Power distribution line fault identification and positioning method and system based on embedded terminal

Abstract

The application discloses a power distribution line fault identification and positioning method and system based on an embedded terminal, relates to the technical field of power distribution lines, and comprises the following steps: deploying embedded terminals at a key node set of a power distribution line to obtain a distributed sensing network; constructing a multi-source causal clue graph; obtaining a burst representation feature vector; performing causal clue path retrieval to generate a multi-source causal clue path set; performing window backtracking to output an implied root cause network node-fault type label set; and taking the key nodes corresponding to the burst representation feature vector and the key nodes corresponding to the implied root cause network node-fault type label set as fault identification and positioning results. The application solves the technical problem that the fault identification and positioning of the power distribution line in the prior art mainly depends on explicit abnormalities, cannot mine and trace the root cause of the fault, and leads to inaccurate positioning, and achieves the technical effect of improving the reliability of power distribution line fault identification.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power distribution lines, in particular to a power distribution line fault identification and positioning method and system based on embedded terminals. BACKGROUND

[0002] In the operation process of a power distribution network, a fault of a power distribution line directly affects the reliability of user power supply and the safe and stable operation of the power grid. Common faults of the power distribution line include grounding, short circuit, wire breakage and the like, and the faults have the characteristics of strong burstness, dispersed range and complex environment. At present, the fault identification of the power distribution line mainly positions the faults that have been represented, lacks in-depth mining, and thus the hidden faults have great security risks. SUMMARY

[0003] The application provides a power distribution line fault identification and positioning method and system based on embedded terminals, which is used to solve the technical problem that the fault identification and positioning of the power distribution line in the prior art mainly depends on explicit abnormalities, cannot mine and trace the root causes of the faults, and thus the positioning is inaccurate.

[0004] In view of the above problems, the application provides a power distribution line fault identification and positioning method and system based on embedded terminals.

[0005] In a first aspect, the application provides a power distribution line fault identification and positioning method based on embedded terminals, which comprises the following steps:

[0006] An embedded terminal is deployed at a key node set of a power distribution line to obtain a distributed sensing network, wherein each network node in the distributed sensing network corresponds to an embedded terminal; historical operation fault data sets and a power distribution line knowledge base are acquired to perform causal clue analysis, and a multi-source causal clue graph is constructed; a burst representation feature detection is performed on each embedded terminal in the distributed sensing network, and a burst representation feature vector is obtained; a causal clue path retrieval is performed on the multi-source causal clue graph based on the burst representation feature vector, and a multi-source causal clue path set is generated; a chain root cause backtracking instruction is acquired, the distributed sensing network is backtracked according to the multi-source causal clue path set, and an implicit root cause network node-fault type label set is output; and the key node corresponding to the burst representation feature vector and the key node corresponding to the implicit root cause network node-fault type label set are taken as a fault identification and positioning result.

[0007] Preferably, each embedded terminal has a node data buffer area, wherein the node data buffer area supports backtracking of historical data.

[0008] Preferably, the chain root cause backtracking instruction is acquired, the distributed sensing network is windowed backtracked according to the multi-source causal clue path set, and an implicit root cause network node-fault type label set is output, including: based on a detection time point of a burst representation feature vector, a historical data backtracking is performed on an embedded terminal node data buffer area in each multi-source causal clue path in the multi-source causal clue path set according to a preset backtracking window, and a path backtracking historical data sequence cluster is obtained; an implicit fault rate identification is performed on the path backtracking historical data sequence cluster, and an implicit root cause network node set is determined according to an identification result, and the implicit root cause network node-fault type label set is generated.

[0009] Preferably, the implicit fault rate identification is performed on the path backtracking historical data sequence cluster, and the implicit root cause network node set is determined according to an identification result, and the implicit root cause network node-fault type label set is generated, including: a first path backtracking historical data sequence is extracted from the path backtracking historical data sequence cluster; a long-short fluctuation trend analysis is performed on the first path backtracking historical data sequence, and a first long fluctuation trend feature vector and a first short fluctuation trend feature vector are generated; a feature enhancement is performed on the first short fluctuation trend feature vector based on the first long fluctuation trend feature vector, a first enhanced fluctuation trend feature vector is generated, and a first fault rate is identified according to the first enhanced fluctuation trend feature vector; in a similar manner, the fault rate identification is performed on the path backtracking historical data sequence cluster, a fault rate greater than or equal to a preset fault rate threshold is taken as an implicit fault rate, and a node corresponding to the implicit fault rate is taken as the implicit root cause network node set; a fault type identification is performed on an enhanced fluctuation trend feature vector corresponding to each implicit root cause network node, and the implicit root cause network node-fault type label set is generated.

[0010] Preferably, the long-short fluctuation trend analysis is performed on the first path backtracking historical data sequence, and the first long fluctuation trend feature vector and the first short fluctuation trend feature vector are generated, including: an abnormal duration set of a power distribution line in a historical time is extracted; a maximum value in the abnormal duration set is taken as a first identification scale; a minimum value in the abnormal duration set is taken as a second identification scale; a feature extraction is performed on the first path backtracking historical data sequence according to the first identification scale and the second identification scale respectively, and the first long fluctuation trend feature vector and the first short fluctuation trend feature vector are obtained.

[0011] Preferably, the first short fluctuation trend feature vector is feature enhanced based on the first long fluctuation trend feature vector to generate a first enhanced fluctuation trend feature vector, including: mapping element similarity recognition between the first long fluctuation trend feature vector and the first short fluctuation trend feature vector, and filling the recognition result into an initially empty matrix to construct a feature enhancement matrix; using the feature enhancement matrix to perform interactive convolution on the first short fluctuation trend feature vector to generate the first enhanced fluctuation trend feature vector.

