Power distribution network fault location method based on artificial intelligence and storage medium

Abstract

The application provides a power distribution network fault positioning method based on artificial intelligence and a storage medium. The method comprises the following steps: determining electrical parameters based on data collected by electrical energy sensors in a multi-level power distribution network, the electrical energy sensors being installed on each level branch and node, the electrical parameters including current and voltage; obtaining preset working parameters, the preset working parameters being preset working parameters for maintaining stable operation of the multi-level power distribution network; determining characteristic values of each level according to the electrical parameters, and identifying the fault type of the fault point according to the deviation of the electrical parameters from the preset working parameters, in combination with the power grid topology relationship and the fault identification model; determining a fault positioning method according to the fault type and the level information, and positioning the fault point, the fault positioning method comprising coarse positioning based on impedance method and fine positioning based on traveling wave method. The method can optimize real-time communication, minimize the transmission delay between fault data and control commands, and respond quickly when a fault occurs.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, in particular to a power distribution network fault positioning method based on artificial intelligence and a computer readable storage medium. BACKGROUND

[0002] Under the fine multi-level division system of the modern power distribution network, the current and voltage data monitoring and collection work of each level faces its own challenges, especially the precision and sampling frequency of data collection are significantly different between different levels. This inconsistency directly maps to the fault diagnosis stage, resulting in a large difference in the accuracy of analysis and the reliability of conclusions when different levels make fault judgments. The diversity of electrical configuration, line parameters and network topology structure of each level makes the manifestation and characteristic signals of the same type of fault in different level networks each have its own characteristics, greatly increasing the difficulty and complexity of fault type identification. And the existing fault positioning algorithm is usually designed based on a specific network topology and fault type, which cannot adapt to the flexibility and robustness of different levels and fault scenarios. SUMMARY

[0003] The main purpose of the present application is to provide a power distribution network fault positioning method based on artificial intelligence and a computer readable storage medium, to at least solve the problem that the existing power grid fault positioning algorithm can only be used for specific network topology and fault type, and cannot accurately identify the fault type of different levels and different fault scenarios.

[0004] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a power distribution network fault positioning method based on artificial intelligence is provided, comprising: determining electrical parameters of a multi-level power distribution network based on data collected by electrical energy sensors in the multi-level power distribution network, the electrical energy sensors being installed on branch lines and nodes of each level, the electrical parameters including current and voltage; obtaining preset working parameters of the multi-level power distribution network, the preset working parameters being preset working parameters for maintaining stable operation of the multi-level power distribution network; determining characteristic values of each level of the multi-level power distribution network according to the electrical parameters, and identifying the fault type of the fault point according to the deviation of the electrical parameters from the preset working parameters, in combination with the power grid topology relationship and the fault identification model of the multi-level power distribution network; determining a fault positioning method according to the fault type of the fault point and the level information of the multi-level power distribution network, and positioning the fault point using the fault positioning method, the fault positioning method including coarse positioning based on impedance method and fine positioning based on traveling wave method.

[0005] According to another aspect of the present application, a computer readable storage medium is provided, which comprises a stored program, wherein the computer readable storage medium controls the device where the computer readable storage medium is located to execute any one of the artificial intelligence-based power distribution network fault locating methods when the program is running.

[0006] With the technical solution of the present application, the above-mentioned artificial intelligence-based power distribution network fault locating method first determines the electrical parameters of the multi-level power distribution network based on the data collected by the power sensors in the multi-level power distribution network. The power sensors are installed on each level branch and node, and the electrical parameters include current and voltage. Then, the preset working parameters of the multi-level power distribution network are obtained, which are the preset working parameters for maintaining the stable operation of the multi-level power distribution network. Then, according to the electrical parameters, the characteristic values of each level of the multi-level power distribution network are determined, and according to the deviation of the electrical parameters from the preset working parameters, the fault type of the fault point is identified in combination with the power grid topology relationship and the fault identification model of the multi-level power distribution network. Finally, according to the fault type of the fault point and the level information of the multi-level power distribution network, the fault locating method is determined, and the fault locating method is used to locate the fault point. The fault locating method includes coarse positioning based on impedance method and precise positioning based on traveling wave method. This method can optimize real-time communication, minimize the transmission delay between fault data and control commands, so as to respond quickly when a fault occurs, and solve the problem that the power grid fault locating algorithm in the prior art can only be used for specific network topology and fault type, and cannot accurately identify the fault type of different levels and different fault scenes. BRIEF DESCRIPTION OF DRAWINGS

[0007] The drawings accompanying the specification of the present application serve to provide a further understanding of the present application, and the illustrative embodiments of the present application and their descriptions serve to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:

[0008] Figure 1 A flowchart of an artificial intelligence-based power distribution network fault locating method according to an embodiment of the present application is shown;

[0009] Figure 2 A flowchart of locating a fault position according to an embodiment of the present application is shown;

[0010] Figure 3 A flowchart of configuring an isolation switch operation according to an embodiment of the present application is shown;

[0011] Figure 4 A flowchart of another artificial intelligence-based power distribution network fault locating method according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0012] It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other in the case of no conflict. The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0013] In order for those skilled in the technical field to better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0014] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0015] As introduced in the background, the fault location algorithm in the prior art is usually designed based on a specific network topology and fault type, and cannot adapt to the flexibility and robustness of different levels and fault scenarios. In order to solve the problem that the power grid fault location algorithm in the prior art can only be used for a specific network topology and fault type, and cannot accurately identify the fault type of different levels and different fault scenarios, the embodiments of the present application provide a power distribution network fault location method based on artificial intelligence and a computer readable storage medium.

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application.

[0017] In the present embodiment, a power distribution network fault location method based on artificial intelligence running on a mobile terminal, a computer terminal or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0018] Figure 1 is a flowchart of the power distribution network fault location method based on artificial intelligence according to the embodiments of the present application. AsFigure 1 As shown, the method includes the following steps:

[0019] Step S201: Based on the data collected by the power sensors in the multi-level distribution network, determine the electrical parameters of the multi-level distribution network. The power sensors are installed on branches and nodes at each level. The electrical parameters include current and voltage.

[0020] Specifically, based on the topology of the multi-level distribution network, voltage and current sensors are rationally arranged to cover branches and nodes at each level. This method can obtain accurate and reliable monitoring data, ensuring the comprehensiveness and representativeness of the monitoring points.

[0021] The determination of electrical parameters of the multi-level distribution network based on data collected by power sensors in the multi-level distribution network includes the following steps:

[0022] Step S301: Based on the topology of the multi-level distribution network, determine the installation location of the power sensor. The installation location of the power sensor covers the branches and nodes of each level of the multi-level distribution network. The power sensor includes a voltage sensor and a current sensor.

[0023] Among them, an abnormal situation identification model is established, and abnormal situations in the operation of the distribution network are discovered through real-time analysis of electrical parameters;

[0024] By combining the distribution network topology and fault feature database, the changes in electrical parameters at each monitoring point under abnormal conditions are analyzed to locate the fault location.

[0025] Step S302: Establish a unified data acquisition and transmission protocol, and based on the above data acquisition and transmission protocol, summarize and synchronize the data collected by the above power sensors at each monitoring point to obtain the initial power data;

[0026] In some instances, by establishing unified data acquisition and transmission protocols, such as Modbus and IEC61850, sensor data distributed across various monitoring points can be aggregated and synchronized, enabling the systematic integration and centralized management of electrical parameters.

[0027] Step S303: Filter, denoise, and extract features from the initial power data to obtain the electrical parameters of the multi-level distribution network.

[0028] Specifically, the data collected by the sensors is preliminarily processed and summarized, and the data is synchronized in time and converted in format. This method can obtain accurate and reliable monitoring data, laying the foundation for subsequent analysis.

[0029] like Figure 2 As shown, Figure 2A flowchart for locating fault position is shown according to an embodiment of the application; first, arrange voltage and current sensors, cover each level branch and node, establish data collection and transmission protocol, then summarize and synchronize sensor data, filter and denoise voltage and current data, finally extract data features, establish abnormal situation identification model, and locate fault position.

[0030] In some examples, for specific interference and noise types of the power distribution network, such as electromagnetic interference and harmonic interference, the collected voltage and current data are filtered, denoised and feature extracted, key electrical parameter features such as voltage waveform distortion and current harmonic content are extracted, data quality is improved, and reliable input is provided for abnormal situation identification and fault location.

[0031] Based on machine learning algorithms such as support vector machine SVM and random forest and pattern recognition techniques, an abnormal situation identification model is established. Through real-time analysis of electrical parameters, key features are extracted and compared with normal operation mode, abnormal situations in power distribution network operation are discovered in time, and intelligent level of monitoring system is improved. Combined with power distribution network topology and fault feature library, fault location algorithm is used. When abnormal situation is monitored, real-time monitoring data is matched and analyzed according to typical fault mode in fault feature library, and fault position is inferred and located combined with power distribution network topology. By analyzing the trend of electrical parameters of monitoring points around the fault point, the fault position is accurately located, and the fault handling time is shortened. When arranging voltage and current sensors, centralized and distributed combination can be used.

[0032] In the data preprocessing stage, wavelet transform and Fourier transform are combined to filter and denoise voltage and current signals, and key features such as fundamental amplitude, phase angle and harmonic content are extracted. Then, support vector machine SVM algorithm is used to establish abnormal situation identification model, and optimal decision boundary is obtained by training historical operation data. When real-time monitoring data deviates from normal operation mode, it is determined as abnormal situation. For fault location, first, according to the topology of power distribution network, a fault feature library is constructed, which contains voltage and current feature modes of various typical faults. When fault occurs, dynamic time warping algorithm is used to align and compare voltage and current waveforms before and after fault, and fault feature sequence is extracted. Then, convolutional neural network algorithm is used to classify and identify fault feature sequence, and determine fault type. Combined with the topology of power distribution network, the position coordinates of fault point are calculated by combining impedance method and traveling wave method, and the accuracy can reach within 50 meters. In terms of data storage, time series database InfluxDB is used to store monitoring data every 5 minutes, and Kafka message queue is used for data backup and distribution. At the same time, role-based access control RBAC mechanism is used to manage the access and operation permissions of monitoring data, and ensure data security.

