knowledge graph based on the fault recognition model, and the accuracy is higher than that of the initial fault recognition result.

[0091] Optionally, when the fault recognition result is that there is a fault, the fault feature entity associated with the power grid fault entity can be queried from the power grid knowledge graph. For example, it is queried that the benchmark feature entity corresponding to the power grid normal operation entity in the power grid knowledge graph includes benchmark feature conditions such as electrical parameters and health state parameters, the electrical parameters, health state parameters and the like in the current device data are matched with the queried benchmark feature conditions, and the cosine similarity, Euclidean distance and the like algorithm is used to calculate the matching degree of the benchmark feature conditions and each parameter in the current device data. The first matching degree is calculated by using a weighted algorithm to process these matching degrees. If the first matching degree is greater than or equal to the first threshold, it indicates that the fault recognition result determined by the fault recognition model is accurate, and the current distribution network has a fault; if the first matching degree is less than the first threshold, it indicates that the fault recognition result determined by the fault recognition model has deviation.

[0092] S304, in the case that the post-verification recognition result represents that the distribution network has a fault, the current operation state data of the distribution network is obtained.

[0093] In the case that the post-verification recognition result represents that the distribution network has a fault, the current operation state data of the distribution network in the fault state is further obtained, so as to determine the protection strategy of the distribution network in the fault state according to S201-S203.

[0094] In the embodiment, the current device data of the switch device is obtained first, the preliminary fault recognition is performed by using the fault recognition model, and the recognition result is verified based on the power grid knowledge graph, so as to effectively filter the misrecognition result of the fault recognition model and improve the accuracy of fault recognition.

[0095] On the basis of the above embodiments, in an exemplary embodiment, the determination of the post-verification recognition result in S201 is further refined. Optionally, as shown in Figure 4 The method can include the following steps:

[0096] S401, querying the benchmark feature entity associated with the power grid normal operation entity from the power grid knowledge graph.

[0097] Optionally, the benchmark feature entity associated with the power grid normal operation entity is queried based on the power grid knowledge graph.

[0098] S402, matching the current device data with the benchmark feature conditions of the benchmark feature entity to obtain a first matching degree.

[0099] The reference feature condition refers to an allowable range or a standard value of each parameter in the reference feature entity when the power distribution network is in normal operation. The first matching degree refers to a degree of fit between the current device data and the reference feature condition of the reference feature entity, and is used to determine whether the power distribution network is indeed in a normal operation state.

[0100] Optionally, the operating current, device temperature, voltage and other parameters in the current device data are extracted, and the matching degrees are calculated respectively with the corresponding reference feature conditions, and the first matching degree is calculated by weighted summation of all matching degrees.

[0101] S403, according to the first matching degree and the first preset threshold, the fault recognition result is verified, and the verification result is determined.

[0102] The first preset threshold refers to a critical value for determining whether the first matching degree meets the requirements, which can be determined according to the operation accuracy requirements of the power distribution network, historical fault data statistics and other factors.

[0103] Optionally, the first matching degree and the first preset threshold are compared. If the first matching degree is greater than or equal to the first preset threshold, and the fault recognition result is that there is no fault in the power distribution network, it is determined that the fault recognition result is correct, and the verification result is that the power distribution network is in normal operation. If the first matching degree is less than the first preset threshold, and the fault recognition result is that there is no fault in the power distribution network, it is determined that the fault recognition result is biased, and the verification result is that the power distribution network has a fault. If the first matching degree is less than the first preset threshold, and the fault recognition result is that there is a fault in the power distribution network, it is determined that the fault recognition result is correct, and the verification result is that the power distribution network has a fault. If the first matching degree is greater than or equal to the first preset threshold, and the fault recognition result is that there is a fault in the power distribution network, it is determined that the fault recognition result is biased, and the verification result is that the power distribution network has no fault.

[0104] In the embodiment, by querying the reference feature entity associated with the power grid normal operation entity, the current device data is matched with the reference feature condition to calculate the first matching degree, and the verification result is determined according to the comparison between the first matching degree and the first preset threshold, which can effectively avoid the missed judgment of the fault recognition model.

[0105] On the basis of the above embodiments, in an exemplary embodiment, after obtaining the verification result of S303. Optionally, as shown in Figure 5 The method can include the following steps:

[0106] S501, in the case that the verification result indicates that the power distribution network has a fault, the current fault type of the power distribution network is determined according to the current device data.

[0107] wherein the current fault type refers to a specific fault category of the power distribution network, such as a single-phase ground short-circuit fault, a transformer winding short-circuit fault, a power generation unit overload fault, etc.

[0108] Optionally, in the case where it is determined that the power distribution network has a fault, the current device data can be input into the fault type identification model to obtain the current fault type of the power distribution network. The fault type identification model can be a machine learning model with historical device data as input and the fault type of the power distribution network as label.

