Fault positioning method and system for high-voltage transmission line

Abstract

The invention discloses a fault positioning method and system for a high-voltage transmission line, and relates to the technical field of fault positioning, and the method comprises the steps: connecting a fault positioning system, and extracting a fault positioning log; tracing and analyzing the fault positioning log to obtain a positioning response module; carrying out the difference identification of the fault positioning precision, and obtaining a fault positioning target compensation parameter; decomposing the execution model to obtain weak parameters and missing parameters; and carrying out retention layer extraction on the execution model, building a transfer learning framework, carrying out processing layer training, connecting the processing layer obtained by training with the retention layer, and replacing the execution model. According to the method, the technical problems of low fault positioning precision of the high-voltage transmission line and insufficient parameter processing and model optimization means in the prior art are solved, and the technical effects of optimizing the execution model of the fault positioning system and effectively improving the fault positioning precision are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault location, and in particular to a method and system for locating a fault of a high-voltage transmission line. Background Art

[0002] At present, the fault location technology of high-voltage transmission lines faces severe challenges. On the one hand, the existing fault location execution model is difficult to accurately adapt to the complex working conditions of the line. When faced with special situations such as the superposition of complex fault types and multi-point faults, the positioning accuracy is seriously insufficient, resulting in a lot of time spent on fault troubleshooting and repair work, which greatly affects the reliability of power supply and causes huge economic losses. On the other hand, the analysis methods of fault location logs are limited, and it is impossible to fully explore the key information such as fault characteristics and line parameter changes contained therein, making it difficult to effectively guide the optimization of the execution model. In addition, when dealing with various parameters that affect the accuracy of fault location, there is a lack of efficient identification and compensation mechanisms, making it difficult to make targeted improvements to the execution model, which seriously restricts the development and application of high-voltage transmission line fault location technology.

[0003] The existing technology has technical problems such as low accuracy of high-voltage transmission line fault location and insufficient parameter processing and model optimization means. Summary of the invention

[0004] The present application provides a method and system for locating a fault of a high-voltage transmission line, which is used to solve the technical problems of low accuracy of fault locating of high-voltage transmission lines, insufficient parameter processing and model optimization means in the prior art.

[0005] In view of the above problems, the present application provides a method and system for locating a fault in a high-voltage transmission line.

[0006] In a first aspect of the present application, a method for locating a fault of a high-voltage transmission line is provided, the method comprising: Connect to the fault location system, extract the fault location log, the fault location log has a fault location evaluation result; according to the fault location evaluation result, perform traceability analysis on the fault location log, obtain the location response module, and extract the execution model of the location response module; perform differential identification of fault location accuracy on the fault location evaluation result, and obtain the fault location target compensation parameter; take the fault location target compensation parameter as the target, decompose the execution model to obtain weak parameters and missing parameters, and the weak parameters are parameters whose fault location accuracy does not meet the requirements; perform reserved layer extraction on the execution model according to the weak parameters and the missing parameters, build a transfer learning framework based on the reserved layer and the weak parameters and the missing parameters, perform processing layer training on the weak parameters and the missing parameters through the transfer learning framework, connect the training obtained processing layer with the reserved layer, and replace the execution model.

[0007] A second aspect of the present application provides a fault location system for a high voltage transmission line, the system comprising: A fault location log extraction unit is used to connect to a fault location system and extract a fault location log, wherein the fault location log has a fault location evaluation result; a tracing and parsing unit is used to perform tracing and parsing of the fault location log according to the fault location evaluation result, obtain a location response module, and extract an execution model of the location response module; a fault location target compensation parameter acquisition unit is used to perform differential identification of fault location accuracy on the fault location evaluation result and obtain a fault location target compensation parameter; an execution model decomposition unit is used to obtain weak parameters and missing parameters by decomposing the execution model with the fault location target compensation parameter as the target, wherein the weak parameters are parameters whose fault location accuracy does not meet the requirements; an execution model replacement unit is used to perform reserved layer extraction on the execution model according to the weak parameters and the missing parameters, and to build a transfer learning framework based on the reserved layer and the weak parameters and the missing parameters, and to perform processing layer training on the weak parameters and the missing parameters through the transfer learning framework, and to connect the training obtained processing layer with the reserved layer to replace the execution model.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: Connect to the fault location system and extract the fault location log; according to the fault location evaluation result, perform traceability analysis on the fault location log, obtain the location response module, and extract the execution model of the location response module; perform differential identification of fault location accuracy and obtain the fault location target compensation parameter; decompose the execution model to obtain weak parameters and missing parameters; extract the reserved layer of the execution model according to the weak parameters and missing parameters, build a transfer learning framework based on the reserved layer and weak parameters and missing parameters, train the processing layer of the weak parameters and missing parameters through the transfer learning framework, connect the training obtained processing layer with the reserved layer, and replace the execution model. The technical effect of optimizing the execution model of the fault location system and effectively improving the fault location accuracy is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a flow chart of a method for locating a fault in a high-voltage transmission line provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a fault location system for a high-voltage transmission line provided in an embodiment of the present application; Description of reference numerals: fault location log extraction unit 10 , tracing analysis unit 20 , fault location target compensation parameter acquisition unit 30 , execution model decomposition unit 40 , execution model replacement unit 50 . DETAILED DESCRIPTION