[0033] On the main line of the distribution network, a centralized monitoring unit is installed every 500 meters, each unit containing 3 voltage sensors and 3 current sensors to measure three-phase voltage and current. On branch lines and user terminals, a distributed monitoring method is used, with an intelligent micro-monitoring unit installed every 200 meters, containing high-precision voltage and current sensors. The sampling frequency of the sensor can be set to 5 kHz, with a measurement range of 0-500 V and 0-100 A, and an accuracy of 0.1 level. Through the Modbus communication protocol, the collected data is transmitted in real time to the monitoring center, and time synchronization and format conversion are performed.

[0034] Step S202, obtaining the preset working parameters of the multi-level distribution network, the preset working parameters being the preset working parameters for maintaining stable operation of the multi-level distribution network;

[0035] Specifically, for the preprocessed data, feature extraction algorithms such as frequency domain feature extraction based on Fast Fourier Transform (FFT) and time-frequency domain feature extraction based on wavelet transform are used to extract key feature values reflecting the operation state of the distribution network from the current and voltage data. The feature values are compared with the preset normal operation parameters of the power grid. This method can improve the performance and efficiency of the system and ensure stable operation of the system by designing and optimizing the multi-level distribution network according to the preset working parameters.

[0036] Among them, according to the topology structure and load distribution of each level of the distribution network, an anomaly detection model is established, and by training historical operation data, the normal operation mode of the power grid is learned. When the real-time feature values deviate from the normal mode, it is judged as a potential abnormal situation;

[0037] In some examples, for the preprocessed data, feature extraction algorithms such as frequency domain feature extraction based on Fast Fourier Transform (FFT) and time-frequency domain feature extraction based on wavelet transform are used to extract key feature values reflecting the operation state of the distribution network from the current and voltage data. Frequency domain features can include fundamental amplitude, harmonic amplitude, and total harmonic distortion, etc., and time-frequency domain features can include wavelet energy distribution, singular value, etc. The extracted feature values are compared with the preset normal operation parameters of the power grid to identify potential abnormal situations. The detected abnormal data is clustered, classified and backtracked to determine the time, location and impact range of the abnormality.

[0038] The abnormal data is clustered using a density-based clustering algorithm such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), to identify different types of abnormal patterns. Classification models such as SVM (Support Vector Machine) and random forest are used to classify abnormal data and determine the type of anomaly. Through the analysis of time series data, the propagation path and impact range of abnormal signals are tracked, providing a basis for subsequent fault diagnosis. According to the results of abnormal situation analysis, combined with expert knowledge and fault case library, a fault diagnosis model is constructed. A decision tree-based fault diagnosis method is used to train historical fault data and generate a fault diagnosis decision tree. The input features of the decision tree include abnormal feature values, power grid topology, device parameters, etc., and the output result is the fault type and location.

[0039] At the same time, the method of case-based reasoning is used to match the current abnormal situation with historical fault cases, find the most similar case, and infer the cause and disposal measures of the current fault according to the diagnosis result of the case. The collected data, extracted feature values, abnormal analysis results and diagnosis results, etc. Information is stored and managed. Distributed time series databases such as InfluxDB or OpenTSDB are used to design reasonable data table structures and tag systems to support efficient data writing and querying. For structured metadata and diagnosis results, they can be stored in relational databases such as MySQL or PostgreSQL. By establishing appropriate indexes and query optimization strategies, the performance of data queries is improved. Using data visualization technologies such as Grafana and other open source tools, the running state of the distribution network, abnormal situations and diagnosis results, etc. Information is visually presented. Through the design of intuitive dashboards and charts, real-time monitoring and trend analysis of key indicators such as voltage, current and power are achieved.

[0040] According to the results of abnormality detection and fault diagnosis, set reasonable alarm thresholds and notification strategies, when the indicators exceed the threshold range or a specific abnormal pattern occurs, timely alarm to the on-duty personnel, notify the relevant personnel through SMS, email and other ways, improve the response speed and processing efficiency of the system. Regularly evaluate the effectiveness of the alarm strategy, optimize the threshold setting and notification method, reduce the occurrence of false positives and false negatives.

[0041] At each node of the power distribution network, current and voltage sensors such as Hall current sensors and resistance voltage divider voltage sensors are deployed, with a sampling frequency set to 10 kHz, collecting 10,000 data points per second, and transmitting data in real time to the edge computing gateway through the ModBus communication protocol. The gateway uses Kalman filtering algorithm to denoise the data, filters out high-frequency noise, and then forwards the data to the cloud server at a frequency of one data packet per second. The cloud server uses ApacheSpark streaming processing framework to extract features in real time, uses FFT algorithm to extract fundamental wave amplitude and phase angle of voltage and current, and uses wavelet transform to extract transient features such as voltage sag and sudden rise. The extracted feature data is stored in the time series database InfluxDB and compared with the preset normal working parameter threshold. If the feature value is found to be outside the normal range, an abnormal alarm is triggered and the abnormal data is transmitted to the anomaly analysis module.

[0042] The anomaly analysis module uses DBSCAN clustering algorithm based on density to cluster and analyze abnormal data, identifies different types of abnormal patterns such as single-phase grounding and two-phase short circuit, and stores the clustering results in the MongoDB database. The clustering results are input into the fault diagnosis model, which uses decision tree algorithm to infer the fault type and location based on clustering results, power grid topology, historical fault cases, etc. The diagnosis results are stored in the MySQL database in JSON format and trigger an alarm, which notifies the on-duty personnel through SMS, email, etc. Use Grafana visualization tool to design power distribution network operation status monitoring dashboard, real-time display of voltage, current, power and other key indicators of each node, and display the trend of indicators in the form of curve chart. When an abnormality or fault occurs, the dashboard highlights the abnormal nodes and faulty equipment in red and pops up the abnormal analysis and fault diagnosis report to assist the on-duty personnel in quickly locating and handling the problem.

[0043] Step S203, according to the above electrical parameters, determine the characteristic value of each level of the multi-level power distribution network, and according to the deviation of the above electrical parameters from the above preset working parameters, combine the power grid topology relationship of the above multi-level power distribution network and the fault recognition model, identify the fault type of the fault point;

[0044] Specifically, for the detected abnormal situation, combine expert knowledge and fault case library to build fault diagnosis rules, analyze and diagnose the abnormal situation through rule reasoning engine to determine the fault type and location; improve the response speed and processing efficiency of the system.

[0045] The characteristic values of each level of the multi-level power distribution network are determined according to the electrical parameters, and the fault type of the fault point is identified according to the deviation of the electrical parameters from the preset working parameters, in combination with the power grid topology relationship and the fault identification model of the multi-level power distribution network, including the following steps:

[0046] In step S401, the characteristic values reflecting the operating state of the multi-level power distribution network are extracted from the electrical parameters, and the characteristic values at least include voltage deviation and current harmonic content.

[0047] In step S402, the target deviation value of the electrical parameters of each detection point from the preset working parameters is calculated by using the Euclidean distance or Manhattan distance measurement method, the preset working parameters at least include rated voltage, rated current and power factor, and the target deviation value at least includes current deviation value and voltage deviation value.

[0048] In step S403, the power grid topology relationship of the multi-level power distribution network is stored in the graph database in the form of an adjacency matrix, wherein each node represents a bus or a switch, each edge represents the connection relationship between two nodes, and the properties of the edge include the electrical parameters of the conductor, and the electrical parameters of the conductor at least include resistance and reactance.

[0049] In step S404, the decision tree algorithm is used to construct the fault identification model, the power grid topology relationship and the target deviation value of each detection point are input into the fault identification model, and the fault type of the fault point is obtained, and the fault type includes one of overload, short circuit, open circuit and grounding.

[0050] Specifically, the operating state and abnormal situation of the power distribution network and the diagnostic result information are presented through a curve diagram or a heat map, and an alarm threshold is set, and when the characteristic value exceeds the threshold range, an alarm is sent to the on-duty personnel. This method improves the response speed and processing efficiency of the system.

[0051] In some examples, according to the normal working parameters of the power grid, such as rated voltage, rated current, power factor, etc., the deviation between the current voltage data and the normal value is calculated, the deviation calculation adopts Mahalanobis distance or cosine similarity measurement method, and different attributes are normalized to eliminate the dimension influence, and the deviation value of each measurement point is obtained. The power grid topology relationship is stored in the graph database in the form of an adjacency table or a compressed sparse matrix, such as Neo4j, each node represents a bus or a switch, each edge represents the connection relationship between nodes, and the properties of the edge include the electrical parameters of the conductor, such as resistance, reactance, etc., reducing memory occupation and calculation overhead.

[0052] The CART decision tree algorithm is used to construct a fault diagnosis model. Through feature selection, tree generation and pruning, the Gini coefficient or information gain ratio is used as the division criterion, and the current and voltage deviation values of each measuring point, the power grid topology structure, historical fault cases, etc. are used as input features, and the output result is the fault type, such as overload, short circuit, open circuit, grounding, etc. For different fault types, the corresponding fault features are extracted, including current duration exceeding rated value, temperature rise, etc. for overload fault features, current surge, voltage drop, etc. for short circuit fault features, current being zero, voltage rising, etc. for open circuit fault features, and zero sequence current increasing, zero sequence voltage rising, etc. for grounding fault features. Wavelet transform and other time-frequency domain analysis methods are used for feature extraction, such as using Daubechies wavelet to decompose the current and voltage signals by 5 layers, extracting the energy features of each frequency band, and splicing the feature vectors of multiple measuring points into a high-dimensional feature matrix. Random forest algorithm is used to integrate multiple decision tree models, and multiple decision trees are generated through bootstrap sampling and attribute randomization, and integrated prediction is performed through voting or averaging to improve the accuracy and robustness of fault diagnosis.