[0109] S502, from the power grid knowledge graph, query the fault feature entity associated with the current fault type.

[0110] Optionally, when the verification result is that the power distribution network has a fault, and it is determined that the current fault of the power distribution network is a specific fault type (such as a line short-circuit fault), the fault feature entity corresponding to the fault type is queried from the power grid knowledge graph. This entity can not only include core electrical feature conditions (such as the current threshold of line short-circuit fault, voltage sudden drop range), but also include typical health state features (such as abnormal temperature rise of switch device contact, temporary decrease of insulation resistance) and scene constraint features (such as short-circuit fault often occurs in line load sudden increase period under specific scene such as high load peak period, bad weather) under this fault type.

[0111] S503, match the current device data with the fault feature conditions of the queried fault feature entity to obtain a second matching degree.

[0112] Wherein, the second matching degree refers to the degree of fit between the current device data and the fault feature conditions set by the fault feature entity associated with the current fault type in the power grid knowledge graph.

[0113] Optionally, the current device data is matched with the fault feature entity queried from the knowledge graph in multiple dimensions. The core electrical feature condition, typical health state feature and scene constraint feature of the fault feature entity are matched with electrical parameters (current, voltage) to determine the core electrical feature matching degree, verify whether the current fault type conforms to the basic electrical rule of this type of fault; the typical health state feature is matched with health state parameters (device temperature, insulation resistance) to determine the typical health state feature matching degree, to supplement the verification of whether the fault has caused the corresponding impact on the device, and the scene constraint feature is matched with environmental parameters (such as whether it is a load peak period, weather condition) to determine the scene constraint feature matching degree, to judge whether the current operating scene conforms to the high-incidence scene of this fault type.

[0114] The weighted algorithm is used to process the core electrical feature matching degree, the typical health state feature matching degree and the scene constraint feature matching degree to calculate a second matching degree, wherein the core electrical feature matching degree has the highest weight, the typical health state feature matching degree has the second highest weight, and the scene constraint feature matching degree has the lowest weight.

[0115] S504, determining a fault type recognition result according to the second matching degree and a second preset threshold.

[0116] The second preset threshold refers to a critical value for judging whether the second matching degree meets the requirements, and the second preset threshold can be determined according to the operation accuracy requirements of the power distribution network, historical fault data statistics and other factors.

[0117] Optionally, the second preset threshold is set, and if the second matching degree is greater than or equal to the second preset threshold, it means that the current fault type determination is correct, and the fault type recognition result is the current fault type; if the comprehensive matching degree is less than the threshold, it means that the current fault type determined by the fault type recognition model has deviation, at this time, the fault type recognition result of the power distribution network has a fault but the fault type needs to be re-identified, and the fault type recognition model is triggered to combine the corresponding fault feature conditions of the knowledge graph to re-analyze the current device data to correct the fault type, avoiding the deviation of the protection strategy caused by the single dimension misjudgment.

[0118] In the embodiment, by further determining the fault type in the case that the power distribution network has a fault, and querying the fault feature conditions set by the fault feature entities associated with the current fault type in the power grid knowledge graph, the current device data is matched with the fault feature conditions to calculate the second matching degree, and the fault type recognition result is determined according to the comparison between the second matching degree and the second preset threshold, which can effectively avoid the missed judgment of the fault type recognition model.

[0119] On the basis of the above embodiments, in an exemplary embodiment, as shown in Figure 6 the method can further include the following steps:

[0120] S601, obtaining initial simulation running state data of the power distribution network in a simulation environment.

[0121] The so-called initial simulation running state data refers to the running data generated by simulating an initial running scene (including a normal initial running scene, an initial fault scene, etc.) in the power distribution network simulation environment, which is the initial input data for training the initial agent model. The so-called initial agent model refers to an agent model that has not been fully trained and has a basic decision framework but the model parameters have not been optimized, which is the initial form of the target agent model.

[0122] Optionally, a simulation environment of the power distribution network is constructed, which is simulated based on the topological structure model of the power distribution network. The initial state of the simulation environment is set, such as the normal operation state of the power grid, the rated value of each device parameter, the normal monitoring state of the protection device, etc. The initial simulation running state data is simulated and generated, which can comprehensively reflect the health status and operation performance of the power grid, including but not limited to the voltage amplitude and phase angle of each node (i.e. key connection point), line current, output power of power generation equipment, transformer oil temperature, state of protection device (whether to act, action time, etc.) and the like.