[0011] The present application provides a method and system for locating a fault in a high-voltage transmission line, aiming to solve the technical problems in the prior art of low accuracy in locating faults in high-voltage transmission lines, and insufficient means for parameter processing and model optimization.

[0012] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0013] Example 1: Figure 1 As shown, the present application provides a method for locating a fault of a high-voltage transmission line, the method comprising: Step S100: connecting to a fault location system and extracting a fault location log, wherein the fault location log has a fault location evaluation result.

[0014] Specifically, when the fault location work of the high-voltage transmission line is started, the terminal used for data analysis and processing is stably connected to the fault location system using special connection equipment and adapted communication protocols. The fault location system runs in high-voltage transmission lines for a long time, continuously monitoring the line status and recording various types of data. After successful connection, the fault location log is exported from the storage module of the fault location system through the preset log extraction instructions and data transmission channels inside the system. These logs record the whole process information of each fault location operation in detail, including the crucial fault location evaluation results, which are generated according to certain evaluation standards and algorithms, and comprehensively reflect the key indicators such as the accuracy and reliability of the fault location, providing basic data support for the subsequent in-depth analysis of the fault location process, finding the cause of the location error, and optimizing the location strategy.

[0015] Step S200: According to the fault location evaluation result, the fault location log is retroactively analyzed to obtain a location response module and extract an execution model of the location response module.

[0016] Specifically, after obtaining the fault location log and the fault location evaluation results it contains, we start to conduct in-depth traceability analysis of the fault location log. First, we conduct a time series analysis on each log, sort out the processing data chain of each link of the fault location in chronological order, and clearly present the process details of the positioning operation. Compare the fault location evaluation results with the preset fault location accuracy requirements, accurately find out the differences between the two, and form deviation information. Take these deviation information as key clues to conduct correlation mining on the processing data chain. By analyzing the deviation parameters in the deviation information and performing correlation analysis with the help of historical data, we can find the key factors affecting the fault location accuracy, namely the influencing factors and the influence relationship between them. With the influencing factors as indexes, we search for relevant participating nodes in the processing data chain, calculate the error influence value of each participating node on the fault location accuracy according to the influence relationship, and locate the participating nodes according to the time series relationship of the processing data chain, and accumulate the error influence values ​​according to the time series. When the cumulative amount reaches the deviation amount in the deviation information, the execution model can be accurately located, and then the corresponding positioning response module can be determined. This module plays a key role in the fault location process, and its execution model is the core basis for subsequent optimization and improvement of the fault location method.

[0017] Step S300: performing fault location accuracy difference identification on the fault location evaluation result to obtain a fault location target compensation parameter.

[0018] Specifically, a fault location accuracy evaluation model is constructed. This model will establish a multi-dimensional benchmark data set based on historical fault location data, line parameters, environmental factors, etc. For example, accurate fault location data for different lines and different environmental conditions in the past period of time, as well as corresponding line length, tower distribution, weather conditions and other information are collected. When a new fault location evaluation result is obtained, key data is extracted from the result, such as the coordinates of the actual fault location, the time spent on positioning, etc. These data are compared with the ideal fault location data of the same type of line and similar environmental conditions in the benchmark data set. For example, if the ideal fault location error range of a certain line in the benchmark data set is within ±50 meters under specific weather conditions, and the current fault location evaluation result shows that the actual location error is 80 meters. The difference calculation algorithm is used to calculate the difference between the actual location result and the ideal location result. Taking the location error distance as an example, the actual error distance is directly subtracted from the upper limit of the ideal error distance (if the actual error is greater than the upper limit of the ideal error) to obtain the positive difference; if the actual error is less than the lower limit of the ideal error, the lower limit of the ideal error is subtracted from the actual error distance to obtain the negative difference. In the above example, the forward difference is 80-50=30 meters. Then, the difference is analyzed by a machine learning algorithm. For example, a linear regression model is used, with the difference as the independent variable and the adjustment parameters that have been manually adjusted or optimized to make the positioning more accurate in the past as the dependent variable for model training. After the training is completed, the currently calculated difference is input into the trained model, and the result of the model output is the fault location target compensation parameter. These parameters may include adjustment values ​​for certain weight coefficients in the positioning algorithm, correction coefficients for sensor data, etc. In this way, the fault location target compensation parameters used to improve the fault location accuracy can be accurately obtained based on the fault location evaluation results.