[0012] Preferably, a historical operation fault data set and a distribution line knowledge base are obtained to sort out causal clues and construct a multi-source causal clue map, including: extracting the connection relationship between key nodes of the distribution line, the circuit breaker action logic and the line abnormality rules in the distribution line knowledge base; combining the connection relationship between key nodes of the distribution line, the circuit breaker action logic and the line abnormality rules, extracting and pairing events from the time series records in the historical operation fault data set to obtain a cause-effect event pair set; deduplicating the cause-effect event pair set to obtain a cleaned cause-effect event pair set; and constructing the multi-source causal clue map based on the cleaned cause-effect event pair set.

[0013] Preferably, each cleaning cause-effect event pair in the cleaning cause-effect event pair set is used as an edge in the graph, and the fault feature vector corresponding to each cleaning cause-effect event pair is used as a node in the graph to construct the multi-source causal clue graph.

[0014] Preferably, in combination with the connection relationship between the key nodes of the distribution line, the circuit breaker action logic and the line abnormality rules, the time series records in the historical operation fault data set are subjected to event extraction and pairing to obtain a cause-effect event pair set, including: extracting the time series records in the historical operation fault data set and performing structured conversion to obtain a historical operation fault structured sequence set; based on the connection relationship between the key nodes of the distribution line, the circuit breaker action logic and the line abnormality rules, the causes in the historical operation fault structured sequence set are retrieved to generate a cause-effect event pair set.

[0015] A second aspect of the present application provides a distribution line fault identification and positioning system based on an embedded terminal, the system comprising:

[0016] The distributed network obtaining module is configured to deploy embedded terminals at a set of key nodes of a power distribution line to obtain a distributed sensing network, wherein each network node in the distributed sensing network corresponds to an embedded terminal; the clue map library construction module is configured to obtain a set of historical operation fault data and a power distribution line knowledge base to perform causal clue analysis and construct a multi-source causal clue map; the burst characterization feature vector obtaining module is configured to traverse each embedded terminal in the distributed sensing network to perform burst characterization feature detection and obtain a burst characterization feature vector; the clue path set generation module is configured to perform causal clue path retrieval in the multi-source causal clue map based on the burst characterization feature vector to generate a multi-source causal clue path set; the label set obtaining module is configured to obtain a chain root cause backtracking instruction, perform window backtracking on the distributed sensing network according to the multi-source causal clue path set, and output a hidden root cause network node-fault type label set; and the fault identification and positioning result obtaining module is configured to take the key node corresponding to the burst characterization feature vector and the key node corresponding to the hidden root cause network node-fault type label set as a fault identification and positioning result.

[0017] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0018] In the present application, embedded terminals are deployed at a set of key nodes of a power distribution line to obtain a distributed sensing network, wherein each network node in the distributed sensing network corresponds to an embedded terminal. Then, a set of historical operation fault data and a power distribution line knowledge base are obtained to perform causal clue analysis and construct a multi-source causal clue map. Further, each embedded terminal in the distributed sensing network is traversed to perform burst characterization feature detection and obtain a burst characterization feature vector. Then, causal clue path retrieval is performed in the multi-source causal clue map based on the burst characterization feature vector to generate a multi-source causal clue path set. A chain root cause backtracking instruction is obtained, window backtracking is performed on the distributed sensing network according to the multi-source causal clue path set, and a hidden root cause network node-fault type label set is output. The key node corresponding to the burst characterization feature vector and the key node corresponding to the hidden root cause network node-fault type label set are taken as a fault identification and positioning result. The technical effect of reliably determining a fault position through deployment of distributed embedded terminals at key nodes of a power distribution line and performing data backtracking analysis to improve positioning accuracy is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort.

[0020] Figure 1 A schematic diagram of a flow chart of a method for identifying and locating a power distribution line fault based on an embedded terminal provided in an embodiment of the present application;

[0021] Figure 2 A schematic diagram of the structure of a distribution line fault identification and positioning system based on an embedded terminal provided in an embodiment of the present application.

[0022] Explanation of the accompanying symbols: distributed network acquisition module 11, clue graph library construction module 12, burst characterization feature vector acquisition module 13, clue path set generation module 14, label set acquisition module 15, fault identification and positioning result acquisition module 16. DETAILED DESCRIPTION

[0023] This application provides a distribution line fault identification and positioning method and system based on an embedded terminal, which is used to solve the technical problem that the fault identification and positioning of distribution lines in the existing technology mainly relies on obvious anomalies, cannot explore and trace the root cause of the fault, and leads to inaccurate positioning.

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0025] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0026] Example 1, as Figure 1 As shown, the present application provides a method for identifying and locating distribution line faults based on an embedded terminal, wherein the method includes:

[0027] Step S100: deploying embedded terminals at key node sets of the power distribution line to obtain a distributed sensing network, wherein each network node in the distributed sensing network corresponds to an embedded terminal;

[0028] Furthermore, each embedded terminal has a node data cache area, wherein the node data cache area supports backtracking historical data.

[0029] In a possible embodiment, the power distribution line is any line that needs to be subjected to fault identification positioning. The embedded terminal is an intelligent device deployed at a key node of the power distribution line for real-time monitoring and data collection. Each embedded terminal can sample current, voltage, frequency spectrum and other signals, and has certain processing and storage capabilities for data preprocessing, feature extraction and other tasks. Each embedded terminal has a node data cache area, which is an internal storage unit of the embedded terminal, used to cache real-time data collected by the node, and can save historical data within a time window set by a person skilled in the art to support subsequent backtracking operations. By configuring the cache area to support backtracking operations, historical data within a certain time window (such as 10 seconds, 30 seconds or longer) is stored in the cache area, achieving the technical effect of providing backtracking data for cause analysis and positioning when a subsequent terminal fails.