[0123] As an optional implementation, a topological structure model of the power grid is constructed, including power generation units (such as thermal power plants, hydropower stations, wind power plants, etc.), substations, transmission lines, distribution transformers, and various load nodes and the like components. The electrical parameters of each component are determined, such as the rated power of the power generation unit, the voltage level, the impedance characteristic, the resistance, the inductance, the capacitance of the transmission line, the load type (industrial, commercial, residential, etc.) of the load node and its power demand variation law, etc. These parameters will be used for subsequent calculation and simulation of the operation state of the power grid.

[0124] Various fault scenarios that may occur are defined, such as line short circuit fault (single-phase ground fault, two-phase short circuit, three-phase short circuit, etc.), line open circuit fault, transformer fault (internal winding short circuit, insulation breakdown, etc.), power generation unit fault (generator loss of excitation, turbine fault, etc.), etc. The probability distribution of the occurrence of each fault scenario is determined, which can be obtained based on historical fault data statistical analysis. The topological structure model simulates the simulation environment under these fault scenarios.

[0125] S602, based on the initial agent model, determining a target action according to the initial simulation running state data and the power grid knowledge graph.

[0126] Among them, the so-called target action is the protection action selected by the initial agent model from the action space to cope with the current simulation running state according to the initial simulation running state data and the power grid knowledge graph.

[0127] Optionally, a suitable reinforcement learning algorithm is selected as the decision core of the initial agent model, such as Deep Q-Network (DQN) or Proximal Policy Optimization (PPO), etc. If DQN is used, a neural network is constructed including an input layer (receiving the power grid state vector), several hidden layers (for feature extraction and policy learning), and an output layer (outputting the Q-value of each action, i.e. Q-value, which refers to the expected return of taking a certain action in a certain state), and the neural network is trained using experience replay mechanism and target network technology, so that it can accurately estimate the value of each action according to the state of the power grid, and thus select the optimal action. If PPO is used, a policy network and a value network are constructed, and the policy gradient method is used for optimization, which controls the amplitude of policy update while ensuring policy improvement, thereby improving the stability and efficiency of training.

[0128] The model parameters of the initial agent model are initialized, including the weights of the neural network (if it is a neural network-based algorithm), a pre-set buffer, etc.

[0129] The initial simulation running state data is converted into a state vector recognizable by the initial agent model, and the correlation of each entity in the power grid knowledge graph (such as the protection action entity corresponding to the normal operating state) is used to select the target action from the action space by the initial agent model.

[0130] As an optional implementation, the initial simulation running state data is converted into a state vector as the input of the reinforcement learning agent. The state vector corresponding to the initial simulation running data can be represented as:

[0131]

[0132] wherein, and represent the voltage amplitude and phase angle of the i-th node, represents the current of the j-th line, represents the output power of the k-th power generation equipment, represents the temperature of the l-th transformer, represents the state of the m-th power grid protection device.

[0133] The actions that the agent can take, i.e. different protection strategies, are determined. These strategies include switching operations of switching devices (such as tripping and reclosing operations of circuit breakers, fusing of fuses, etc.), setting value adjustment of relay protection (such as setting of action current and action time of overcurrent protection, etc.), and input of backup power or backup equipment, etc. Each possible action is numbered or coded to construct an action space.

[0134] For example, the action space wherein a1 represents a tripping operation of the circuit breaker 1, a2 represents an operation of adjusting the overcurrent protection action current of the line 2 to , and a3 represents an operation of switching on a backup transformer.

[0135] S603, determining a reward value after the target action is performed according to the reward function.

[0136] wherein the so-called reward function refers to a function for evaluating the pros and cons of the target action, and the reward value after the target action is performed is calculated by setting multiple evaluation indexes and assigning corresponding weights, so as to guide the intelligent agent model to learn the optimal protection strategy. The so-called reward value refers to the output result of the reward function, which is a quantitative evaluation of the effect of the target action, and the higher the reward value, the better the target action.

[0137] Optionally, a reasonable reward function is designed to evaluate the pros and cons of the action (protection strategy) taken by the intelligent agent. The reward function should comprehensively consider multiple factors, such as the stability of the power grid (measured by indexes such as node voltage fluctuation and frequency deviation), the range and time of power outage caused by the fault, the damage degree of equipment, the action accuracy of the protection device (whether misoperation or refusal to operate), etc.

[0138] For example, the expression of the reward function is:

[0139]

[0140] wherein represents the stability of the power grid, which can be specifically the average value of the node voltage fluctuation of the power grid; represents the frequency deviation of the power grid; is the power outage time, represents the damage degree of the equipment of the power grid, which is specifically a quantitative index of the damage degree of the equipment, such as the repair cost of the faulty equipment; is the action accuracy index of the protection device, 1 for correct action and 0 for misoperation or refusal to operate, are the corresponding weight coefficients respectively, which are determined according to the importance of each factor.