[0053] At the same time, other types of machine learning models such as support vector machines and neural networks are introduced to further improve the performance of fault diagnosis through model fusion. Cross-validation and other methods are used to evaluate and optimize the model, and a complete model updating mechanism is established to continuously update the model parameters to adapt to the dynamic changes of power grid operating conditions. After identifying the fault type, the spatial distribution information of the fault measuring points and the power grid topology relationship are used to infer the location or range of the fault through graph theory algorithms such as shortest path and connectivity analysis, to generate fault reports and maintenance suggestions, and to provide decision support for maintenance personnel, improving the efficiency and accuracy of fault handling.

[0054] Current and voltage sensors, such as Hall current sensors and voltage transformers, are installed at the main nodes of the power distribution network, with a sampling frequency of 10 kHz. The data is transmitted in real time to the edge gateway through the RS485 bus. The gateway uses an adaptive Kalman filter algorithm to denoise the data, filtering out the 50 Hz power frequency and its harmonic interference. Then, the current and voltage data are normalized using the maximum absolute value normalization method, mapping the data to the [-1, 1] interval. The normalized data are compared with the pre-set normal operating threshold. The Mahalanobis distance is used to calculate the distance between the data points and the threshold center. If the distance exceeds 3 standard deviations, it is determined as abnormal data. The current and voltage values at that moment are extracted as abnormal features. The abnormal feature data are sent to the cloud server in JSON format through the MQTT protocol and stored in the time series database InfluxDB. The topology structure of the power distribution network, including the connection relationship and electrical parameters of busbars, switches, transformers, etc., is stored in the Neo4j graph database. The abnormal feature data and topology structure data are input into the pre-trained CART decision tree model for fault diagnosis. The decision tree divides the features by the information gain ratio criterion, generating a binary tree with a depth of 5. The leaf nodes of the tree correspond to different fault types, such as single-phase grounding and two-phase short circuit.

[0055] For the diagnosed fault type, the typical waveform and wavelet energy distribution features of this type of fault are extracted by querying the fault feature library. The Daubechies4 wavelet is used to decompose the fault waveform into 4 layers, and the root mean square value of each scale coefficient is extracted as the wavelet energy feature. The wavelet energy feature is matched with the current fault waveform feature to calculate the cosine similarity. If the similarity is greater than 0.8, the fault type is confirmed. Otherwise, the data sample is added to the training set, and the decision tree model is retrained. According to the fault type and topology structure, the Dijkstra shortest path algorithm is used to calculate the electrical distance from the fault point to each monitoring point. The location with the smallest electrical distance and consistent with the fault feature is selected as the fault point. The fault report and repair plan are generated and pushed to the mobile terminal of the maintenance personnel through Web service, guiding the on-site fault handling work.

[0056] Step S204, according to the fault type of the above fault point and the hierarchical information of the above multi-level power distribution network, determine the fault location method, and use the above fault location method to locate the above fault point. The fault location method includes coarse positioning based on impedance method and precise positioning based on traveling wave method.

[0057] Specifically, according to the fault type information obtained by the fault diagnosis module, such as single-phase grounding, two-phase short circuit, etc., combined with the hierarchical structure of the power grid, such as the main line, branch line, etc., the appropriate fault location technology is adaptively selected to improve the accuracy and reliability of the traveling wave parameter measurement. Through simulation test and field test, the positioning algorithm is optimized and verified to ensure its performance and stability under actual working conditions.

[0058] Among them, for different fault types, time-frequency domain analysis method is used to extract corresponding fault features.

[0059] Among them, according to the fault type of the fault point and the hierarchical information of the multi-level distribution network, the fault location method is determined, and the fault point is located by using the fault location method, including the following steps:

[0060] Step S501, according to the fault type of the fault point and the hierarchical information of the multi-level distribution network, the coarse positioning method based on impedance method is used to calculate the first position information of the fault point on the main line of the multi-level distribution network, and the fine positioning method based on traveling wave method is used to calculate the second position information of the fault point on the branch line of the multi-level distribution network;

[0061] Step S502, on the main line of the multi-level distribution network, according to the voltage data and current data at the fault time, the equivalent impedance between the fault point and the detection point is calculated, and according to the electrical parameters of the line where the fault point is located, the impedance equation set between the fault point and each detection point is established, the impedance equation set is solved, and the first position information of the fault point is obtained. The electrical parameters of the line include resistivity and reactance rate;

[0062] Step S503, on the branch line of the multi-level distribution network, according to the propagation speed and time difference of the traveling wave on the line where the fault point is located, the distance between the fault point and each detection point is calculated, and the multi-point ranging method is used to determine the second position information of the fault point according to the distance between the fault point and each detection point. The accuracy of the second position information is higher than that of the first position information.

[0063] Specifically, different fault location methods are determined according to different hierarchical information and fault types to improve the positioning accuracy and adaptability, and ensure its performance and stability under actual working conditions.

[0064] In some examples, rough positioning based on impedance method is used on the main line, by measuring the voltage and current at the time of failure, the equivalent impedance of the fault point is calculated, and then the fault distance is estimated according to the line parameters. Considering the double-end and multi-end measurement data, the positioning accuracy can reach within 100 meters. On the branch line, accurate positioning based on traveling wave method is used, according to the propagation time difference of traveling wave generated by fault on the line, the fault point position is calculated, by accurately identifying the traveling wave head, the problem of wave speed changing with frequency is solved, and the positioning accuracy is strived to reach within 50 meters. For complex line structure, such as multi-branch, non-homogeneous, etc., improved impedance method and traveling wave method are used. For impedance method, positioning algorithms considering fault resistance are introduced, such as Takagi method, Eriksson method, etc., to improve positioning accuracy and adaptability. For traveling wave method, signal processing techniques such as wavelet transform and Hilbert transform are introduced to improve the accuracy and reliability of traveling wave parameter measurement. Through simulation test and field test, the positioning algorithm is optimized and verified to ensure its performance and stability under actual working conditions.

[0065] A high-precision distribution network topology model is constructed, including the length of the line, the type of the conductor, the position of the tower, and the parameters of the switch and transformer. Through state estimation and topology analysis algorithm, the running state of the network is updated in real time, providing accurate input data for fault location. Through a large number of simulation and measured data, the law and influencing factors of positioning error are summarized, and the positioning result is self-adaptively corrected, continuously improving the positioning accuracy and reliability.

[0066] The advantages of impedance method and traveling wave method are comprehensively utilized to realize fast and accurate fault location. The fault location result is connected with geographic information system GIS, and the geographic coordinates and surrounding environment information of the fault point are automatically generated, providing intuitive and reliable support for repair decision, and finally realizing fault location to the level of tower, providing key support for intelligent operation and management of distribution network. In the distribution automation master station system, fault recording data is collected from intelligent electronic device IED of substation in real time through IEC61850 protocol, the sampling frequency can reach 10kHz, and the voltage and current waveforms of three stages before, during and after fault are included. For the main line, two-end impedance method is used for rough positioning, first calculating the positive sequence voltage drop ΔU and the positive sequence current I, then according to the line unit length impedance parameters R and X, the fault distance L = |ΔU| / (I×sqrt(R 2+ X2), the positioning accuracy is about 100 meters. For branch lines, the traveling wave method is used for accurate positioning, the wavelet transform is used for fault traveling wave detection and extraction, the sampling data is decomposed by Db4 wavelet for 5 layers, the mutation position of scale coefficient amplitude is extracted, the traveling wave head arrival time t is determined, and then the fault distance L = v x t / 2 is calculated according to the line wave speed v and the line length S, and the positioning accuracy can reach 50 meters. For complex lines with inhomogeneity and branches, the Takagi method and the wavelet traveling wave method are combined, the parameters such as fault resistance and distributed capacitance are introduced, and the positioning accuracy and reliability are improved through iterative calculation and optimization solution.

[0067] At the same time, through artificial intelligence algorithms such as support vector machine SVM, historical fault location data is trained and learned to establish a nonlinear mapping relationship between fault distance and influencing factors, and the fault point position is predicted online, and the results of impedance method and traveling wave method are fused and corrected, and the error is controlled within 20 meters. The positioning results of the fault point are associated with geographic coordinates and line resource information, and are displayed intuitively on the electronic map, and a fault analysis report is generated to guide the repair personnel to quickly and accurately find the fault point, and realize intelligent fault location and disposal.

[0068] Among them, the distribution network topology model is constructed, including the length of the line, the type of the conductor and the position attribute of the tower, as well as the parameters of the switch and transformer equipment, and the running state of the network is updated in real time through topology analysis and power flow calculation method.

[0069] Among them, after the above fault point is positioned by the above fault positioning method, the above method further includes the following steps:

[0070] Step S601, according to the fault condition of the above fault point and the corresponding fault influence evaluation model, the fault influence range and the influence severity corresponding to the fault condition are determined, and the fault isolation priority of the fault point is determined according to the fault influence range and the influence severity, and the fault condition includes the fault type of the fault point and the position information of the fault point;

[0071] Among them, according to the results of fault diagnosis and positioning, the type and position of the fault are determined, including single-phase grounding fault and two-phase short-circuit fault, as well as the line, bus and transformer equipment where the fault occurs, and the fault point information is obtained as the input data of the fault influence domain evaluation model.