[0030] Preferably, the set of key nodes includes, for example, power distribution transformers, switching devices and feeder outlets. An embedded terminal is deployed at each key node, and each embedded terminal transmits data with other embedded terminals through a standard communication protocol (such as LoRa, NB-IoT, 5G, etc.) to form a distributed sensing network. That is, the distributed sensing network refers to a network architecture composed of multiple embedded terminals, wherein each node (embedded terminal) is interconnected through a communication protocol to jointly complete real-time monitoring and fault identification of the power distribution line. The number of nodes in the distributed sensing network is consistent with the number of key nodes in the set of key nodes. Each network node corresponds to an embedded terminal.

[0031] For example, when voltage drops occur at a key node of the power distribution network, the embedded terminal can collect the voltage value in real time and store it in the data cache area. If voltage recovery is abnormal, the system can find signs of current fluctuations or harmonic increases 10 seconds before the voltage drop in the historical data in the cache area, which may indicate that the cable has a poor contact problem. By backtracking the cache area data, the system can accurately infer the root cause of the fault.

[0032] Step S200: Obtain a set of historical operation fault data and a power distribution line knowledge base to comb causal clues and construct a multi-source causal clue map;

[0033] In one possible embodiment, the historical operation fault data set assigns fault data collected in historical operation of the power distribution line, usually including fault type, time of fault occurrence, duration, impact range, device state, alarm record, etc. These data can reflect the performance of the power distribution line under different fault situations. The power distribution line knowledge base contains basic information and related rules of the power distribution line. Optionally, it includes the topology structure of the power distribution line (such as the connection relationship between nodes), device parameters (such as the action logic of circuit breakers, switch operation rules), environmental factors (such as temperature and humidity, load fluctuation, etc.), and empirical operation rules (such as the regularity of some historical faults). The multi-source causal clue graph is a causal graph structure constructed by multiple information sources, which represents the causal relationship between the associated key nodes corresponding to the fault, and displays the relationship between different fault types and their related factors in a graphical manner.

[0034] Further, the historical operation fault data set and the power distribution line knowledge base are acquired for causal clue sorting to construct a multi-source causal clue graph. The step S200 of the embodiment of the present application further includes:

[0035] Extracting the connection relationship between key nodes of the power distribution line, the circuit breaker action logic, and the line abnormality rules in the power distribution line knowledge base;

[0036] Combining the connection relationship between key nodes of the power distribution line, the circuit breaker action logic, and the line abnormality rules, performing event extraction and pairing on the time sequence records in the historical operation fault data set to obtain a cause-effect event pair set;

[0037] De-duplicating the cause-effect event pair set to obtain a cleaned cause-effect event pair set;

[0038] Constructing the multi-source causal clue graph based on the cleaned cause-effect event pair set.

[0039] Further, each cleaned cause-effect event pair in the cleaned cause-effect event pair set is taken as an edge in the graph, and a fault feature vector corresponding to each cleaned cause-effect event pair is taken as a node in the graph to construct the multi-source causal clue graph.

[0040] Further, the step S200 of the embodiment of the present application further includes:

[0041] Extracting the time sequence records in the historical operation fault data set for structured conversion to obtain a historical operation fault structured sequence set;

[0042] Searching for reasons in the historical operation fault structured sequence set based on the connection relationship between key nodes of the power distribution line, the circuit breaker action logic, and the line abnormality rules to generate a cause-effect event pair set.

[0043] Preferably, the connection relationship between the key nodes of the power distribution line in the power distribution line knowledge base, the circuit breaker action logic and the line abnormal rule are extracted, wherein the circuit breaker action logic includes the action condition and recovery strategy of the circuit breaker, and the line abnormal rule includes the correlation rule of the key node abnormality, such as that the device aging can cause the cable failure and the load fluctuation can cause the voltage instability.

[0044] The time sequence data related to the fault is extracted from the historical fault data set, including the time sequence record of the monitoring signals such as current, voltage, harmonic and load. Each record usually contains the following information: the timestamp of the fault occurrence; the parameters such as current and voltage at the corresponding moment; the type, node and device state of the fault.

[0045] The original time sequence data is structured and converted, that is, the time sequence data is converted into a standard format (such as a database record or a table), so that the data can be conveniently processed, queried and analyzed. For example, each parameter of each time sequence record is stored as a column in a data table, and each record is taken as a row. The historical fault data after the structured conversion is analyzed, and the data is deeply mined in combination with the topology structure of the power distribution line, the action logic of the circuit breaker and the line abnormal rule.

[0046] Through the topology structure of the power distribution network, the connection between the fault occurrence node and other nodes is identified, the path of the fault propagation is understood, and then the root cause of the fault occurrence is analyzed according to the action rule of the circuit breaker (for example: the circuit breaker trips when the current exceeds the set value), whether the fault is caused by the non-action or response delay of the circuit breaker is determined, and further, the operation rule of the power distribution line is compared, and whether there is an abnormal behavior (such as load fluctuation, current anomaly, harmonic increase, etc.) that violates the normal operation is checked. After the retrieval and analysis, the cause-and-effect event pair set shown in Table 1 is generated, that is, the cause and effect of each fault are identified.