[0141] As another optional implementation, the reward function can also be determined according to the stability of the power grid, the frequency deviation of the power grid, the action accuracy, the power outage time, the damage degree of the equipment of the power grid, and the matching degree. The matching degree includes the matching degree of the action and the power grid knowledge graph, specifically the matching degree of the protection action and the corresponding entity in the power grid knowledge graph, that is, whether the action conforms to the protection action suggestion or specification in the power grid knowledge graph under the running state.

[0142] S604, simulating the execution of the target action in the simulation environment to obtain the next simulation running state data of the distribution network.

[0143] The next simulation running state data refers to new simulation running state data of the power distribution network after the target action is executed in the simulation environment, reflecting the influence of the target action on the running state of the power distribution network.

[0144] Optionally, after simulating the execution of the target action in the simulation environment, the power distribution network continues to operate normally, and the simulation running state data at this time is collected as the next simulation running state data, which has no significant deviation from the initial simulation running state data.

[0145] At each time step t, the agent selects a target action at from the action space A according to the power grid running state St, for example, the agent observes that the current of a certain line is too large, and may select a target action of adjusting the overcurrent protection setting value of the line or tripping the corresponding circuit breaker, applies the selected target action at to the simulation environment, and the simulation environment simulates the running change of the power grid under the target action according to the pre-set topological structure model and fault scene, to obtain the power grid state St+1 at the next time step and the corresponding reward Rt+1.

[0146] S605, store the state transition data containing the initial simulation running state data, the target action, the reward value and the next simulation running state data to a preset buffer.

[0147] The preset buffer refers to a cache space for storing state transition data, and adopts a first-in-first-out or other data management strategy to ensure that sufficient training data can be obtained when updating the model.

[0148] The state transition information is stored in the preset buffer.

[0149] S606, in the case where the amount of data in the preset buffer reaches a set threshold, batch data is obtained from the preset buffer, and the model parameters of the initial agent are updated according to the batch data.

[0150] The set threshold refers to a preset buffer data amount threshold for triggering the update of the model parameters of the initial agent, which can be determined according to the model training efficiency, data processing capacity and other factors.

[0151] Optionally, when the data in the preset buffer reaches a certain amount, a batch of sample data (i.e. batch data) is randomly extracted from the preset buffer, for the agent based on DQN, the target Q value is calculated using these sample data, and the weights of the neural network are updated by minimizing the mean square error loss function between the target Q value and the current predicted Q value. For the agent based on PPO, the gradients of the policy network and the value network are calculated using the policy gradient method, and the parameters are updated to improve the performance of the policy.

[0152] S607, in the case where the updated initial agent model meets the convergence condition, the updated initial agent model is taken as the target agent model.

[0153] Optionally, in the case where a preset number of iterations or reward value convergence is reached, the updated initial agent model is taken as the target agent model.

[0154] S608, in the case where the updated initial agent model does not meet the convergence condition, the next simulation running state data is taken as new initial simulation running state data, and the operation of determining the target action according to the initial simulation running state data and the power grid knowledge graph is returned to be performed.

[0155] The so-called convergence condition refers to a condition for judging whether the initial agent model is trained, including a preset number of iterations, reward value convergence (reward value fluctuation range of continuous multiple iterations is less than a set value), etc.

[0156] In the case where the preset number of iterations or reward value convergence is not reached, the above environment interaction, data collection and model training update process is repeated, and through a large number of training iterations, the agent gradually learns the protection strategy under different power grid states, i.e., the strategy that can maximize the reward function.

[0157] As an optional implementation, a series of evaluation indexes can be defined to measure the performance of the protection strategy (i.e., the optimal protection strategy) output by the agent model, such as average outage time, average outage frequency, power system average interruption duration index, power shortage expectation, etc. These indexes can comprehensively reflect the ability and reliability level of the power grid in coping with faults under the optimal protection strategy.

[0158] For example, the average outage time refers to the ratio of the sum of the average outage time of each user in the statistical period to the total number of users, which is calculated by counting the outage time under a large number of simulated fault scenarios, and is used to evaluate the power supply continuity of the power grid.

[0159] The optimal strategy is compared and evaluated with the traditional protection strategy or other benchmark strategies. Under the same virtual environment and fault scenario settings, different strategies are applied for simulation running, and the corresponding evaluation index values are calculated. Through comparison, the advantages of the optimal strategy in reducing the outage range, shortening the outage time, reducing the equipment damage risk, etc. are found.

[0160] For example, compared with the traditional fixed value protection strategy, the optimal strategy can dynamically adjust the protection action according to the real-time state of the power grid when coping with complex faults, so that the average outage time is reduced by 30%, the power shortage expectation is reduced by 20%, and the reliability of the power grid is significantly improved.