[0019] Step S400: Taking the fault location target compensation parameter as a target, the execution model is decomposed to obtain weak parameters and missing parameters, wherein the weak parameters are parameters whose fault location accuracy does not meet the requirement.

[0020] Specifically, after obtaining the fault location target compensation parameters, the fault location target compensation parameters are used as key guidance to conduct an in-depth analysis of the execution model of the positioning response module. The execution model is a complex system that contains many parameters that affect the fault location accuracy. Using specialized model decomposition tools and techniques, the execution model is gradually disassembled according to its functional structure and parameter characteristics. During the disassembly process, the role of each parameter in the positioning process is evaluated according to the pre-set fault location accuracy standard. Those parameters that cause the fault location accuracy to fail to meet the expected requirements are identified as weak parameters. For example, if a sensor parameter is used in the model to measure the line current to assist in positioning, but due to its insufficient accuracy or deviation, the positioning result based on the parameter deviates from the actual fault location, this sensor parameter is a weak parameter. At the same time, check whether there are missing necessary parameters in the model. These parameters are indispensable to ensure the fault location accuracy, but they are not considered or set in the current execution model. Through comprehensive and detailed analysis and screening, the weak and missing parameters in the execution model are finally accurately identified, laying the foundation for the subsequent targeted optimization of the execution model.

[0021] Step S500: extract the reserved layer of the execution model according to the weak parameters and the missing parameters, build a transfer learning framework based on the reserved layer and the weak parameters and the missing parameters, train the processing layer of the weak parameters and the missing parameters through the transfer learning framework, connect the processing layer obtained by training with the reserved layer, and replace the execution model.

[0022] Specifically, after identifying the weak parameters and missing parameters in the execution model, the execution model is extracted from the reserved layers. With the help of professional model analysis tools, according to the structure and functional characteristics of the model, the parts of the model that are stable and still of great value for fault location are selected as the reserved layers. These reserved layers will serve as the basic framework for the subsequent optimization model. For example, some layers used for data preprocessing and preliminary feature extraction are universal and stable when processing high-voltage transmission line data. Even if the model parameters change, these layers can still play an effective role, so they are retained. Based on the reserved layers and the identified weak parameters and missing parameters, a transfer learning framework is built. In this framework, relevant transfer learning algorithms and tools are used to build corresponding processing layers for processing weak parameters and missing parameters. For weak parameters, according to their characteristics and problems, appropriate optimization algorithms are selected, such as using the gradient descent algorithm to adjust the processing layers corresponding to the weak parameters, and gradually improve their contribution to the fault location accuracy through multiple iterative training; for missing parameters, a special processing layer is rebuilt, and suitable data are selected from a large amount of historical data and other data sets related to high-voltage transmission line fault location as training samples to train the missing parameter processing layer. After sufficient training, the processing layer for weak and missing parameters is connected to the previously extracted retention layer. The connection process strictly follows the model structure and data flow requirements to ensure that the newly connected processing layer can work with the retention layer. After the connection is completed, the original execution model is replaced as a whole with the newly constructed model. The new execution model incorporates the optimization of weak parameters and the supplementary processing of missing parameters. It can process data more accurately and improve the positioning accuracy during the fault location process, effectively improving the performance of the high-voltage transmission line fault location system.

[0023] In a possible implementation, step S200 further includes: Step S210: Perform time series analysis on each fault location log to obtain a processing data chain of each fault location evaluation result; Step S220: matching the fault location evaluation result with the fault location accuracy requirement to obtain deviation information; Step S230: using the deviation information as an identification index, performing association mining according to the processing data chain, tracing back the execution model, and obtaining the positioning response module.

[0024] Specifically, for each fault location log, the time series analysis method is used to sort out the data in the log in the chronological order of the log. Starting from the starting time of the fault, the operation records, measurement data, and various intermediate processing results related to the fault location are sorted out in sequence, and these data are connected in chronological order to form a processing data chain of each fault location evaluation result. This processing data chain fully presents the data changes from the beginning of fault location to the evaluation result.

[0025] The obtained fault location evaluation results are compared with the pre-set fault location accuracy requirements. The fault location accuracy requirements are determined based on the operating standards, safety specifications, and actual engineering experience of high-voltage transmission lines, and have clear quantitative indicators. By comparing the actual fault location results with the ideal accuracy requirements, the differences between the two are found, and these differences constitute deviation information. For example, if the ideal fault location error should be within 50 meters, and the actual location error reaches 80 meters, then the excess 30 meters and the related location result information are deviation information.