[0047] Table 1 Cause-and-effect event pair list

[0048]

[0049] Only one of the repeated false pairs in the set of cause-effect event pairs is retained, thereby obtaining a set of cleaned cause-effect event pairs after deduplication. By constructing the set of cause-effect event pairs, the system can trace the occurrence path and root cause of the fault, thereby improving the accuracy and efficiency of fault diagnosis. For example, assume that an abnormal current occurs at a certain node, the circuit breaker fails to act in time, causing the overload of the device and the final damage of the device. By extracting historical data and combining the action logic of the circuit breaker, the system can identify the following cause-effect event pair: the cause event is a sudden surge of current; the effect event is device overload. This event pair provides a clue for subsequent root cause tracing. By analyzing the relationship between the current anomaly and the device overload, it can be inferred that the sudden surge of current is caused by the current fluctuation of the upstream node, thereby completing the positioning of the fault.

[0050] In one embodiment, the set of cleaned cause-effect event pairs is a set of cause-effect event pairs after cleaning, deduplication, and data consistency verification. Each event pair describes the relationship between the cause and result of a fault in the power distribution system. For example, current overload (cause) leads to circuit breaker tripping (effect). The multi-source cause-effect clue graph is a graph composed of multiple cause-effect event pairs, containing the cause-effect relationship between each node in the power distribution system. This graph combines the topology of the power distribution line, historical fault data, and device information to reason about the cause and propagation path of the fault. In the multi-source cause-effect clue graph, each node represents a fault feature or a monitored device state, usually represented by a feature vector, such as current, voltage, harmonic, etc. Each edge represents a cause-effect event pair, connecting two fault nodes and indicating how one fault feature leads to another.

[0051] Each cleaned cause-effect event pair represents an edge, which connects two nodes and indicates the cause-effect relationship between them. For example, a sudden surge of current (node A) may cause device overload (node B), so there will be an edge from node A to node B in the graph. For each event pair, connect its cause event and effect event to form an edge, and set the weight of the edge to the likelihood or strength of the event occurrence (e.g., the probability of current overload causing circuit breaker tripping). Therefore, each cleaned cause-effect event pair in the set of cleaned cause-effect event pairs is taken as an edge in the graph, and the fault feature vector corresponding to each cleaned cause-effect event pair is taken as a node in the graph, to construct the multi-source cause-effect clue graph. By constructing the multi-source cause-effect clue graph, historical fault data and power distribution line rules are presented in a graphical manner, achieving the technical effect of providing reliable support for the identification, prediction, and positioning of power distribution line faults.

[0052] Step S300: Traverse each embedded terminal in the distributed sensing network to detect burst characterization features and obtain a burst characterization feature vector;

[0053] Step S400: performing causal clue path retrieval in the multi-source causal clue graph based on the burst representation feature vector, to generate a multi-source causal clue path set;

[0054] In one possible embodiment, each embedded terminal is responsible for collecting real-time data of the corresponding key node in operation, including but not limited to current, voltage, frequency, power factor and other parameters. When an abnormality occurs in a certain key node (for example, sudden surge of current, sudden drop of voltage, etc.), the embedded terminal quickly extracts the corresponding burst representation feature vector from the real-time data through short-time Fourier transform. The burst representation feature vector is used to reflect the abnormal situation of the key node in real time. The burst representation feature vector is usually composed of a plurality of data points, representing the change trend of the relevant parameters when the fault occurs. For example, rapid rise of current, abnormal fluctuation of temperature, etc.

[0055] Further, the burst representation feature vector and the fault feature vector in the multi-source causal clue graph are used to calculate the similarity by using the cosine calculation formula, and the nodes located in the first m positions in the calculation result are added to the path center node set. Then, the clue paths of the path center node set in the multi-source causal clue graph are summarized and added to the multi-source causal clue path set, where m is an integer greater than or equal to 3.

[0056] For example, current overload can be caused by multiple factors, and different factors correspond to different clue paths. For example, the clue path of circuit breaker tripping includes: path 1: current surge → device overload → circuit breaker tripping. Path 2: sudden increase in load → current surge → device overload → circuit breaker tripping. Path 3: power fluctuation → current fluctuation → unstable device operation → circuit breaker tripping.

[0057] By performing causal path retrieval in the multi-source causal clue graph, the potential causes of the fault can be analyzed from multiple angles. The retrieval result of each causal path provides a different perspective for fault diagnosis.

[0058] Step S500: acquiring a chain root cause backtracking instruction, performing window backtracking on the distributed sensing network according to the multi-source causal clue path set, and outputting a hidden root cause network node-fault type label set;

[0059] Step S600: taking the key node corresponding to the burst representation feature vector and the key node corresponding to the hidden root cause network node-fault type label set as the fault identification and positioning result.

[0060] In one possible embodiment, the chain root cause backtracking instruction is used to backtrack the node data buffer area of the embedded terminal of each key node in the power distribution line according to the multi-source causal clue path set obtained according to the above steps to trace the root cause of the fault. Chain root cause backtracking means not only backtracking to the node directly causing the fault, but also backtracking to the potential factor upstream causing the fault of the node. Through window backtracking analysis, an implicit root cause network node is obtained, each implicit root cause node is associated with a corresponding fault type label (such as overload, short circuit, ground fault, etc.), a fault type label is formed, and a set of implicit root cause network nodes-fault type labels is constructed. The implicit root cause network node refers to a node that is not directly represented but indeed causes the current fault. For example, current fluctuation may not directly cause the fault, but it may be caused by upstream equipment aging. Through causal backtracking, the system helps to find the potential cause behind the direct fault event, especially the tracing of implicit faults. It is ensured that not only the current occurring fault can be located, but also the root cause thereof can be analyzed.