[0161] In this embodiment, by obtaining initial simulation running state data in a simulation environment, selecting a target action from an action space based on an initial agent model and a power grid knowledge graph, calculating a reward value using a reward function, simulating action execution to obtain next simulation state data, storing state transition data and updating model parameters, until the model meets the convergence condition to obtain a target agent model, so that the target agent model can learn the corresponding relationship between different running states and optimal protection actions from a large number of simulation scenarios. The setting of the reward function guides the model to optimize in the direction of stable power grid, accurate action and minimum loss. The power grid knowledge graph provides domain knowledge support for the model, improving the rationality of the model decision. The trained target agent model has the ability to adapt to different running states and can output the optimal protection strategy, effectively overcoming the defects of traditional protection strategies based on fixed thresholds, lack of adaptability and optimization ability, and providing a more intelligent and efficient protection scheme for the distribution network.

[0162] On the basis of the above embodiments, in one exemplary embodiment, the selection of the target action in S602 is further refined. As shown in Figure 7 may include the following steps:

[0163] S701, input the initial simulation running state data into the initial agent model to obtain the initial action and the probability distribution of the initial action.

[0164] Wherein, the so-called initial action refers to various possible actions output by the initial agent model based on the initial simulation running state data; the so-called probability distribution refers to the probability of selection of each possible action output by the initial agent model, and the higher the probability, the more the model considers that the action is more suitable for the current simulation state.

[0165] Optionally, the initial simulation running state data is obtained, and the initial simulation running state data is standardized to form a simulation state vector. The simulation state vector is input into the initial agent model (such as a deep neural network model), and the initial agent model outputs the probability distribution of the initial action through feature extraction and analysis. Assuming that the action space includes action A (breaker tripping), action B (reclosing), action C (adjusting the overcurrent protection current value), and action D (putting in standby power supply), the initial agent model outputs the probability distribution of the initial action as: action A (0.6), action B (0.1), action C (0.2), and action D (0.1).

[0166] S702, from the power grid knowledge graph, query the protection action matched with the initial simulation running state data.

[0167] Optionally, the operation state represented by the simulation state vector is analyzed to be a specific fault type (such as single-phase ground fault of a transmission line) of the power distribution network, a protection action matched with the fault type is queried from the power grid knowledge graph, and the matched protection actions are obtained as action A (breaker tripping) and action E (breaker tripping and reclosing delay of 2 seconds).

[0168] In S703, the target action is selected from the initial action and the protection action according to the probability distribution of the initial action.

[0169] Optionally, the protection actions (action A and action E) queried from the power grid knowledge graph and the probability distribution of the initial action are combined, and the weighted probability method is used to select the target action. For the protection actions matched in the power grid knowledge graph, the probability of the initial action is multiplied by a weight coefficient (such as 1.5), and the adjusted probability distribution is: action A (0.6x1.5=0.9), action B (0.1), action C (0.2), action D (0.1), and action E (new probability 0.3). The adjusted probability is normalized, and the action A with the highest probability is finally selected as the target action.

[0170] In this embodiment, the probability distribution of the initial action is obtained by analyzing the simulation state vector through the initial intelligent agent model, which ensures the intelligence of the action selection. Then, the matched protection actions are queried in combination with the power grid knowledge graph, the constraints and guidance of the domain knowledge are introduced, and the unreasonable action selection caused by the data deviation of the initial intelligent agent model is avoided. Finally, the target action is selected according to the protection action and the probability distribution of the initial action, which takes into account the autonomous decision of the initial intelligent agent model and the reliability of the domain knowledge. This combined action selection mechanism makes the target action not only conform to the rules learned by the initial intelligent agent model, but also comply with the professional knowledge in the power grid field, thereby improving the accuracy and rationality of the action selection, providing a more reliable basis for the subsequent model training and the formulation of protection strategies in practical applications, and further overcoming the limitations of traditional protection strategies.

[0171] On the basis of the above embodiments, in an exemplary embodiment, the method can further include:

[0172] A deep learning model is constructed to deeply analyze the current device data of the switch device, identify the fault mode of the switch device, and predict the fault development trend. The knowledge graph technology is used to integrate the professional knowledge in the low-voltage switch field and provide auxiliary decision for fault diagnosis. According to the fault severity, different levels of warning thresholds are set to realize graded response.

[0173] According to the fault severity, different levels of warning thresholds are set to realize graded response, which is as follows:

[0174] The warning levels are set, including:

[0175] Primary warning (minor fault): Set a lower threshold to capture early signs of potential failure, such as small fluctuations in current, voltage, or slight changes in temperature, humidity, when the monitoring data reaches or exceeds this threshold, the system issues a minor fault warning, prompting the operation and maintenance personnel to pay attention and prepare for preliminary inspection.