[0072] Among them, according to the fault condition of the above fault point and the corresponding fault influence evaluation model, the fault influence range and the influence severity corresponding to the fault condition are determined, and the fault isolation priority of the fault point is determined according to the fault influence range and the influence severity, including the following steps:

[0073] Step S701, starting from the fault point, traversing the topology structure of the multi-level power distribution network, searching for all lines and devices directly or indirectly connected with the fault point, forming a fault influence tree with the fault point as the root node, to determine the fault influence range, wherein each node of the fault influence tree represents an affected element, and the number of layers of the fault influence tree represents the indirect degree of influence of the element;

[0074] Step S702, according to the fault influence tree, calculating the fault current and voltage deviation index of each affected element, using the node voltage method and / or branch current method for power flow calculation, analyzing the influence of the fault on the voltage, current and power parameters of each node and branch, and determining the influence severity;

[0075] Step S703, constructing a fault sensitivity matrix, and evaluating the fault influence range and key influencing factors under different fault conditions through the fault sensitivity matrix, wherein the rows of the fault sensitivity matrix represent the fault type of the fault point and the position information of the fault point, the columns of the fault sensitivity matrix represent the affected elements, and the elements of the fault sensitivity matrix represent the influence coefficient of a certain fault on a certain element;

[0076] Step S704, determining the fault isolation priority of the fault point based on the fault sensitivity matrix.

[0077] Specifically, according to the results of fault diagnosis and positioning, the type and location of the fault are determined, the detailed information of the fault point is obtained, and input conditions are provided for subsequent influence domain evaluation, thereby improving the safety, reliability and resilience of the power distribution network.

[0078] In some examples, according to the results of fault diagnosis and positioning, the type and location of the fault are determined, such as single-phase ground fault, two-phase short-circuit fault, etc., and the line, bus, transformer and other devices where the fault occurs, the detailed information of the fault point is obtained, and input conditions are provided for subsequent influence domain evaluation. A breadth-first search (BFS) algorithm is used, starting from the fault point, traversing the topology structure of the power distribution network layer by layer, searching for all lines and devices directly or indirectly connected with the fault point, until all related nodes are visited, forming a fault influence tree with the fault point as the root node, each node of the tree representing an affected element, and the number of layers of the tree representing the indirect degree of influence of the element, thereby determining the influence range of the fault. On the basis of the fault influence tree, impedance matrix method is used for fault analysis, according to the fault type and location, the impedance matrix of the power distribution network is modified, the node voltage and branch current distribution under fault are calculated, the voltage deviation and overcurrent severity indexes are obtained, and the influence of the fault on each node and branch is quantitatively evaluated.

[0079] A fault sensitivity matrix is constructed, and the elements of the matrix are voltage sensitivity coefficients and current sensitivity coefficient The effects of the change in active power at node j on the voltage at node i and the current in branch ij are represented by the Jacobian matrix, respectively.

[0080] Important loads in the distribution network, such as hospitals and transportation hubs, are identified and classified. A hierarchical structure model is established by combining the Analytic Hierarchy Process (AHP) and the Delphi method. Pairwise comparison matrices are constructed, weight vectors are calculated, and the consistency of the judgment matrix is ​​verified. Taking into account factors such as load capacity, type, and importance, the importance weight of each load node is obtained and incorporated into the fault impact assessment model.

[0081] Based on fault impact assessment, and combined with the topological constraints and operational objectives of the distribution network, such as minimum outage range and fastest recovery time, a multi-objective optimization model is constructed. State variables for disconnecting switches and load transfer switches are defined, constraint equations for voltage, current, and power are established, and weighting coefficients for different fault types are set. Objective functions include minimizing the impact range, minimizing severity, and prioritizing critical loads. Multi-objective optimization algorithms such as NSGA-II and MOPSO are used to solve the model. The fault impact assessment model undergoes self-learning and updating. By collecting and analyzing historical fault event records, distribution network power flow snapshots, and expert experience data, machine learning algorithms such as support vector machines and decision trees are used to adaptively adjust and optimize the model's parameters and structure. Evaluation indicators such as accuracy, recall, and F1 score are selected to quantitatively evaluate the model's performance and generalization ability.

[0082] The time dimension is introduced to dynamically model the propagation and evolution process of fault influence in the space-time scale, analyze the key path of fault propagation and cascading failure mode, identify the key nodes and edges, and optimize the resilience and vulnerability of the distribution network. The complex network theory is used to characterize the topological properties and dynamic behavior of the distribution network, reveal the propagation mechanism and evolution law of fault influence, and provide theoretical basis and technical support for real-time monitoring, risk warning and defense control of the distribution network, and improve the safety, reliability and resilience of the distribution network. Assuming that the distribution network consists of 10 feeders and 100 nodes, when a single-phase ground fault occurs at node 27, the fault influence range is searched by BFS algorithm, and the search depth is set to 3 layers. The search results show that the fault affects 3 nodes in the first layer, 8 nodes in the second layer and 12 nodes in the third layer, a total of 23 nodes are affected. Then the impedance matrix method is used to calculate the short-circuit current and voltage deviation of the affected nodes, and it is found that the short-circuit current of 2 nodes exceeds 1.5 times of the rated value, and the voltage deviation of 5 nodes exceeds 10% of the rated value, which is judged as a serious influence node. Then the fault sensitivity matrix is calculated, based on the active power injection change rate of 0.1 MW, the sensitivity coefficient of node 27 fault to all nodes voltage is calculated, the maximum value is-0.85, that is, node 27 increases 0.1 MW active power injection, the voltage of a node decreases 0.085 kV. For the 10 key load nodes, the analytic hierarchy process (AHP) is used for evaluation, by considering 8 key factors such as fault rate, load, customer type, electricity price, etc., a hierarchical evaluation system including three levels is constructed. In this process, 5 experts in the field are invited to carefully score the relative importance between factors, the scoring standard is 1 to 9 points, according to which the weight vector reflecting the importance of load is calculated, the highest scoring item has a weight of 0.25, reflecting its prominent position in decision-making.

[0083] On this basis, a mixed integer nonlinear programming (MINLP) optimization model is further established to improve the fault response capability of the power grid through optimization decisions. The decision variables of the model involve the switching states of 23 disconnectors, while a series of basic constraints are set, including ensuring voltage stability, maintaining a single radial network structure, and prioritizing the restoration of important load power supply, etc. To solve this model, the extremal climbing algorithm is used for iterative search. After 500 generations of operation, each generation has a population size of 50, the crossover probability is set to 0.8, and the mutation probability is 0.1. Finally, a set of Pareto optimal solutions is identified. Through sorting analysis of these solutions, a compromise scheme that meets the requirements of multiple objectives and is relatively balanced is selected, and the corresponding fault isolation operation sequence is determined. In order to verify the effectiveness of the model, 2000 real fault cases accumulated in the past 10 years are used for training, and 10-fold cross-validation is implemented. The results show that the average prediction accuracy is more than 95%, indicating that the model has high practical value. Based on the analysis of complex network theory, the average degree of the original power grid network is 〈k〉 = 3.8, and once a fault occurs, the average degree decreases to 2.2, revealing a significant weakening of network connectivity, emphasizing the urgency of improving the resilience of the power grid and taking effective measures. In addition, by simulating the removal of the top 10% key nodes in terms of degree, betweenness centrality and closeness centrality, it is found that the maximum decrease in network efficiency can reach 60%. This result clearly points out these structurally vulnerable key nodes, emphasizing that they are the focus of future monitoring and protection, and are of great significance to maintaining the stable operation of the power grid.

[0084] In step S602, a fault isolation control strategy is determined based on the fault conditions and the fault isolation priorities of each of the fault points, and the fault isolation control strategy is used to sequentially control the switching devices in the fault area to minimize the fault influence range and the power outage time.

[0085] In step S602, a fault isolation control strategy is determined based on the fault conditions and the fault isolation priorities of each of the fault points, and the fault isolation control strategy is used to sequentially control the switching devices in the fault area to minimize the fault influence range and the power outage time.

[0086] In step S801, the fault time-effect type of the fault point is determined according to the position coordinates and the topological region to which the fault point belongs. The fault time-effect type is either a permanent fault or a transient fault.

[0087] In step S802, in the case where the fault time-effect type is the permanent fault, the emergency degree index of the fault is calculated in combination with the fault isolation priority, the fault severity, the influence range, and the important load factor.

[0088] Step S803: Based on the topology and real-time operating status of the multi-level distribution network, obtain the location and status information of all switching devices in the fault area, establish decision variables and constraints for switching operations, and form an optimized configuration of disconnecting switches. The aforementioned switching devices include manual switches, automatic switches, and disconnecting switches.

[0089] Step S804: Based on the urgency index of the fault and the optimized configuration of the disconnecting switch, set the corresponding state space, action space and reward function, and generate the optimal switch operation sequence with the goal of minimizing the impact range and power outage time of the fault, to obtain the fault isolation control strategy.

[0090] like Figure 3 As shown, Figure 3 The diagram illustrates a flowchart of configuring a disconnect switch operation according to an embodiment of this application. First, the fault location coordinates and area are obtained through diagnosis and positioning. The fault type is determined to be permanent or transient. If it is a permanent fault, the isolation control strategy is activated. If it is a transient fault, the state change is recorded and detected. Then, the fault isolation priority is determined, the fault urgency index is calculated, and different levels of faults are handled according to the index. Finally, the disconnect switch operation is optimized.

[0091] Specifically, based on the fault impact assessment, and combined with the topological constraints and operational objectives of the distribution network, such as the minimum outage range and the fastest recovery time, a multi-objective optimization model is constructed. Constraint equations such as voltage, current, and power are established, and weight coefficients for different fault types are set. This method introduces the time dimension to dynamically model the propagation and evolution of fault impacts in the spatiotemporal scale, thereby optimizing the resilience and vulnerability of the distribution network.