[0061] Further, the key node corresponding to the burst representation feature vector and the key node corresponding to the set of implicit root cause network nodes-fault type labels are taken as the fault identification and positioning result. That is, the fault identification and positioning result not only contains the fault position that has been represented, but also includes the implicit fault position that causes the represented fault. The technical effects of improving the fault identification and positioning accuracy of the power distribution line and improving the fault response efficiency are achieved.

[0062] Further, the chain root cause backtracking instruction is obtained, the distributed sensing network is windowed backtracked according to the multi-source causal clue path set, and a set of implicit root cause network nodes-fault type labels is output. The step S500 of the embodiment of the application further includes:

[0063] Based on the detection time point of the burst representation feature vector, the embedded terminal node data buffer area in each multi-source causal clue path in the multi-source causal clue path set is backtracked according to a preset backtracking window, and a path backtracking history data sequence cluster is obtained;

[0064] The path backtracking history data sequence cluster is traversed to identify an implicit fault rate, and an implicit root cause network node set is determined according to the identification result to generate the set of implicit root cause network nodes-fault type labels.

[0065] Further, the step S500 of the embodiment of the application further includes:

[0066] A first path backtracking history data sequence is extracted from the path backtracking history data sequence cluster;

[0067] perform long-short fluctuation trend analysis on the first path backtracking history data sequence to generate a first long fluctuation trend feature vector and a first short fluctuation trend feature vector;

[0068] perform feature enhancement on the first short fluctuation trend feature vector based on the first long fluctuation trend feature vector to generate a first enhanced fluctuation trend feature vector, and identify a failure rate according to the first enhanced fluctuation trend feature vector to obtain a first failure rate;

[0069] In this way, the failure rate is identified by traversing the path backtracking history data sequence cluster, and a failure rate greater than or equal to a preset failure rate threshold is taken as an implied failure rate, and a node corresponding to the failure rate is taken as an implied root cause network node set.

[0070] The failure type identification is performed in combination with an enhanced fluctuation trend feature vector corresponding to each implied root cause network node to generate an implied root cause network node-failure type label set.

[0071] In an embodiment of the present application, the preset backtracking window is a time period for historical data backtracking preset by a person skilled in the art, which can be 5s, 15s, etc. The detection time point of the burst representation feature vector is obtained, and the length of the preset backtracking window is added to the detection time point to obtain a time window for historical data backtracking. The data stored in the time window in the embedded terminal node data buffer area in each multi-source causal clue path in the multi-source causal clue path set is subjected to historical data backtracking to obtain a path backtracking history data sequence cluster. The path backtracking history data sequence cluster includes backtracking data of different key nodes in each multi-source causal clue path in the multi-source causal clue path set, and includes a plurality of path backtracking history data sequence sets, each path backtracking history data sequence set corresponding to one multi-source causal clue path. Each path backtracking history data sequence reflects the data change of a key node in the time window.

[0072] In an embodiment, the path backtracking history data sequence cluster is taken as analysis data to analyze the implied failure in the multi-source causal clue path to obtain an identification result. The identification result contains the existence of an implied failure of a key node in each multi-source causal clue path. A failure rate greater than or equal to a preset failure rate threshold (the minimum failure rate at which an implied failure exists, preset by a person skilled in the art) is taken as an implied failure rate, and a node corresponding to the failure rate is taken as an implied root cause network node set. Furthermore, the failure type identification is performed in combination with an enhanced fluctuation trend feature vector corresponding to each implied root cause network node to generate an implied root cause network node-failure type label set.

[0073] Through path tracing and feature analysis, it is possible to identify those fault sources that are not obvious, achieving the technical effect of analyzing faults from multiple angles and ensuring the accuracy and reliability of fault identification and positioning.

[0074] Preferably, a first path backtracking historical data sequence is extracted from the path backtracking historical data sequence cluster, and then a long-short fluctuation trend analysis is performed on the first path backtracking historical data sequence to generate a first long fluctuation trend feature vector and a first short fluctuation trend feature vector. The first long fluctuation trend feature vector is a spatial feature vector that globally reflects the first path backtracking historical data sequence using a larger analysis scale. The first short fluctuation trend feature vector is a temporal feature vector obtained by performing a local feature analysis on the first path backtracking historical data sequence using a smaller analysis scale.

[0075] The spatial feature vector, that is, the first long fluctuation trend feature vector, is used to enhance the features of the first short fluctuation trend feature vector, thereby achieving the goal of iterating the overall fluctuation trend and the local fluctuation trend, thereby obtaining the first enhanced fluctuation trend feature vector that can fully reflect the fluctuation situation of the first path backtracing historical data sequence.

[0076] Preferably, a fault identification rate network layer is obtained, and the fault identification rate network layer is used to identify the fault rate of the first enhanced fluctuation trend feature vector to obtain the first fault rate. A plurality of sample first enhanced fluctuation trend feature vectors and a plurality of sample first failure rates are obtained as training data, and the training data are used to perform supervised training on a framework constructed based on a feedforward neural network, and the network parameters of the framework are continuously iteratively updated during training until convergence, thereby obtaining the trained fault identification rate network layer. Based on the same principle of obtaining the implicit failure rate, the path is traversed back to the historical data sequence cluster to identify the failure rate, and the failure rate greater than or equal to the preset failure rate threshold is used as the implicit failure rate, and its corresponding node is used as the implicit root cause network node set.

[0077] A fault type sample table constructed by those skilled in the art based on experience is obtained as shown in Table 2, wherein each fault type sample has a corresponding fault trend feature vector identifier.

[0078] Table 2 Fault type sample table

[0079]

[0080] Using the cosine calculation formula, the enhanced fluctuation trend feature vector corresponding to each implicit root cause network node is matched with the fault trend feature vector in the fault type sample table for similarity, and the fault type corresponding to the maximum matching similarity is used as the type recognition result to generate the implicit root cause network node-fault type label set.