[0176] Secondary warning (moderate fault): Set a higher threshold to identify obvious signs of failure, such as current overload, voltage anomalies, or rapid temperature rise, when the monitoring data reaches or exceeds this threshold, the system issues a moderate fault warning, requiring operation and maintenance personnel to immediately conduct on-site inspection and prepare to take necessary maintenance measures.

[0177] Tertiary warning (serious fault): Set the highest threshold to capture serious fault conditions that may cause equipment damage or system failure, such as short circuits, fires, etc., when the monitoring data reaches or exceeds this threshold, the system issues a serious fault warning and automatically triggers remote control instructions, such as cutting off the fault source, starting emergency response mechanisms, etc., to minimize losses.

[0178] Through the remote communication and data synchronization layer, the warning information is sent to the mobile devices of the operation and maintenance personnel or the control center in real time. After receiving the warning information, the operation and maintenance personnel judge whether immediate action is needed according to the warning level and fault type.

[0179] For primary and secondary warnings, operation and maintenance personnel go to the site for detailed inspection according to fault location and preliminary diagnosis information. For tertiary warnings, operation and maintenance personnel may need to immediately start the emergency response mechanism and handle the situation on site under the premise of ensuring safety.

[0180] According to the inspection results, the operation and maintenance personnel take appropriate maintenance measures, such as replacing damaged parts, adjusting device parameters, etc. During the handling process, the operation and maintenance personnel can maintain real-time communication with the server 104 through the remote control and feedback layer to obtain necessary support and guidance.

[0181] The operation and maintenance personnel feed back the handling results and fault cause analysis to the server 104, and the server 104 optimizes the protection strategy according to these information. Through continuous learning and optimization, the server 104 can gradually improve the accuracy and response speed of the warning, providing more reliable protection for the safe and stable operation of the switch device.

[0182] On the basis of the above embodiments, in an exemplary embodiment, the method can further include:

[0183] Utilizing AR technology, immersive remote operation guidance is provided for operation and maintenance personnel, improving operation accuracy and efficiency, establishing an intelligent learning process based on user feedback, continuously optimizing remote operation processes and interface design, and improving user experience. In emergency situations, automatic triggering of remote control instructions quickly cuts off the source of failure, reducing losses.

[0184] The method provided by the embodiment introduces an Internet of Things sensor network and big data technology to monitor the running state of the switch device in real time, including key parameters such as current, voltage, temperature, and humidity, and uses an artificial intelligence algorithm to intelligently analyze the monitoring data, automatically adjust the threshold of the protection strategy, and dynamically adapt to the operation state of the power grid. This effectively avoids false positives or false negatives caused by improper fixed threshold settings, improving the accuracy and reliability of protection;

[0185] Further, by using big data analysis and machine learning algorithms, the power grid operation data is deeply mined and analyzed to identify fault patterns and predict fault development trends. Through the construction of an intelligent diagnosis model, real-time diagnosis and early warning of power grid faults are realized, providing timely fault information and repair suggestions for operation and maintenance personnel, reducing fault handling time and power grid outage risk.

[0186] Further, by constructing a remote control system and an AR operation and maintenance platform, remote monitoring and operation of the switch device are realized. Operation and maintenance personnel can remotely obtain real-time fault positioning and repair guidance through AR technology, improving operation accuracy and efficiency. At the same time, remote control can also realize automated operation and maintenance, reducing the risk and cost of on-site operation and maintenance;

[0187] Further, by obtaining the current operation state data of the distribution network, inputting the target intelligent agent model trained based on the simulation operation state data and the power grid knowledge graph, and obtaining and executing the current protection strategy, the defects of traditional low-voltage switch protection strategies based on fixed electrical parameter threshold design and relying on manual on-site operation and maintenance can be effectively overcome. In the data acquisition stage, the fault recognition model preliminarily identifies the fault, and double-checks it in combination with the power grid knowledge graph to ensure the accuracy and reliability of the operation state data, providing a solid foundation for the formulation of subsequent protection strategies. The training process of the target intelligent agent model combines a large amount of simulation data and domain knowledge, breaking free from the limitations of fixed thresholds, and can flexibly adjust the protection strategy according to the dynamic operation state of the distribution network. The multi-index design of the reward function guides the model to learn the optimal protection action, taking into account power grid stability, action accuracy, outage time, and other core requirements. In practical applications, this method does not require manual on-site decision-making, has fast response speed, can quickly respond to dynamic changes and complex fault scenarios of the power grid, effectively reduces protection misoperation and refusal, shortens outage time, reduces equipment damage, improves power supply reliability and operation safety of the distribution network, and fully adapts to the current development needs of intelligent and distributed power grids.