[0026] The obtained deviation information is used as the key identification index, and the association mining is carried out in combination with the previously generated processing data chain. According to the key features in the deviation information, such as the type, size, and time point of the deviation, the relevant data nodes and operation links are searched in the processing data chain. By analyzing the inherent connection between these related data, the execution model that plays a key role in the fault location process is traced. With the deepening of mining, the various factors and links involved in the fault location decision and calculation are gradually determined, and then the location response module responsible for response and processing in the entire fault location process is obtained, which provides an important basis for the subsequent in-depth analysis of the fault location mechanism and the optimization of the location method.

[0027] In a possible implementation, step S230 further includes: Step S231: Based on the deviation parameters of the deviation information, correlation analysis is performed through historical data to obtain influencing factors and influencing relationships; Step S232: using the impact factor as an index to search in the processing data chain to obtain participating nodes; Step S233: performing influence calculation on the participating nodes according to the influence relationship to obtain an error influence value; Step S234: positioning the participating nodes based on the processing data link timing relationship, accumulating and calculating the error impact value according to the timing relationship, and locating the execution model when the accumulated amount reaches the deviation amount of the deviation information.

[0028] Specifically, deviation parameters are extracted from the deviation information, such as the distance deviation value and angle deviation value of fault location. With these deviation parameters as the core, correlation analysis is performed in a database that stores a large amount of historical fault location data. By calculating the degree of correlation between each factor in the historical data and the deviation parameters, those factors that have a significant impact on the deviation are screened out. These factors are the influencing factors, and the mutual influence relationship between them and the deviation parameters is determined. For example, a specific sensor measurement error is positively correlated with the fault location distance deviation, and the quantitative relationship of the degree of influence is also determined.

[0029] After obtaining the influencing factors, the processing data chain records in detail the entire process data from data collection, transmission, processing to the final positioning result output during the fault location process. Taking the influencing factor as the key identifier of the search, each data node is checked and compared in turn along the processing data chain. For example, if the influencing factor is the measurement deviation of a specific model of sensor, the processing data chain contains data collection and processing nodes of multiple sensors. At this time, these nodes are checked one by one to find data records related to the influencing factor. Once it is found that the key information such as the sensor model, measurement time, and measurement environment involved in a data node matches the influencing factor, or the data processing logic of the node is related to the influencing factor, then this data node is identified as a participating node. Through this comprehensive and detailed search method, all participating nodes related to the influencing factor are accurately screened out from the processing data chain, laying the foundation for the subsequent analysis of the impact of these nodes on fault location.

[0030] After determining the participating nodes, according to the influence relationship, these influence relationships are determined based on the historical data correlation analysis, reflecting the intrinsic connection between the influence factor and the fault location deviation, usually presented in the form of a mathematical model or a quantitative relationship. For each participating node, according to its corresponding influence factor, it is substituted into the influence relationship model for calculation. Assume that the influence relationship shows that for every unit change of a certain influence factor, the fault location error will change by a specific value in a certain direction, and the actual change of the influence factor corresponding to the participating node is known. For example, the influence relationship is that for every increase of influence factor A by 1, the fault location distance error increases by 3 meters. At a participating node, the influence factor A increases by 2. Then, according to this relationship, the error impact value of the participating node on the fault location distance is calculated to be 6 meters. If there are multiple influence factors related to the participating node, they are calculated separately according to their respective influence relationships, and then comprehensively superimposed according to the actual situation. By performing such calculations on each participating node, the error impact values ​​of each participating node are fully obtained. These values ​​intuitively show the role played by each participating node in causing the fault location deviation, providing a quantitative basis for the subsequent traceability execution model.

[0031] After obtaining the error impact value of the participating nodes, further analysis is carried out based on the timing relationship of the processing data chain. The processing data chain records the changes of various types of data in the time sequence of the fault location process. Each participating node has its corresponding timestamp on this data chain, reflecting the order in which it appears in the fault location process. Using a timing model similar to the Markov chain, starting from the beginning of the processing data chain, the participating nodes are located in turn according to their time sequence. In this process, the error impact value of each participating node is accumulated and calculated in time sequence. With the Markov chain, it is assumed that the error impact value of the current participating node is not only related to itself, but also affected by the state of the previous participating node (in actual situations, that is, the previous data processing link may affect the contribution of the current link to the fault location error). As the calculation proceeds, the error impact value is continuously accumulated. This accumulated amount is continuously monitored and compared with the deviation amount in the deviation information. The deviation amount in the deviation information is obtained by matching the fault location evaluation result with the fault location accuracy requirement, and represents the gap between the actual fault location result and the ideal result. When the accumulated error impact value reaches the deviation amount, it means that the key link combination that causes the current fault location deviation has been found, and these key links constitute the core part of the execution model, thereby realizing the positioning of the execution model. This means that through the analysis of the timing relationship of the processing data chain and the cumulative calculation of the error impact value, the key mechanism that causes the positioning deviation in the fault location process has been successfully found, providing a precise direction for the subsequent optimization and improvement of the execution model.