[0081] Further, long-short fluctuation trend analysis is performed on the first path backtracking history data sequence to generate a first long fluctuation trend feature vector and a first short fluctuation trend feature vector. The step S500 of the embodiment of the present application further includes:

[0082] Extracting an abnormal duration set of the power distribution line in the historical time;

[0083] Taking the maximum value in the abnormal duration set as a first identification scale;

[0084] Taking the minimum value in the abnormal duration set as a second identification scale;

[0085] Performing feature extraction on the first path backtracking history data sequence according to the first identification scale and the second identification scale respectively to obtain the first long fluctuation trend feature vector and the first short fluctuation trend feature vector.

[0086] Further, feature enhancement is performed on the first short fluctuation trend feature vector based on the first long fluctuation trend feature vector to generate a first enhanced fluctuation trend feature vector. The step S500 of the embodiment of the present application further includes:

[0087] Performing mapping element similarity identification on the first long fluctuation trend feature vector and the first short fluctuation trend feature vector, and filling the identification result into an initially empty matrix to construct a feature enhancement matrix;

[0088] Performing interactive convolution on the first short fluctuation trend feature vector by using the feature enhancement matrix to generate the first enhanced fluctuation trend feature vector.

[0089] In one possible embodiment, the abnormal duration set refers to a set of all abnormal durations recorded in the backtracking history data process. For example, a certain device is in an abnormal current or voltage state for a long time, and the system records the duration of these abnormalities. The identification scale refers to the window range during data analysis, which is used to determine the duration of interest. Taking the maximum value as the first identification scale to focus on long-term trends, the first identification scale is used to identify long-term failure signs of the device (such as gradual increase of load and gradual rise of current, etc.). Taking the minimum value as the second identification scale to focus on short-term abnormalities. The second identification scale is used to help the system capture sudden fluctuations of current and voltage, and to identify transient faults in time.

[0090] Preferably, a plurality of sample path backtracking history data sequences are obtained, and a plurality of sample first long-term fluctuation trend feature vectors and a plurality of sample first short-term fluctuation trend feature vectors obtained by feature extraction using a first identification scale and a second identification scale are used as a training sample set. The framework based on the feedforward neural network is supervised trained using the plurality of sample path backtracking history data sequences and the plurality of sample first long-term fluctuation trend feature vectors in the training sample set, to learn the mapping relationship between the path backtracking history data sequence and the long-term fluctuation trend feature vector, until the training converges, to obtain a trained first feature extractor. Based on the same principle, the second feature extractor is obtained by training based on the plurality of sample path backtracking history data sequences and the plurality of sample first short-term fluctuation trend feature vectors in the training sample set. The first long-term fluctuation trend feature vector and the first short-term fluctuation trend feature vector are obtained by feature extraction using the first feature extractor and the second feature extractor on the first path backtracking history data sequence, respectively.

[0091] In one possible embodiment, the cosine similarity calculation formula is used to identify the mapping similarity of the same type of elements in the first long-term fluctuation trend feature vector and the first short-term fluctuation trend feature vector, and the identification result is normalized using a normalization formula, and then filled into an initially empty matrix to construct a feature enhancement matrix. The first enhanced fluctuation trend feature vector is generated by using the feature enhancement matrix to interactively convolve the first short-term fluctuation trend feature vector.

[0092] Preferably, each element of the feature enhancement matrix is multiplied by the corresponding element in the first short-term fluctuation trend feature vector to complete the interactive convolution, obtain a weighted sum, and compose the first enhanced fluctuation trend feature vector. For example, if a feature enhancement matrix element represents the relationship between current fluctuation and voltage fluctuation, it will be multiplied by the current or voltage value in the short-term fluctuation feature vector, thereby weighting the influence of the short-term fluctuation feature.

[0093] Through the interactive convolution, the system can combine the long-term and short-term feature information to accurately identify the short-term fluctuation when the fault occurs. This enhancement can improve the sensitivity and accuracy of fault detection, especially when facing complex fault patterns.

[0094] In summary, the embodiments of the present application have at least the following technical effects:

[0095] By deploying embedded terminals at key nodes of distribution lines, and combining historical operation fault data and causal clue analysis of the distribution line knowledge base, the present application can realize real-time fault monitoring and positioning, and through steps such as multi-source causal clue graph, sudden representation feature detection, and chain root cause backtracking, the fault type can be accurately identified and the hidden fault root cause can be traced, thereby achieving the technical effects of improving fault diagnosis accuracy and fault positioning reliability.

[0096] Embodiment two, based on the same inventive concept as the embedded terminal-based power distribution line fault identification and positioning method in the preceding embodiment, as Figure 2 shown, the application provides an embedded terminal-based power distribution line fault identification and positioning system, and the system and method embodiments in the application are based on the same inventive concept. The system comprises:

[0097] A distributed network obtaining module 11 is configured to deploy embedded terminals at a set of key nodes of a power distribution line and obtain a distributed sensing network, wherein each network node in the distributed sensing network corresponds to an embedded terminal.

[0098] A clue map library construction module 12 is configured to obtain a set of historical operation fault data and a power distribution line knowledge base, perform causal clue analysis, and construct a multi-source causal clue map.

[0099] A burst characterization feature vector obtaining module 13 is configured to traverse each embedded terminal in the distributed sensing network to detect burst characterization features and obtain a burst characterization feature vector.

[0100] A clue path set generation module 14 is configured to perform causal clue path retrieval in the multi-source causal clue map based on the burst characterization feature vector and generate a multi-source causal clue path set.