[0188] It should be understood that, although each step in the flowchart involved in the above embodiments is shown in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in the above embodiments can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or steps or stages in other steps. It can be understood that the steps in different embodiments can be freely combined as needed, and various non-contradictory schemes formed by the combination are within the scope of protection of the present application.

[0189] Based on the same inventive concept, the embodiments of the present application also provide a switch protection strategy determination device for implementing the above-mentioned switch protection strategy determination method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more switch protection strategy determination device embodiments provided below can refer to the limitations of the switch protection strategy determination method in the above text, which will not be repeated here.

[0190] In one exemplary embodiment, as shown in Figure 8 A switch protection strategy determination device is provided, comprising: a first acquisition module 801, a strategy determination module 802, and an action execution module 803, wherein:

[0191] The first acquisition module 801 is configured to acquire current operating state data of the power distribution network.

[0192] The strategy determination module 802 is configured to input the current operating state data into a target agent model to obtain a current protection strategy of the power distribution network. The target agent model is trained based on simulation operating state data of the power distribution network and a power grid knowledge graph of the power distribution network. The power grid knowledge graph includes power grid fault type entities, fault feature entities, power grid normal operating entities, benchmark feature entities, and protection action entities.

[0193] The action execution module 803 is configured to control a power grid protection device in the power distribution network to perform a corresponding action according to the current protection strategy.

[0194] In one embodiment, the first acquisition module 801 comprises:

[0195] The first acquisition unit is configured to acquire current device data of the switch device.

[0196] The identification unit is configured to input the current device data into the fault identification model to obtain a fault identification result of the power distribution network.

[0197] The verification unit is configured to verify the fault identification result based on the power grid knowledge graph to obtain a verified identification result.

[0198] The second obtaining unit is configured to obtain current operation state data of the power distribution network in a case where the verified identification result indicates that the power distribution network has a fault.

[0199] In one of the embodiments, the verification unit is specifically configured to: query, from the power grid knowledge graph, a reference feature entity associated with a power grid normal operation entity; match the current device data with a reference feature condition of the reference feature entity to obtain a first matching degree; and verify the fault identification result according to the first matching degree and a first preset threshold to determine the verified identification result.

[0200] In one of the embodiments, the apparatus further includes:

[0201] The fault type identification module is configured to determine a current fault type of the power distribution network in a case where the verified identification result indicates that the power distribution network has a fault.

[0202] The query module is configured to query, from the power grid knowledge graph, a fault feature entity associated with the current fault type.

[0203] The matching module is configured to match the current device data with a fault feature condition of the queried fault feature entity to obtain a second matching degree.

[0204] The determination module is configured to determine a fault type identification result according to the second matching degree and a second preset threshold.

[0205] In one of the embodiments, the apparatus further includes:

[0206] The second obtaining module is configured to obtain initial simulation operation state data of the power distribution network in a simulation environment.

[0207] The selection module is configured to determine a target action based on the initial agent model and according to the initial simulation operation state data and the power grid knowledge graph, where an action space includes actions of different power grid protection devices.

[0208] The reward module is configured to determine a reward value after the target action is performed according to a reward function.

[0209] The simulation module is configured to simulate the power distribution network performing the target action in the simulation environment to obtain next simulation operation state data of the power distribution network.

[0210] The storage module is configured to store state transition data including initial simulation running state data, target action, reward value and next simulation running state data to a preset buffer;

[0211] The first updating module is configured to acquire batch data from the preset buffer when the amount of data in the preset buffer reaches a set threshold, and update the model parameters of the initial agent according to the batch data.

[0212] The second updating module is configured to, when the updated initial agent model meets a convergence condition, take the updated initial agent model as the target agent model.

[0213] The returning module is configured to, when the updated initial agent model does not meet the convergence condition, take the next simulation running state data as new initial simulation running state data, and return to perform the operation of determining the target action according to the initial simulation running state data and the power grid knowledge graph.

[0214] In one of the embodiments, the reward function is determined according to power grid stability, power grid frequency deviation, action accuracy, power outage time, power grid equipment damage degree and matching degree, and the matching degree includes the matching degree of the action and the power grid knowledge graph.

[0215] In one of the embodiments, the selecting module is specifically configured to: input the initial simulation running state data into the initial agent model to obtain an initial action and a probability distribution of the initial action; query a protection action matched with the initial simulation running state data from the power grid knowledge graph; and select the target action from the action space according to the initial action probability distribution.

[0216] Each of the above modules of the switch protection strategy determination apparatus can be realized by software, hardware and a combination thereof in whole or in part. Each of the above modules can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform the operations corresponding to each of the above modules.

[0217] In one exemplary embodiment, a computer device, which can be a server, is provided, and an internal structure diagram of the computer device can be as shown in Figure 9As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the protection strategy. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with the terminal outside through the network connection. The computer program is executed by the processor to realize a switch protection strategy determination method.