[0092] In some examples, based on the fault impact assessment model, the isolation priority of the fault is determined, considering factors such as fault severity S, impact range A, number of important loads N, etc., calculating the fault urgency index I = S x A x N, using the analytic hierarchy process to determine the weight of each factor, and processing faults of different levels in order from high to low. According to the topology and real-time operating state of the distribution network, the position and state information of all switch devices in the fault area are obtained, including manual switches, automatic switches, disconnectors, etc., the decision variables and constraint conditions of switch operation are established, the decision variables are the states (0 / 1) of the switches, such as 0 indicating that the switch is open and 1 indicating that the switch is closed, and the constraint conditions include voltage amplitude U_min≤U≤U_max, power flow P_ij≤P_max, feeder topology G_t∈T, etc., forming the optimal configuration problem of disconnectors. U_min represents the lower limit of voltage, which is the minimum voltage value allowed in the operation of a node or line in the distribution network, and below this value may affect the power supply quality and equipment safety. U_max represents the upper limit of voltage, and correspondingly, it is the maximum allowed voltage value, and exceeding this value may cause equipment overvoltage damage or power quality problems.

[0093] P_ij refers to the power flow from node i to node j, indicating the active power flowing between the two nodes. P_max represents the maximum allowed power flow of a certain line or device. If the actual power flow exceeds this value, it may cause overload, and measures such as network reconfiguration or load restriction need to be taken to avoid it. G_t represents the topology of the distribution network at time t (i.e. the real-time operating state), represented by a graph in graph theory, containing nodes (such as substations, load points) and edges (such as transmission lines). ∈T represents that G_t belongs to a certain topology in the set of all possible topologies T.

[0094] T contains all legal connection modes of the power distribution network, that is, the network structure that meets the physical connection rules and operation rules. The switch operation problem is modeled and solved by using the deep reinforcement learning algorithm DQN. The input state is the fault type, the status of the disconnecting switch, etc. The output action is the switch operation combination, and the reward function R is the weighted sum of the fault isolation range A and the outage time T, R = -w1xA-w2XT, the weights w1 and w2 are used to balance the importance of the two factors. The neural network contains 3 hidden layers, each layer has 64 neurons, the activation function is ReLU, the learning rate is 0.001, the discount factor is 0.9, the experience replay capacity is 10000, the target network update frequency is 100 steps, and the intelligent agent is trained to minimize the fault influence range and the outage time as the goal to generate the optimal switch operation sequence. In the process of generating the switch operation sequence, the impact of each step operation on the power grid topology is evaluated in real time. Through power flow calculation and fault analysis, it is judged whether the operation will cause a new fault or violate the constraint condition. The Newton-Raphson method is used for power flow calculation, and the error is less than 1e-6. The sequential Monte Carlo method is used for fault analysis, and the number of fault scenarios is 1000. If the constraint condition is not met, the operation is backtracked and modified. For important loads such as hospitals and transportation hubs, the standby power source is started or the temporary transfer strategy is adopted. The standby power source capacity is not less than 80% of the important load, and the switching logic is to put in when the fault area and the adjacent area are simultaneously powered off. The load priority is sorted according to the importance, and the transfer line selection is based on the minimum cut set algorithm. Through distributed energy, emergency power generation vehicles, etc., the power supply of important loads is restored as soon as possible to reduce the impact of faults on key users.

[0095] While performing the fault isolation operation, the operating state and load recovery of the power grid are continuously monitored, the isolation control strategy is dynamically adjusted according to the actual effect and environmental changes, and an adaptive dynamic programming algorithm is used. The state transition equation is s_t+1=f(s_t,a_t), which represents the rule of transition from the current state s_t to the next state s_t+1 after action a_t. The function f defines how the state changes with the action, and embodies the model of the dynamic behavior of the power grid, where s_t represents the state of the power grid at time t, a_t represents the specific operation taken at time t to isolate the fault or restore the normal operation of the power grid, such as opening or closing a certain isolation switch, the value function is approximated using a neural network, the policy iteration period is 10 minutes, the time complexity is O(nlogn), the space complexity is O(n), and the adaptive capability index is the fault isolation range reduction rate and the load recovery time shortening rate. The strategy is optimized and updated online to improve the real-time and adaptability of fault handling. Through the information of the fault indicator and the microcomputer protection device, it is judged that the No. 2 main transformer has a B-phase transient ground fault, the fault duration is 0.2 seconds, the fault current is 3.5 times the rated current, it is a transient fault, and it is not isolated for the time being. Reclosing after a delay of 1 second, continue to monitor the fault state. If the fault occurs again within 1 minute, it is determined to be a permanent fault, and the isolation control is started.

[0096] According to the fault severity S=3 (current exceeds 3 times the rated value), the influence range A=2 (2 10kV feeder lines), the number of important loads N=1 (1 three-A-level hospital), the fault emergency index I=6 is calculated, which belongs to high priority fault and needs to be isolated preferentially. The state information of 5 boundary switches and 10 disconnectors in the fault area is collected through the all-fiber communication network, and a disconnector optimization model is established, with the constraint conditions being that the feeder voltage deviation is not more than 5%, the line flow is not more than 80% of the rated value, and the feeder maintains a tree topology. A DQN reinforcement learning network is built using TensorFlow, with the input state dimension being 15, the output action dimension being 10, the hidden layer being 3 layers, each layer having 64 nodes, the activation function being ReLU, the optimizer being Adam, the learning rate being 0.001, the discount factor being 0.9, the replay capacity being 10,000, and the target network update frequency being 100 steps. Through 10,000 training iterations, the loss function converges to below 0.05, and the optimal switch operation sequence is generated. The operation sequence is evaluated in real time using the Pandapower power flow calculation program and the fault analysis program, and a hybrid algorithm of Newton method and fast decoupling method is adopted, with an error of less than 1e-6. 1,000 fault scenarios are sampled for risk assessment, and it is found that the third step operation will cause overload of disconnector No. 6, which is modified to disconnect switch No. 6 and close switch No. 7. For the only three-A-level hospital load in the fault area, the power supply is transferred through the 500kW diesel generator on site and the standby line of the adjacent feeder. The diesel generator switching logic is to start when the hospital suddenly loses power and the standby line also loses power, and the load recovery time is not more than 30 seconds. During the fault isolation process, the power grid state and isolation / recovery strategy are re-evaluated every 5 minutes, the Q value function is updated using the policy iteration algorithm, the state space dimension is 30, the action space dimension is 20, the discount factor is 0.95, and the convergence precision is 1e-4. By adaptively adjusting the control strategy of the disconnectors and standby power sources, the fault isolation range is reduced by 15% and the important load recovery time is shortened.

[0097] In step S603, real-time communication optimization is performed on the multi-level distribution network, and a data transmission scheduling strategy of the multi-level distribution network is established.

[0098] In step S603, real-time communication optimization is performed on the multi-level distribution network, and a data transmission scheduling strategy of the multi-level distribution network is established.

[0099] In step S901, a communication network for power distribution automation is established by combining optical fiber communication and 5G wireless communication.

[0100] In step S902, fault data is compressed, encoded and encrypted, and classified and prioritized according to the type and priority of the data.

[0101] Step S903, a redundant transmission link is established between the key nodes of the communication network, and when the main link fails or is congested, the backup link is switched to continue transmitting data;

[0102] Step S904, a communication delay prediction model is established, and a communication delay prediction result is obtained according to the communication delay prediction model and delay influencing factors, wherein the communication delay prediction model is constructed according to historical delay data of the communication link, the delay influencing factors include network topology, data volume and priority, and the communication delay prediction model is used to predict the transmission delay under the current network state;

[0103] Step S905, the scheduling of data transmission is dynamically adjusted according to the communication delay prediction result and the priority of the data;

[0104] Step S906, a multi-objective optimization algorithm is used to balance the two objectives of minimizing delay and meeting priority, and the data transmission scheduling strategy is generated.

[0105] Specifically, the method can minimize the transmission delay between fault data and control commands, so as to respond quickly when a fault occurs.

[0106] In some examples, a combination of optical fiber communication and 5G wireless communication is used to construct a communication network for power distribution automation. Optical fiber transmission is used to transmit large-capacity fault recording data, and 5G network transmission is used to transmit low-latency control commands. The deployment location and density of 5G base stations are reasonably planned, and network slicing technology is used to allocate dedicated transmission resources for fault data and control commands. Through the complementation of multiple communication methods, the communication reliability and real-time performance are improved.

[0107] The fault data is compressed and encrypted for transmission. Wavelet transform and Huffman coding are combined to compress the fault recording data, for example, 1MB of original data is compressed to less than 100KB, which greatly reduces the data transmission volume and shortens the transmission time. At the same time, the AES-256 encryption algorithm is used to ensure the security of data transmission while controlling the additional delay introduced by encryption operations to be within 1ms. The communication protocol stack and parameter settings are optimized, and a lightweight transmission protocol based on UDP, such as GOOSE, SV, etc., is used to reduce protocol overhead and encapsulation time. Compared with traditional ModbusTCP and IEC104 protocols, the transmission delay of GOOSE protocol can be reduced to less than 1ms, meeting the real-time requirements of fault protection action. By optimizing TCP / IP parameters, such as setting the buffer size to 1MB and selecting BBR as the congestion control algorithm, the network throughput and transmission efficiency are improved while ensuring reliable transmission.

[0108] According to the type and priority of data, the DiffServ mechanism is adopted to classify and mark the fault data and control commands, and to schedule the priority. The fault data is marked as gold service, and the control command is marked as platinum service, and the priority of platinum service is higher than that of gold service. In the process of network transmission, through the mechanisms of queue scheduling and traffic shaping, the end-to-end delay of platinum service is guaranteed to be less than 10 ms, and the delay of gold service is less than 50 ms, and the key data is given priority processing. Between the key nodes of the communication network, redundant transmission links are established, and SDN technology is used to realize flexible configuration and fast switching of the primary and backup links. Combined with the topology of the power distribution network, double-link access is set at each substation and switching station, and the link capacity meets the demand of 120% of the peak period. When the primary link fails or is congested, the backup link can complete switching within 20 ms, ensuring the continuity and reliability of data transmission. Through big data analysis and machine learning technology, a communication delay prediction model is established, historical delay data of communication links are collected, and key features affecting delay are extracted, such as network topology, routing hop count, channel utilization, etc.