[0101] A label set obtaining module 15 is configured to obtain a chain root cause backtracking instruction, perform window backtracking on the distributed sensing network based on the multi-source causal clue path set, and output a set of hidden root cause network node-fault type labels.

[0102] A fault identification and positioning result obtaining module 16 is configured to take the key nodes corresponding to the burst characterization feature vector and the key nodes corresponding to the set of hidden root cause network node-fault type labels as fault identification and positioning results.

[0103] Further, each embedded terminal has a node data buffer area, wherein the node data buffer area supports backtracking of historical data.

[0104] Further, the label set obtaining module 15 is configured to perform the following steps:

[0105] Based on the detection time point of the burst characterization feature vector, historical data in the embedded terminal node data buffer area within each multi-source causal clue path in the multi-source causal clue path set is backtracked according to a preset backtracking window, and a path backtracking historical data sequence cluster is obtained.

[0106] The path backtracking history data sequence cluster is traversed to identify an implicit fault rate, and an implicit root cause network node set is determined according to an identification result, and a node-fault type label set of the implicit root cause network is generated.

[0107] Further, the label set obtaining module 15 is configured to perform the following steps:

[0108] A first path backtracking history data sequence is extracted from the path backtracking history data sequence cluster;

[0109] Long and short fluctuation trend analysis is performed on the first path backtracking history data sequence to generate a first long fluctuation trend feature vector and a first short fluctuation trend feature vector;

[0110] Feature enhancement is performed on the first short fluctuation trend feature vector based on the first long fluctuation trend feature vector to generate a first enhanced fluctuation trend feature vector, and a fault rate is identified according to the first enhanced fluctuation trend feature vector to obtain a first fault rate;

[0111] By analogy, the path backtracking history data sequence cluster is traversed to identify a fault rate, and a fault rate greater than or equal to a preset fault rate threshold is taken as an implicit fault rate, and a node corresponding to the fault rate is taken as an implicit root cause network node set;

[0112] A fault type is identified in combination with an enhanced fluctuation trend feature vector corresponding to each implicit root cause network node to generate a node-fault type label set of the implicit root cause network.

[0113] Further, the label set obtaining module 15 is configured to perform the following steps:

[0114] A set of abnormal duration lengths of the power distribution line in the historical time is extracted;

[0115] A maximum value in the set of abnormal duration lengths is taken as a first identification scale;

[0116] A minimum value in the set of abnormal duration lengths is taken as a second identification scale;

[0117] Feature extraction is performed on the first path backtracking history data sequence according to the first identification scale and the second identification scale respectively to obtain the first long fluctuation trend feature vector and the first short fluctuation trend feature vector.

[0118] Further, the label set obtaining module 15 is configured to perform the following steps:

[0119] Mapping element similarity is identified for the first long fluctuation trend feature vector and the first short fluctuation trend feature vector, and an identification result is filled into an initially empty matrix to construct a feature enhancement matrix;

[0120] interactively convolve the first short-wave fluctuation trend feature vector with the feature enhancement matrix to generate the first enhanced fluctuation trend feature vector.

[0121] Further, the clue graph library construction module 12 is configured to perform the following steps:

[0122] extracting connection relationships between key nodes of the power distribution line, breaker action logic, and line abnormality rules in the power distribution line knowledge base;

[0123] combining the connection relationships between key nodes of the power distribution line, the breaker action logic, and the line abnormality rules, performing event extraction and pairing on the time sequence records in the historical operation fault data set to obtain a set of cause-effect event pairs;

[0124] de-duplicating the set of cause-effect event pairs to obtain a set of cleaned cause-effect event pairs;

[0125] constructing the multi-source cause-effect clue graph based on the set of cleaned cause-effect event pairs.

[0126] Further, each cleaned cause-effect event pair in the set of cleaned cause-effect event pairs is taken as an edge in the graph, and a fault feature vector corresponding to each cleaned cause-effect event pair is taken as a node in the graph, to construct the multi-source cause-effect clue graph.

[0127] Further, the clue graph library construction module 12 is configured to perform the following steps:

[0128] extracting the time sequence records in the historical operation fault data set to perform structured conversion to obtain a set of historical operation fault structured sequences;

[0129] performing retrieval on reasons in the set of historical operation fault structured sequences based on the connection relationships between key nodes of the power distribution line, the breaker action logic, and the line abnormality rules to generate a set of cause-effect event pairs.

[0130] It should be noted that the above sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. Moreover, the above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0131] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0132] The specification and drawings are, of course, to be regarded in an illustrative rather than a restrictive sense. It is to be understood that any such modifications, variations, combinations or equivalents that fall within the scope of the application are intended to be embraced herein.