[0032] In a possible implementation, step S231 further includes: Step S2311: using the deviation parameter as a keyword, capturing memory in historical data, and constructing a first record library, wherein the first record library is a historical data set including the deviation parameter; Step S2312: performing variable relationship fitting on the deviation parameter according to the first record library to obtain a first record influencing factor and a first influencing relationship thereof; Step S2313: extracting a variable combination according to the first influence relationship in the first record library, capturing the variable combination as a keyword in a historical database, and constructing a second record library, wherein the second record library is a historical data set of variable combinations including deviation parameters; Step S2314: performing correlation interaction on the deviation parameters according to the second record library to obtain a second record impact factor and a second impact relationship; Step S2315: cross-merge the first recorded impact factor and its first impact relationship, the second recorded impact factor and its second impact relationship to obtain the impact factor and impact relationship.

[0033] Specifically, in the fault location analysis process of high-voltage transmission lines, when the deviation parameter is obtained, the memory capture of historical data is started. The historical data contains a large amount of information related to past fault location. These data are stored in various forms, which may be table records in the database or log data saved in the form of files. With the deviation parameter as the keyword for accurate retrieval, a comprehensive and detailed search is performed in the huge historical data storage system with the help of a special search algorithm or data query tool. During the search process, each data record is screened and judged. As long as the data record contains the deviation parameter, whether it appears directly in numerical form or in related descriptions, calculation results and other fields, the data will be selected. For example, if the deviation parameter is the voltage deviation value at a specific tower, then the search will traverse all records about the tower voltage measurement and fault location calculation involving the tower voltage. Collect all qualified data, organize and store them according to certain rules (such as time sequence, data type, etc.) to build the first record library. This first record library becomes a historical data set specifically containing the deviation parameter, providing basic data support for subsequent in-depth analysis of the relationship between the deviation parameter and other factors, and for exploring potential influencing factors and influencing relationships.

[0034] After the construction of the first record library is completed, a large amount of historical data containing deviation parameters is stored in the first record library. These data involve multiple variables, such as the length of the transmission line, the type of tower, ambient temperature, humidity, etc., and there may be various complex relationships between them and the deviation parameters. In order to explore these potential relationships, data analysis tools and related algorithms, such as polynomial regression algorithms, are used to conduct in-depth analysis of the data in the first record library. Taking polynomial regression as an example, the deviation parameter is set as the dependent variable, and other related variables are used as independent variables to construct a polynomial regression model. When constructing the model, not only the simple linear relationship between the variables is considered, but also quadratic terms, cross terms, etc. are added to capture the second-order interaction relationship between the variables. For example, assuming that the deviation parameter is the distance deviation of fault location, and the independent variables include the length of the transmission line and the ambient temperature, the model may contain the first-order terms of the line length and the ambient temperature, and may also contain the product term of the line length and the ambient temperature and their respective quadratic terms. By fitting the data in the first record library, the parameters of the model are continuously adjusted so that the model can describe the relationship between the deviation parameter and other variables as accurately as possible. After repeated fitting and optimization, the variables that have a significant impact on the deviation parameters are screened out from the model. These variables are the first recorded influencing factors. At the same time, the specific mathematical relationship or action mode between these influencing factors and the deviation parameters is determined, that is, the first influencing relationship. For example, it may be concluded that for every 1°C increase in ambient temperature, the fault location distance deviation increases by 0.5 meters, and for every 1 km increase in line length, the distance deviation increases by 0.2 meters on the basis of considering the temperature effect. These first recorded influencing factors and their first influencing relationships preliminarily reveal the influencing factors and action mechanisms directly related to the deviation parameters, laying the foundation for further mining of more complex relationships.

[0035] After obtaining the first influence relationship in the first record library, the first influence relationship shows the association pattern between the deviation parameter and other variables, from which some variables with greater influence on the deviation parameter can be found. Based on these variables, variable combinations are extracted according to certain rules. These combinations not only contain simple single variables, but also focus on variable combinations containing second-order interactions, such as product terms between variables (such as the combination of variables A and B A×B), cross terms, etc. This is because in a complex high-voltage transmission line fault location system, the interaction between variables is often not a simple linear relationship, and the second-order interaction may have an important impact on the deviation parameter. After extracting the variable combination, these combinations are used as new keywords and searched in a huge historical database using data retrieval tools. The historical database contains a large amount of data related to high-voltage transmission lines, covering various types of information at different times and under different working conditions. During the search process, each record in the database is compared one by one. As long as the record contains these variable combinations, no matter what form it exists in, the record will be screened out. For example, if the variable combination is "measurement value of a certain model sensor × ambient humidity", then when searching, all records involving the measurement value of the sensor of this model and the ambient humidity, and the two are multiplied or related calculations involving this combination will be found. Collect all the records that meet the conditions, organize and store them according to specific logic (such as time sequence, data type, etc.), and finally build a second record library. This second record library specifically includes historical data of variable combinations containing deviation parameters. Compared with the first record library, it mines more complex correlation information between data, and provides richer data support for further analysis of the influencing factors of deviation parameters, especially those influencing relationships with indirect interactive responses, which helps to have a deeper understanding of the interaction mechanism of various factors in the fault location process.