[0218] Those skilled in the art can understand that, Figure 9 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0219] In an exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to realize the steps in each of the method embodiments described above.

[0220] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to realize the steps in each of the method embodiments described above.

[0221] In an exemplary embodiment, a computer program product is provided, including a computer program, and the computer program is executed by the processor to realize the steps in each of the method embodiments described above.

[0222] It should be noted that the data (including but not limited to data for analysis, stored data, displayed data, etc., such as current running state data) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0223] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0224] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0225] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method for determining a switch protection strategy, characterized in that, The method includes: Obtain the current operating status data of the power distribution network; The current operating status data is input into the target intelligent agent model to obtain the current protection strategy of the distribution network; wherein, the target intelligent agent model is trained based on the simulated operating status data of the distribution network and the power grid knowledge graph of the distribution network, and the power grid knowledge graph includes fault type entities, fault feature entities, normal operation of the power grid entities, baseline feature entities, and protection action entities; According to the current protection strategy, control the power grid protection devices in the distribution network to perform corresponding actions.

2. The method according to claim 1, characterized in that, The acquisition of the current operating status data of the distribution network includes: Obtain the current device data of the switching equipment; The current equipment data is input into the fault identification model to obtain the fault identification result of the power distribution network; Based on the power grid knowledge graph, the fault identification result is verified to obtain the verified identification result; If the verification result indicates that there is a fault in the distribution network, the current operating status data of the distribution network is obtained.

3. The method according to claim 2, characterized in that, The step of verifying the fault identification result based on the power grid knowledge graph to obtain a verified identification result includes: From the power grid knowledge graph, query the baseline feature entities associated with the entities operating normally in the power grid; The current device data is matched with the baseline feature conditions of the baseline feature entity to obtain a first matching degree; The fault identification result is verified based on the first matching degree and the first preset threshold to determine the identification result after verification.

4. The method according to claim 2, characterized in that, After verifying the fault identification result based on the power grid knowledge graph to obtain the verified identification result, the method further includes: If the verification result indicates that there is a fault in the distribution network, the current fault type of the distribution network is determined based on the current equipment data. From the power grid knowledge graph, query the fault feature entities associated with the current fault type; The current device data is matched with the fault characteristic conditions of the queried fault characteristic entities to obtain a second matching degree; The fault type identification result is determined based on the second matching degree and the second preset threshold.

5. The method according to claim 1, characterized in that, The method further includes: Obtain the initial simulation operation status data of the power distribution network in the simulation environment; Based on the initial intelligent agent model, the target action is determined according to the initial simulation operation state data and the power grid knowledge graph; The reward value after performing the target action is determined based on the reward function; In the simulation environment, the distribution network is simulated to perform the target action in order to obtain the next simulation operation status data of the distribution network; The state transition data, which includes the initial simulation running state data, the target action, the reward value, and the next simulation running state data, are stored in a preset buffer. When the amount of data in the preset buffer reaches a set threshold, batch data is obtained from the preset buffer, and the model parameters of the initial agent are updated according to the batch data; If the updated initial agent model satisfies the convergence condition, the updated initial agent model shall be used as the target agent model. If the updated initial agent model does not meet the convergence condition, the next simulation running state data is used as the new initial simulation running state data, and the operation of determining the target action based on the initial simulation running state data and the power grid knowledge graph is returned.

6. The method according to claim 5, characterized in that, The reward function is determined based on grid stability, grid frequency deviation, action accuracy, power outage time, grid equipment damage level, and matching degree, where the matching degree includes the degree of matching between the action and the grid knowledge graph.

7. The method according to claim 5, characterized in that, The determination of the target action based on the initial intelligent agent model, according to the initial simulation running state data and the power grid knowledge graph, includes: The initial simulation running state data is input into the initial intelligent agent model to obtain the initial action and the probability distribution of the initial action; From the power grid knowledge graph, query the protection actions that match the initial simulation operation status data; Based on the probability distribution of the initial action, a target action is selected from the initial action and the protection action.

8. A device for determining a switch protection strategy, characterized in that, The device includes: The first acquisition module is used to acquire the current operating status data of the power distribution network; The strategy determination module is used to input the current operating status data into the target intelligent agent model to obtain the current protection strategy of the distribution network; wherein, the target intelligent agent model is trained based on the simulated operating status data of the distribution network and the power grid knowledge graph of the distribution network, and the power grid knowledge graph includes power grid fault type entities, fault feature entities, power grid normal operation entities, baseline feature entities, and protection action entities; The action execution module is used to control the power grid protection devices in the distribution network to perform corresponding actions according to the current protection strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

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

Images