[0109] By using feature normalization and principal component analysis (PCA) methods, the features are preprocessed and reduced in dimension, and a support vector regression (SVR) model is constructed by selecting a Gaussian kernel function, and the model hyperparameters are optimized by grid search, such as penalty coefficient C=10, relaxation variable ξ=0.1, etc. The prediction model is trained and verified, and the average prediction error is controlled within 10%, and the transmission delay under the current network state is predicted in real time, which provides a basis for subsequent scheduling decisions. According to the communication delay prediction results and data priority, the scheduling of data transmission is dynamically adjusted, and the NSGA-III multi-objective optimization algorithm is adopted to balance the two objectives of minimizing delay and satisfying priority.

[0110] The optimal scheduling strategy parameters such as packet size and transmission interval are searched through non-dominated sorting, congestion calculation and other operations. After 100 iterations, the Pareto frontier converges, the latency minimization target improves by 20%, and the priority satisfaction target improves by 15%. Edge computing nodes are deployed at the distribution end to analyze and process local data such as fault recording and alarm events, extract key features and statistical indicators, and only upload the processed result data to the main station, reducing data transmission volume and latency. High-reliability MQTT protocol is used for data synchronization between edge nodes and the main station, with transmission latency controlled within 100 ms. At the same time, a dynamic scheduling mechanism for communication resources is established, considering real-time traffic, priority and network load of business data, dynamically allocating transmission bandwidth and computing resources to ensure the quality of service of critical business and improve the overall efficiency of the communication network. A 24-core single-mode optical cable is laid between the substation and the switchyard, with a transmission bandwidth of 10 Gbps and a delay of 1 ms / km. At the same time, a 5G micro base station with a coverage radius of 500 m is deployed at each switchyard, with an uplink peak rate of 1 Gbps and a latency of 10 ms. When a fault occurs, the nearest switchyard collects 96 kHz, 16-bit current and voltage waveform data with a sampling period of 20 ms, collecting 1.92 MB of data per cycle. DB10 wavelet transform is used to decompose the data into 8 layers, extract fault transient features, and then combine Huffman coding to compress the data volume to 192 KB, with a compression ratio of 10:1. AES-256 algorithm is used to encrypt the data, with a key length of 256 bits and an encryption time of 0.5 ms. The encrypted data is transmitted in groups through the GOOSE protocol, with each group of data having a length of 128 bytes and a transmission latency of 0.2 ms.

[0111] Fault data and control commands are marked as DSCP gold and platinum services respectively, and four priority queues are configured, with platinum queue weight 50% and gold queue weight 30%, ensuring fault data latency within 20 ms and control command latency within 5 ms. SDN and OSPF protocols are used to build dual-active links between the distribution main station and the substation, with a primary link bandwidth of 1 Gbps and a backup link bandwidth of 500 Mbps, a link fault detection time of 10 ms, and a switching time of 20 ms. Historical link latency data for one year is collected, with a 5-minute cycle, extracting 15 features such as net element CPU utilization, memory occupancy, link traffic and packet loss rate, normalizing to 0-1, reducing features to 8 dimensions using principal component analysis, training a Gaussian kernel SVM regression model, 5-fold cross-validation, grid search penalty coefficient C and relaxation variable, optimal parameters C = 12.5, γ = 0.08, R 2The value reaches 0.95. Deploy the edge gateway, configure 4-core CPU, 8G memory, 512G SSD, run Docker container, use Kafka stream processing framework, real-time cleaning, statistical analysis of fault data, extract 10 characteristic parameters such as fault duration, amplitude, frequency, use MQTT protocol compression upload, transmission delay control within 20ms. Train NSGA-III scheduling algorithm, population size 50, crossover probability 0.8, mutation probability 0.1, iteration 200 times, optimal delay reduction 25% in Pareto frontier, priority improvement 10%, generate optimal scheduling parameters of data packet 128-256B, sending interval 5-10ms.

[0112] Step S604, based on the above data transmission scheduling strategy and the above fault isolation control strategy, operation instructions are issued to the switching devices in the fault area to isolate the fault point and restore power supply outside the fault area.

[0113] Wherein, based on the above data transmission scheduling strategy and the above fault isolation control strategy, operation instructions are issued to the switching devices in the fault area to isolate the fault point and restore power supply outside the fault area, including the following steps:

[0114] Step 1001, based on the above data transmission scheduling strategy and the above fault isolation control strategy, operation instructions are issued to the switching devices in the fault area, and the switching devices are controlled to perform disconnection operation to isolate the fault area, and the isolated fault area and non-fault area are obtained;

[0115] Step 1002, reconstructing the load and power supply of the isolated fault area to generate a power supply recovery strategy with the smallest network transmission loss and the least number of operations;

[0116] Step 1003, using real-time data and historical data of the distribution automation master station, using machine learning algorithm, stability evaluation and risk prediction are carried out on each of the isolated non-fault area, to obtain the feasibility and safety of the operation of the non-fault area;

[0117] Step 1004, if the feasibility and safety of the operation of the non-fault area do not meet the preset conditions, start the load reduction or grid operation strategy in the power supply recovery strategy;

[0118] Step 1005, real-time acquisition of voltage, current and power measurement data of the power distribution terminal, calculation of power supply reliability index, and comparison with the predetermined target value;

[0119] Step 1006, when the deviation between the power supply reliability index and the predetermined target value is greater than the deviation threshold, adjust the power supply recovery strategy to obtain the adjusted power supply recovery strategy.

[0120] Specifically, according to the results of fault diagnosis and positioning, the position coordinates of the fault point and the topological region to which it belongs are obtained, it is judged whether the fault type is a permanent fault or a transient fault, and classification is performed according to the fault duration, fault current size and other characteristics. For permanent faults, the fault isolation control strategy is started, for transient faults, it is not isolated for the time being, time delay reclosing is performed, and fault recording and alarm are performed, and the state change of the fault point is continued to be monitored. This method can restore power supply to important loads as soon as possible, reduce the impact of faults on key users, and continuously monitor the operation state of the power grid and the load recovery situation while performing fault isolation operation.

[0121] Among them, the real-time operation model of the distribution network is constructed in combination with the topological structure of the distribution network and the real-time load data;

[0122] According to the results of fault diagnosis and positioning, the control strategy for fault isolation and recovery is generated;

[0123] The generated control strategy is converted into a sequence of switch operation instructions, and the instructions are issued to the equipment outside the fault area to perform the disconnection operation, and the fault area is isolated from the surrounding healthy power grid;

[0124] The load and power of the isolated fault area are reconstructed, and the optimal power supply recovery scheme is found;

[0125] With the minimum network loss and the minimum number of operations as the target, the operation scheme and sequence of power grid reconstruction are generated, and the operation scheme is converted into control instructions;

[0126] Using the real-time data and historical data of the distribution automation master station, a machine learning algorithm is used to evaluate the stability and predict the risk of each isolated island after isolation, and to judge the feasibility and safety of island operation;

[0127] For islands that do not meet the requirements, load reduction or grid-connected operation strategy is started;

[0128] The execution effect of the power grid reconstruction and island operation scheme is dynamically evaluated, the voltage, current and power measurement data of the distribution terminal are obtained in real time, the power supply reliability index is calculated, and compared with the predetermined target value;

[0129] When the deviation exceeds the threshold value, the optimization control process is triggered, and the power supply recovery scheme is adjusted.

[0130] Specifically, this method can optimize real-time communication and minimize the transmission delay between fault data and control commands, so as to respond quickly when a fault occurs.

[0131] After the fault area is determined, the switches directly related to the fault in the fault area need to be disconnected first to prevent the flow of fault current and avoid further expansion of the fault. After the fault point inside the fault area is isolated, the switches located at the boundary of the fault area, which are usually located at the junction of the fault area and the healthy power grid, need to be disconnected. By disconnecting these switches, the fault area can be completely isolated and will not affect the healthy power grid. Controlling the disconnection of switches in the fault area is to isolate the fault source directly. Controlling the disconnection of switches outside the fault area is to prevent the spread of the fault to the healthy power grid. Both operations are to minimize the scope of the fault and restore normal power supply as soon as possible.

[0132] In some examples, by collecting the state information of each switch and monitoring point in real time online, combining the topology structure and real-time load data of the power distribution network, using graph theory methods such as maximum flow-minimum cut algorithm and shortest path algorithm, a real-time operation model of the power distribution network is automatically constructed, considering the radial and weak loop characteristics of the power distribution network, global situational awareness and visual display of the entire power distribution network are realized.

[0133] According to the results of fault diagnosis and positioning, a method combining rule-based reasoning engine and deep reinforcement learning is used to automatically generate control strategies for fault isolation and recovery. The rule base contains processing experience and expert knowledge for each type of fault. The deep reinforcement learning algorithm uses the DDPG model. The state space includes fault type, location, switch state, etc. The action space includes the operation combination of each switch. The reward function considers factors such as fault isolation range, outage time, network loss, etc. Through continuous trial and error and optimization, the control strategy is adjusted and improved online, and the convergence of the algorithm is proved. The generated control strategy is converted into specific switch operation instruction sequence, which is sent to the primary equipment near the fault area, such as circuit breakers, disconnectors, etc., through IEC61850 communication protocol and MMS message service, to perform disconnection operation, isolate the fault area from the surrounding healthy power grid, and prevent the spread of the fault impact.