[0036] After the second record library is built, the second record library stores historical data of variable combinations containing deviation parameters, which covers more complex data associations than the first record library. In order to mine deeper information from these data, the deep neural network in the deep learning model is used to perform association interaction analysis on the data in the second record library. In the analysis process, the focus is on the complex nonlinear relationships and high-order interaction effects between variables in the data. The deep neural network has a strong feature learning ability. It can automatically extract complex feature representations from a large amount of data. By inputting the data in the second record library into the neural network model, the model processes and analyzes the data layer by layer. For example, for data involving multiple variable combinations (including second-order interaction terms), the neural network can learn how these variable combinations affect each other and act on the deviation parameters together. After learning and analyzing a large amount of data, the model will screen out those factors that have a significant impact on the deviation parameters. These factors are the second record influencing factors. At the same time, the specific relationship between these influencing factors and the deviation parameters is determined, that is, the second influencing relationship. These second influencing relationships not only include the direct impact between variables, but also fully consider the impact of indirect interactive responses. For example, it is found that the product of two variables and the sum of another variable, under certain conditions, will affect the deviation parameter in a complex nonlinear way. Through such analysis, the second record influencing factor and the second influencing relationship obtained can more comprehensively and accurately reflect the complex associations hidden behind the data, providing key support for the subsequent more accurate analysis of the influencing factors in the fault location process.

[0037] After obtaining the first record influencing factors and their first influencing relationships, the second record influencing factors and their second influencing relationships, the first record influencing factors and their first influencing relationships mainly reflect the relatively direct and simple connection between the deviation parameters and the relevant variables, focusing on the first-order interaction; while the second record influencing factors and their second influencing relationships dig out the complex associations including nonlinear, indirect interactive responses and high-order interactions. The cross-fusion process is to conduct a comprehensive analysis of the two sets of influencing factors and influencing relationships at different levels. The two sets of data are integrated using methods such as deep learning models. The first record influencing factors and their relationships, and the second record influencing factors and their relationships are used as different input features of the model. Through the multi-layer structure of the neural network, the model automatically learns the synergy and potential connection between the two sets of data. In this process, the model will weight, combine and transform the two sets of influencing factors, and fully consider the possible high-order interactive relationships between the factors. For example, it may be found that a certain influencing factor in the first record combined with a certain variable in the second record will produce a new and more complex impact pattern on the deviation parameter under a specific weight distribution. By continuously adjusting the parameters and structure of the model, the model can accurately capture these complex interactions and finally output the comprehensive influencing factors and influencing relationships. These new influencing factors and influencing relationships integrate the information of direct and indirect influences, first-order interactions and higher-order interactions, and more comprehensively and accurately reveal the various factors and their interaction mechanisms that lead to fault location deviations, providing a more reliable basis for the subsequent optimization and improvement of the fault location model based on these results.

[0038] In a possible implementation, step S234 further includes: Step S2341: establishing a timing chain according to the timing corresponding positions of the participating nodes in the processing data chain timing relationship, wherein the timing chain nodes are participating nodes; Step S2342: Fitting the influencing factors and influencing relationships corresponding to the participating nodes to the time series chain nodes to construct a Markov chain model, wherein the Markov chain model describes the deviation dependency relationship among the participating nodes; Step S2343: performing time-series cumulative calculation on the impact results of the participating nodes according to the Markov chain model.

[0039] Specifically, based on the processing data chain timing information, the participating nodes are arranged in the order in which they appear in the entire fault location process to establish a timing chain. This timing chain uses participating nodes as node elements and clearly presents the distribution of each participating node in the time dimension, laying the foundation for subsequent analysis.