Claims

1. A method for identifying and locating power distribution line faults based on an embedded terminal, characterized in that: The method comprises: Embedded terminals are deployed at key nodes of the distribution line to obtain a distributed sensing network, where each network node in the distributed sensing network corresponds to an embedded terminal; Obtain historical operation fault data sets and distribution line knowledge base to sort out causal clues and construct a multi-source causal clue map; Traversing each embedded terminal in the distributed sensing network to perform burst characterization feature detection to obtain a burst characterization feature vector; Performing causal clue path retrieval in the multi-source causal clue graph based on the burst representation feature vector to generate a multi-source causal clue path set; Obtaining a chain root cause backtracking instruction, performing window backtracking on the distributed perception network according to the multi-source causal clue path set, and outputting an implicit root cause network node-fault type label set; The key nodes corresponding to the burst characterization feature vector and the key nodes corresponding to the implicit root cause network node-fault type label set are used as fault identification and positioning results; The chain root cause backtracking instruction is obtained, and the distributed perception network is windowed backtracked according to the multi-source causal clue path set, and a set of implicit root cause network nodes and fault type labels is output, including: Based on the detection time point of the burst characterization feature vector, historical data backtracking is performed on the embedded terminal node data cache area in each multi-source causal clue path in the multi-source causal clue path set according to a preset backtracking window to obtain a path backtracking historical data sequence cluster; Traversing the path backtracking historical data sequence cluster to identify implicit fault rates, and determining an implicit root cause network node set based on the identification result, and generating the implicit root cause network node-fault type label set; Extracting a first path backtracking historical data sequence from the path backtracking historical data sequence cluster; Performing a long-short fluctuation trend analysis on the first path backtracking historical data sequence to generate a first long-short fluctuation trend feature vector and a first short-short fluctuation trend feature vector; performing feature enhancement on the first short fluctuation trend feature vector based on the first long fluctuation trend feature vector to generate a first enhanced fluctuation trend feature vector, and identifying a failure rate based on the first enhanced fluctuation trend feature vector to obtain a first failure rate; Similarly, the path is traversed back through the historical data sequence cluster to identify the failure rate, and the failure rate with a failure rate greater than or equal to the preset failure rate threshold is taken as the implicit failure rate, and the corresponding node is taken as the implicit root cause network node set; Fault type identification is performed in combination with the enhanced fluctuation trend feature vector corresponding to each implicit root cause network node to generate the implicit root cause network node-fault type label set.

2. The method for identifying and locating a power distribution line fault based on an embedded terminal according to claim 1, wherein: Each embedded terminal has a node data cache area, wherein the node data cache area supports backtracking historical data.

3. The method for identifying and locating a power distribution line fault based on an embedded terminal according to claim 1, wherein: Performing a long-short fluctuation trend analysis on the first path backtracking historical data sequence to generate a first long-short fluctuation trend feature vector and a first short-short fluctuation trend feature vector includes: Extract the abnormal duration of distribution lines in historical time; Taking the maximum value in the abnormal duration set as the first identification metric; Taking the minimum value in the abnormal duration set as the second identification metric; Feature extraction is performed on the first path backtracking historical data sequence according to a first identification scale and a second identification scale respectively to obtain the first long fluctuation trend feature vector and the first short fluctuation trend feature vector.

4. The method for identifying and locating a power distribution line fault based on an embedded terminal according to claim 1, wherein: The method of performing feature enhancement on the first short fluctuation trend feature vector based on the first long fluctuation trend feature vector to generate a first enhanced fluctuation trend feature vector includes: Performing mapping element similarity recognition on the first long fluctuation trend feature vector and the first short fluctuation trend feature vector, and filling the recognition results into an initially empty matrix to construct a feature enhancement matrix; The first short fluctuation trend feature vector is interactively convolved using the feature enhancement matrix to generate the first enhanced fluctuation trend feature vector.

5. The method for identifying and locating a distribution line fault based on an embedded terminal according to claim 1, wherein: Obtain historical operation fault data sets and distribution line knowledge base to sort out causal clues and build a multi-source causal clue map, including: Extract the connection relationship between key nodes of distribution lines, circuit breaker action logic and line abnormality rules from the distribution line knowledge base; Combined with the connection relationship between key nodes of the distribution line, the circuit breaker operation logic and the line abnormality rules, event extraction and pairing are performed on the time series records in the historical operation fault data set to obtain a cause-effect event pair set; Deduplication is performed on the cause-effect event pair set to obtain a cleaned cause-effect event pair set; The multi-source causal clue map is constructed based on the cleaned cause-effect event pair set.

6. The method for identifying and locating a distribution line fault based on an embedded terminal according to claim 5, wherein: Each cleaning cause-effect event pair in the cleaning cause-effect event pair set is used as an edge in a graph, and the fault feature vector corresponding to each cleaning cause-effect event pair is used as a node in the graph to construct the multi-source causal clue graph.

7. The method for identifying and locating a distribution line fault based on an embedded terminal according to claim 5, wherein: include: Extracting the time series records in the historical operation fault data set and performing structured conversion to obtain a historical operation fault structured sequence set; Based on the connection relationship between the key nodes of the distribution line, the circuit breaker action logic and the line abnormality rules, the causes in the historical operation fault structured sequence set are retrieved to generate a cause-effect event pair set.

8. The distribution line fault identification and positioning system based on embedded terminal is characterized by: A method for identifying and locating a distribution line fault based on an embedded terminal according to any one of claims 1 to 7 is implemented, the system comprising: A distributed network acquisition module is used to deploy embedded terminals at key node sets of the distribution line to obtain a distributed perception network, wherein each network node in the distributed perception network corresponds to an embedded terminal; The clue map library construction module is used to obtain the historical operation fault data set and the distribution line knowledge base to sort out the causal clues and build a multi-source causal clue map; A burst characterization feature vector acquisition module is used to traverse each embedded terminal in the distributed sensing network to perform burst characterization feature detection and obtain a burst characterization feature vector; A clue path set generation module, configured to perform causal clue path retrieval in the multi-source causal clue graph based on the burst representation feature vector, and generate a multi-source causal clue path set; A label set acquisition module is used to obtain a chain root cause backtracking instruction, perform window backtracking on the distributed perception network according to the multi-source causal clue path set, and output an implicit root cause network node-fault type label set; The fault identification and positioning result obtaining module is used to take the key nodes corresponding to the burst characterization feature vector and the key nodes corresponding to the implicit root cause network node-fault type label set as the fault identification and positioning results.

Images