[0134] For scenarios requiring fast switching, GOOSE messages are used for horizontal communication to ensure the real-time and reliability of command issuance. The load and power of the isolated fault area are reconstructed, and a mathematical model of power grid reconstruction is established, with the objective function being the minimization of network loss and operation times, and the constraint conditions including power flow constraints, voltage constraints, feeder capacity constraints, etc. The decision variables are the state combinations of switches. The branch and bound method and the tabu search algorithm are used to solve the model, with the branching strategy being according to the depth-first principle, the tabu list length being 50, and the maximum iteration number being 500 times. The optimal reconstruction scheme obtained is converted into control instructions, which are executed through the automation control system. The real-time and historical data of the distribution automation master station are used, and time-domain simulation and small signal analysis methods are used to evaluate the stability of each isolated island after isolation. Considering the load characteristics and new energy penetration rate in the island, the frequency and voltage deviation indicators are calculated and compared with the standard threshold. If the range is exceeded, it is determined to be unstable risk, and the load reduction or grid-connected operation strategy is automatically started. The execution effect of the power grid reconstruction and island operation scheme is dynamically evaluated, and through the service-oriented architecture, the voltage, current, power, and other measurement data of the distribution terminal are obtained in real time, and the system average interruption duration index (SAIDI) and SAIFI are calculated. Among them, SAIDI = ∑U_i*N_i / N_T, U_i is the i-th power outage time, N_i is the number of users affected by the i-th power outage, and N_T is the total number of users.

[0135] According to the actual situation of the distribution network, the threshold values of SAIDI and SAIFI are set to not more than 2 hours / year and 3 times / year respectively. When the actual indicators deviate from the threshold values, the optimization control process is automatically triggered to adjust the recovery scheme. A bidirectional communication mechanism is established between the distribution automation master station and the sub-stations such as substations and switching stations, and the MQTT protocol is used for data transmission. The MQTT message uses JSON format and QoS level 2 to ensure that the message is delivered only once. Through the TLS / SSL secure communication protocol and the AES-256 data encryption algorithm, the confidentiality and integrity of data transmission are ensured. The interval of the heartbeat mechanism is 5 seconds, and the timeout time is 10 seconds. After timeout, it is automatically retransmitted for 3 times. Time synchronization uses the IEEE1588 high-precision time protocol, with a synchronization accuracy of better than 1 millisecond. Active operation control technologies such as reactive power optimization and voltage control are introduced into the distribution network, and emerging technologies such as distribution Internet of Things and edge computing are used to build a distributed self-healing control architecture. Intelligent agents are deployed in distribution terminal devices to realize fast positioning, isolation and recovery of faults. Security communication protocols, intrusion detection, access control and other measures are used to prevent network attacks and data leaks and other risks, and to improve the security and reliability of the distribution automation system.

[0136] The operation state data of 86 intelligent switches, 135 distribution transformer terminals and 450 user smart meters on the 10kV distribution line are collected through the optical fiber Ethernet, uploaded to the distribution master station, and the power supply recovery path when the fault occurs is calculated using the Dijkstra shortest path algorithm to form a real-time simulation model of the distribution network. When a three-phase short-circuit fault occurs in the No. 1 line Yuanbaoshan transformer, based on the 132 rules of the fault diagnosis expert system, the fault type and fault phase are inferred in parallel, and at the same time the DDPG algorithm generates the optimal isolation strategy after 2000 iterations of exploration, isolates the fault transformer, and performs loop operation on the 9 healthy branch feeders to minimize the isolation range to 300 households. The master station issues 18 control commands to the 4 intelligent switches near the fault, and the tripping action is realized through the GOOSE message within 10ms to isolate the fault point.

[0137] The optimal network reconstruction scheme is solved by using the tabu search algorithm, and through deep rule pruning, the optimal topology is searched for 5000 times to operate 12 tie switches to restore power supply in the non-fault area and reduce network loss by 8%. Transient simulation analysis is performed on the three isolated islands, and if the frequency deviation exceeds 0.5Hz or the voltage deviation exceeds 7%, the system is judged to be unstable and 5% of the load is automatically cut off. During the fault isolation process, the SAIDI is 25.6min and the SAIFI is 0.6 times, which is much better than the annual target of 2h and 3 times. This fault forms 36 alarm events in the SCADA system, which are transmitted to the master station through IEC61850 MMS and automatically summarized into the fault management log. Subsequently, machine learning algorithms are used to adaptively optimize the protection setting value and protection logic to realize fault self-healing closed-loop control.

[0138] The above artificial intelligence-based power distribution network fault positioning method of the present application first determines the electrical parameters of the multi-level power distribution network based on the data collected by the power sensors in the multi-level power distribution network. The power sensors are installed on each level branch and node, and the electrical parameters include current and voltage. Then, the preset working parameters of the multi-level power distribution network are obtained, which are the preset working parameters for maintaining the stable operation of the multi-level power distribution network. Then, according to the electrical parameters, the characteristic values of each level of the multi-level power distribution network are determined, and according to the deviation of the electrical parameters from the preset working parameters, combined with the power grid topology relationship and the fault recognition model of the multi-level power distribution network, the fault type of the fault point is recognized. Finally, according to the fault type of the fault point and the level information of the multi-level power distribution network, the fault positioning method is determined, and the fault positioning method is used to position the fault point. The fault positioning method includes coarse positioning based on impedance method and precise positioning based on traveling wave method. This method can optimize real-time communication, minimize the transmission delay between fault data and control commands, so as to respond quickly when a fault occurs, and solves the problem that the existing power grid fault positioning algorithm can only be used for specific network topology and fault type, and cannot accurately identify the fault type of different levels and different fault scenes.

[0139] In order for those skilled in the art to more clearly understand the technical solutions of the present application, the implementation process of the artificial intelligence-based power distribution network fault positioning method of the present application will be described in detail below in conjunction with specific embodiments.

[0140] The present embodiment relates to a specific artificial intelligence-based power distribution network fault positioning method, as shown in Figure 4 The method comprises the following steps:

[0141] Step S1: based on the data collected by the voltage and current sensors in the multi-level power distribution network, integrate each electrical parameter;

[0142] Step S2: analyze the collected current and voltage data, and record the characteristic values of each level and the normal working parameters of the power grid;

[0143] Step S3: according to the deviation of the current and voltage data from the normal working parameters, combined with the power grid topology relationship, identify the fault type, which includes overload, short circuit, open circuit and grounding;

[0144] Step S4: combined with the diagnostic results of the fault type and the level information of the power grid, select the corresponding fault positioning method, including coarse positioning based on impedance method and precise positioning based on traveling wave method;

[0145] Step S5: for each type of fault and position, set the fault impact domain evaluation model;

[0146] Step S6: based on different fault positions, types and priorities, formulate a fault isolation control strategy;

[0147] Step S7: Real-time communication optimization is performed to minimize the transmission delay between the fault data and the control commands so as to respond quickly when the fault occurs.

[0148] Step S8: Operation instructions are issued based on the control strategy to isolate the located fault power grid part and simultaneously enable the preplan to allocate the surrounding healthy power grid resources to restore the power supply service outside the fault area.

[0149] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0150] The embodiment of the present application also provides a power distribution network fault positioning device based on artificial intelligence. It should be noted that the power distribution network fault positioning device based on artificial intelligence of the embodiment of the present application can be used to execute the power distribution network fault positioning method based on artificial intelligence provided by the embodiment of the present application. The device is used to realize the above-mentioned embodiment and preferred embodiment, and the description is not repeated. As used below, the term "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the device described in the following embodiment is preferably realized in software, the realization of hardware, or a combination of software and hardware, is also possible and conceived.

[0151] The power distribution network fault positioning device based on artificial intelligence described above includes a processor and a memory, and the first determination unit and the like are stored in the memory as program units. The corresponding functions are realized by the processor executing the program units stored in the memory. The modules are located in the same processor; or the modules are located in different processors in any combination.

[0152] The processor contains a core, and the core calls the corresponding program unit from the memory. The core can be set to one or more, and the problem that the power grid fault positioning algorithm in the prior art can only be used for a specific network topology and fault type and cannot accurately identify the fault type for different levels and different fault scenarios can be solved by adjusting the core parameters.

[0153] The memory can include a non-permanent memory in a computer readable medium, a random access memory (RAM) and / or a non-volatile memory such as a read-only memory (ROM) or a flash memory (flash RAM), and the memory includes at least one memory chip.

[0154] The embodiment of the present application provides a computer readable storage medium, the computer readable storage medium comprises a stored program, wherein the program controls a device where the computer readable storage medium is located to execute the power distribution network fault positioning method based on artificial intelligence when the program runs.

[0155] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. In addition, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0156] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0157] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0158] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for performing the functions specified in the flowchart

[0159] In a typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memories.

[0160] Memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM), electrically programmable read only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or non-volatile memory of other types in the nature of static or dynamic RAM. Memory is an example of computer readable storage media. A "computer-readable storage medium" can be any available medium or

[0161] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0162] It should also be noted that the terms "comprising", "containing", or any other variant thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not include only those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0163] From the above description, it can be seen that the above-mentioned embodiments of the present application achieve the following technical effects:

[0164] 1)、The above-mentioned power distribution network fault positioning method based on artificial intelligence of the present application first determines the electrical parameters of the multi-level power distribution network based on the data collected by the power sensors in the multi-level power distribution network. The power sensors are installed on each level branch and node. The electrical parameters include current and voltage. Then, the preset working parameters of the multi-level power distribution network are obtained. The preset working parameters are the preset working parameters for maintaining the stable operation of the multi-level power distribution network. Then, according to the electrical parameters, the characteristic values of each level of the multi-level power distribution network are determined, and according to the deviation of the electrical parameters and the preset working parameters, combined with the power grid topology relationship and the fault recognition model of the multi-level power distribution network, the fault type of the fault point is recognized. Finally, according to the fault type of the fault point and the level information of the multi-level power distribution network, the fault positioning method is determined, and the fault positioning method is used to position the fault point. The fault positioning method includes coarse positioning based on impedance method and fine positioning based on traveling wave method. This method can optimize real-time communication, minimize the transmission delay between fault data and control commands, so as to respond quickly when a fault occurs, and solve the problem that the existing power grid fault positioning algorithm can only be used for specific network topology and fault type, and cannot accurately identify the fault type of different levels and different fault scenes.