[0040] The influence factors and influence relationships corresponding to the determined participating nodes are integrated into the nodes of the timing chain to construct a Markov chain model. In the construction process, the state space of the Markov chain is first clarified, and each node in the timing chain is defined as a state. Each state corresponds to a specific value of the influence factor or a specific pattern of the combination of influence factors. Then, based on the historical data and the fitted influence factors, the transition probability between states is calculated. For example, referring to the past fault location data, the influence factor value of the participating node A in the current state is analyzed, and its impact on the subsequent state of the participating node B is judged, and the probability of transferring from the state of the participating node A to the state of the participating node B is determined. This probability reflects the timing dependency between nodes. Based on the historical data of each node, a transfer matrix is ​​constructed, and the elements in the matrix accurately represent the probability of transferring from one node state to another node state. Through these steps, a Markov chain model that can accurately describe the deviation dependency relationship in the participating nodes is successfully constructed. The model reflects the dynamic changes and interdependence characteristics between the nodes of the timing chain.

[0041] After the Markov chain model is constructed, the Markov chain model has clarified the transition probability and dependency between the states of each participating node. At this time, starting from the starting participating node of the timing chain, the initial impact result of the node itself is determined. This initial impact result may be the initial contribution value of the impact factor corresponding to the node to the fault location deviation, such as the initial impact value of a sensor measurement deviation on the positioning error. According to the transition probability in the Markov chain model, the change in the impact result of the next node when the starting node state is transferred to the next participating node state is calculated. This change is added to the initial impact result of the next node itself to obtain the comprehensive impact result of the next node in the current state. In this way, each participating node is calculated in turn along the timing chain. After each node is calculated, its comprehensive impact result is used as the basis for the calculation of the next node and is continuously accumulated. In the calculation process, the timing dependency between each node is fully considered, that is, the state and impact result of the previous node will have an effect on the next node according to the transition probability. Through such a step-by-step timing accumulation calculation, the cumulative impact result of all participating nodes on the entire timing chain on the fault location deviation is finally obtained. This cumulative result comprehensively reflects the combined effect of each participating node on the final deviation in the entire fault location process, and provides key quantitative data support for in-depth analysis of the formation mechanism of fault location deviation and determination of the execution model.

[0042] In a possible implementation, step S2343 further includes: Step S23431: Based on the timing chain node to which the execution model belongs, the deviation probability of all the preceding participating nodes is obtained through the Markov chain model; Step S23432: taking the timing chain node to which the execution model belongs as the segmentation point, calculating the deviation probability of subsequent participating nodes through the Markov chain model to obtain the subsequent deviation impact probability; Step S23433: Using the subsequent deviation impact probability as a screening threshold, screening and replacing the deviation probabilities of all the previous participating nodes, obtaining the participating node with the largest cumulative impact probability, and determining the execution model.

[0043] Specifically, in the process of fault location analysis, after completing the construction of the Markov chain model, the execution model corresponds to a specific node in the timing chain. Taking this node as the benchmark, with the help of the Markov chain model, the deviation probability of all the previous participating nodes is backtracked. These deviation probabilities reflect the possibility of the previous participating nodes affecting the fault location deviation, and are calculated based on the state transition probability of each node in the Markov chain model.

[0044] After the timing chain node to which the execution model belongs is determined, the timing chain node is used as the dividing point to start the in-depth analysis of the subsequent participating nodes. The Markov chain model has been constructed before, which clearly defines the state transition probability between each node. Based on this model, the calculation starts from the first participating node after the node to which the execution model belongs. First, the initial state of the node and its corresponding influencing factor are clarified. According to the state transition probability between the node and the previous node (i.e., the node to which the execution model belongs) in the Markov chain model, combined with the state and impact result of the previous node, the probability of the node affecting the fault location deviation in the current state is calculated. Then, in the same way, according to the Markov chain model, the deviation probability of the next node is calculated when the node is transferred to the next subsequent participating node, and each subsequent participating node is calculated in turn along the timing chain. The deviation probabilities of each of these subsequent participating nodes are summarized and comprehensively analyzed, and finally a value that can reflect the possibility of all subsequent participating nodes affecting the fault location deviation is obtained, which is the subsequent deviation impact probability. This probability value is crucial for evaluating the potential role of subsequent nodes in the entire process of fault location deviation formation, and provides important data basis for subsequent screening of key nodes and determination of execution models.

[0045] The subsequent deviation impact probability obtained is set as the screening threshold, and this threshold is used to screen and replace the deviation probabilities of all the previous participating nodes. In this process, by comparison and calculation, the participating node that makes the cumulative amount of the impact probability the largest is found. Because this participating node has the most significant comprehensive impact on the fault location deviation, it is determined as a key node closely related to the execution model, and then the execution model is determined. After the execution model is determined, it needs to be verified. Since the Markov model is based on probability calculation and accumulation, although the previous calculation may have caused the cumulative probability to reach the deviation amount, there may be other calculation models that affect fault location. Therefore, the probability of subsequent nodes continues to be verified. If it is found that the probability of the subsequent node is greater than the probability when the execution model is currently determined, it is necessary to find the key node again and determine again which node caused the deviation of the positioning result and the problem of insufficient precision, so as to ensure the accuracy and reliability of the execution model.