[0165] The above-mentioned is only the preferred embodiment of the present application and does not limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An artificial intelligence-based power distribution network fault location method, characterized in that, The method comprises the following steps: Based on the data collected by the power sensor in the multi-level power distribution network, the electrical parameters of the multi-level power distribution network are determined, the power sensor is installed on each level branch and node, and the electrical parameters include current and voltage; Obtain the preset working parameters of the multi-level power distribution network, the preset working parameters are the preset working parameters for maintaining the stable operation of the multi-level power distribution network; According to the electrical parameters, the characteristic values of each level of the multi-level power distribution network are determined, and according to the deviation of the electrical parameters and the preset working parameters, the fault type of the fault point is identified by combining the power grid topology relationship and the fault identification model of the multi-level power distribution network; According to the fault type of the fault point and the level information of the multi-level power distribution network, the fault positioning method is determined, and the fault positioning method is used to position the fault point, the fault positioning method includes coarse positioning based on impedance method and precise positioning based on traveling wave method.

2. The method of claim 1, wherein, Based on the data collected by the power sensor in the multi-level power distribution network, the electrical parameters of the multi-level power distribution network are determined, which comprises: According to the topological structure of the multi-level power distribution network, the installation position of the power sensor is determined, the installation position of the power sensor covers the branch and node of each level of the multi-level power distribution network, and the power sensor includes voltage sensor and current sensor; A unified data acquisition and transmission protocol is established, and based on the data acquisition and transmission protocol, the data collected by the power sensor of each monitoring point is summarized and synchronized to obtain initial power data; The initial power data is filtered, denoised and feature extracted to obtain the electrical parameters of the multi-level power distribution network.

3. The method of claim 1, wherein, According to the electrical parameters, the characteristic values of each level of the multi-level power distribution network are determined, and according to the deviation of the electrical parameters and the preset working parameters, the fault type of the fault point is identified by combining the power grid topology relationship and the fault identification model of the multi-level power distribution network, which comprises: The characteristic values reflecting the operating state of the multi-level power distribution network are extracted from the electrical parameters, and the characteristic values at least include voltage deviation and current harmonic content; The target deviation value of the electrical parameters of each detection point and the preset working parameters is calculated by using Euclidean distance or Manhattan distance measurement method, the preset working parameters at least include rated voltage, rated current and power factor, and the target deviation value at least includes current deviation value and voltage deviation value; The power grid topology relationship of the multi-level power distribution network is stored in the graph database in the form of adjacency matrix, wherein each node represents a bus or switch, each edge represents the connection relationship between two nodes, and the properties of the edge include the electrical parameters of the conductor, and the electrical parameters of the conductor at least include resistance and reactance; The decision tree algorithm is used to construct the fault identification model, the power grid topology relationship and the target deviation value of each detection point are input into the fault identification model, and the fault type of the fault point is obtained, and the fault type includes one of overload, short circuit, open circuit and grounding.

4. The method of claim 1, wherein, According to the fault type of the fault point and the hierarchical information of the multi-level power distribution network, a fault positioning method is determined, and the fault positioning method is used to position the fault point, comprising: According to the fault type of the fault point and the hierarchical information of the multi-level power distribution network, a coarse positioning method based on impedance method is used to calculate the first position information of the fault point on the main line of the multi-level power distribution network, and a fine positioning method based on traveling wave method is used to calculate the second position information of the fault point on the branch line of the multi-level power distribution network; On the main line of the multi-level power distribution network, the equivalent impedance between the fault point and the detection point is calculated according to the voltage data and current data at the fault time, and the impedance equation set between the fault point and each detection point is established according to the electrical parameters of the line where the fault point is located, the impedance equation set is solved, and the first position information of the fault point is obtained, the electrical parameters of the line include resistivity and reactance rate; On the branch line of the multi-level power distribution network, the distance between the fault point and each detection point is calculated according to the propagation speed and time difference of the traveling wave on the line where the fault point is located, and the second position information of the fault point is determined according to the distance between the fault point and each detection point by using multi-point ranging method, the accuracy of the second position information is higher than that of the first position information.

5. The method of claim 1, wherein, After the fault point is positioned by using the fault positioning method, the method further comprises: According to the fault condition of the fault point and the corresponding fault impact evaluation model, the fault impact range and the impact severity corresponding to the fault condition are determined, and the fault isolation priority of the fault point is determined according to the fault impact range and the impact severity, the fault condition includes the fault type of the fault point and the position information of the fault point; Based on the fault condition and the fault isolation priority of each fault point, a fault isolation control strategy is determined, which is used to sequence control the switching devices in the fault area to minimize the fault impact range and outage time; Real-time communication optimization is performed on the multi-level power distribution network, and a data transmission scheduling strategy of the multi-level power distribution network is established; Based on the data transmission scheduling strategy and the fault isolation control strategy, operation instructions are sent to the switching devices in the fault area to isolate the fault point and restore power supply outside the fault area.

6. The method of claim 5, wherein, According to the fault condition of the fault point and the corresponding fault impact evaluation model, the fault impact range and the impact severity corresponding to the fault condition are determined, and the fault isolation priority of the fault point is determined according to the fault impact range and the impact severity, comprising: Traverse the topology structure of the multi-level distribution network from the fault point, search all lines and devices directly or indirectly connected with the fault point, form a fault influence tree with the fault point as the root node, and determine the fault influence range, wherein each node of the fault influence tree represents an affected element, and the number of layers of the fault influence tree represents the indirect degree of influence of the element; According to the fault influence tree, calculate the fault current and voltage deviation index of each affected element, perform power flow calculation by using node voltage method and / or branch current method, analyze the influence of the fault on the voltage, current and power parameters of each node and branch, and determine the influence severity; Construct a fault sensitivity matrix, and evaluate the fault influence range and key influencing factors under different fault conditions through the fault sensitivity matrix, wherein the row of the fault sensitivity matrix represents the fault type of the fault point and the position information of the fault point, the column of the fault sensitivity matrix represents the affected element, and the element of the fault sensitivity matrix represents the influence coefficient of a certain fault on a certain element; Determine the fault isolation priority of the fault point based on the fault sensitivity matrix.

7. The method of claim 5, wherein, Based on the fault condition and the fault isolation priority of each fault point, determine the fault isolation control strategy, including: According to the position coordinates and the topological region to which the fault point belongs, determine the fault time limit type of the fault point, which is a permanent fault or a transient fault; In the case of the permanent fault, combine the fault isolation priority, the fault severity, the influence range and the important load factors to calculate the emergency index of the fault; According to the topology structure and real-time operation state of the multi-level distribution network, obtain the position and state information of all switch devices in the fault area, and establish the decision variables and constraint conditions of switch operation to form an isolation switch optimization configuration, wherein the switch devices include manual switches, automatic switches and isolation switches; According to the emergency index of the fault and the isolation switch optimization configuration, set the corresponding state space, action space and reward function, and generate an optimal switch operation sequence to obtain the fault isolation control strategy, with the goal of minimizing the fault influence range and outage time.

8. The method of claim 5, wherein, Optimize real-time communication of the multi-level distribution network, and establish a data transmission scheduling strategy of the multi-level distribution network, including: Establish a communication network of power distribution automation by combining optical fiber communication and 5G wireless communication; Compress and encrypt the fault data, and according to the type and priority of the data, mark and schedule the fault data and control commands by priority; Establish redundant transmission links between key nodes of the communication network, and when the main link fails or is congested, switch to the backup link to continue data transmission; A communication time delay prediction model is established, and a communication time delay prediction result is obtained according to the communication time delay prediction model and time delay influencing factors, wherein the communication time delay prediction model is constructed according to historical time delay data of a communication link, the time delay influencing factors include network topology, data volume and priority, and the communication time delay prediction model is used to predict transmission time delay under a current network state; According to the communication time delay prediction result and the priority of data, a data transmission schedule is dynamically adjusted; A multi-objective optimization algorithm is used to balance time delay minimization and priority satisfaction, and the data transmission schedule strategy is generated.

9. The method of claim 5, wherein, Based on the data transmission schedule strategy and the fault isolation control strategy, operation instructions are sent to switching devices in a fault area to isolate the fault point and restore power supply outside the fault area, including: Based on the data transmission schedule strategy and the fault isolation control strategy, operation instructions are sent to switching devices in a fault area, and the switching devices are controlled to perform disconnection operations to isolate the fault area, obtaining an isolated fault area and a non-fault area; The isolated fault area is reconstructed for load and power supply, and a power supply recovery strategy is generated with the smallest network transmission loss and the least number of operations as the target; Using real-time data and historical data of the distribution automation master station, a machine learning algorithm is used to evaluate the stability and predict the risk of each isolated non-fault area, and the feasibility and safety of the non-fault area operation are obtained. If the feasibility and safety of the non-fault area operation do not meet the preset conditions, the load reduction or grid-connected operation strategy in the power supply recovery strategy is started; Real-time voltage, current and power measurement data of the distribution terminal are obtained, power supply reliability indicators are calculated, and compared with predetermined target values; If the deviation between the power supply reliability indicators and the predetermined target values is greater than the deviation threshold, the power supply recovery strategy is adjusted to obtain an adjusted power supply recovery strategy.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a stored program, wherein when the program runs, the device where the computer readable storage medium is located executes the power distribution network fault positioning method based on artificial intelligence in any one of claims 1 to 7.

Images