[0046] In a possible implementation, step S500 further includes: Step S510: collecting compensation parameter data according to the fault location target compensation parameter and constructing a migration training data set; Step S520: Based on the processing layer corresponding to the weak parameter, sample increment is performed from the migration training data set according to the original training record, and incremental learning is performed on the processing layer corresponding to the weak parameter to enhance the processing layer corresponding to the weak parameter; Step S530: constructing a missing processing layer according to the missing parameters, and screening the migration training data set to establish a training set and a test set based on the missing parameters; Step S540: using the screened training set and test set to train and converge the missing processing layer to obtain the missing processing layer.

[0047] Specifically, in the optimization process of fault location, the fault location target compensation parameters are the core guide of the entire operation, which clarifies the direction of adjustment and optimization required to improve the accuracy of fault location. Based on these target compensation parameters, compensation parameter data is collected in a targeted manner. This process needs to be comprehensive and meticulous, involving multiple data sources. For example, real-time line operation data such as current, voltage, temperature and other parameters are obtained from various sensors, because the changes in these physical quantities are closely related to fault location and may be affected by the target compensation parameters of fault location; at the same time, historical data related to compensation parameters in past fault cases are collected. These data record the actual operating status and adjustment of compensation parameters under different fault conditions, providing valuable experience reference for analysis. After collecting enough data, the data is sorted and preprocessed. Outliers are removed, missing values ​​are filled, and standardized or normalized according to the characteristics of the data and the needs of subsequent model training. Then, the processed data is organized and arranged according to certain rules to construct a migration training data set. This dataset not only contains data directly related to the fault location target compensation parameters, but also integrates data on various factors that may affect fault location, providing a rich and high-quality data foundation for subsequent transfer learning-based model training, to help improve the accuracy and reliability of fault location.

[0048] Focus on the processing layer corresponding to the existing weak parameters. This processing layer already exists in the previous fault location model training, but due to the existence of weak parameters, its performance needs to be further improved. At this time, the original training record plays an important role. It records in detail various information of the processing layer in the previous training process, including the samples used in training, the number of training rounds, and the update of parameters. With the help of the original training record, samples related to weak parameters are screened out from the migration training data set. The migration training data set contains rich and diverse data, which are collected based on the compensation parameters of the fault location target, providing a new source of knowledge for model optimization. Through careful screening, select those samples that can supplement and strengthen the weak parameter processing layer. This is the process of sample increment. Subsequently, these newly added samples are input into the processing layer corresponding to the weak parameters to start incremental learning. Incremental learning is a special learning method that allows the model to gradually learn new samples based on existing knowledge without retraining the entire model. In this process, the processing layer will adjust and optimize its own parameters based on the characteristics of the new samples and the knowledge in the original training record. For example, the weights in the neural network can be adjusted so that the processing layer can better process data related to weak parameters, thereby enhancing the performance of the processing layer corresponding to the weak parameters and enabling it to play a more accurate and stable role in fault location.

[0049] A special missing parameter processing layer is built for missing parameters. This processing layer is designed based on the characteristics of missing parameters and their role in fault location. After the construction is completed, appropriate data is filtered out from the migration training data set again based on the characteristics of missing parameters and divided into training set and test set. The training set is used for model training, and the test set is used to evaluate model performance.

[0050] After the construction of the training set and the test set is completed, the training set is input into the missing processing layer to start the training process. During the training, the missing processing layer will use optimization algorithms such as gradient descent based on the characteristics of the input data to continuously adjust its internal parameters, such as the weights and biases in the neural network, so that the model output results are as close to the true value as possible. With the increase of training rounds, the model's ability to fit the data in the training set gradually increases, and the loss function value continues to decrease. The test set is used regularly to evaluate the missing processing layer in training. By inputting the test set data, the model's performance indicators such as accuracy and recall are calculated to monitor whether the model is overfitting or underfitting. If the performance of the model on the test set no longer improves, or the loss function value converges to a stable minimum value, it means that the missing processing layer has been trained and converged. At this point, the missing processing layer obtained can effectively handle missing parameters, play a key role in fault location, and provide strong support for improving the accuracy of fault location.

[0051] Embodiment 2: Based on the same inventive concept as the method for locating a fault in a high-voltage transmission line in the above embodiment, Figure 2 As shown, the present application provides a fault location system for a high-voltage transmission line, and the system and method embodiments in the present application are based on the same inventive concept. The system includes: A fault location log extraction unit 10 is used to connect to the fault location system and extract a fault location log, wherein the fault location log has a fault location evaluation result; A tracing and parsing unit 20 is used to perform tracing and parsing of the fault location log according to the fault location evaluation result, obtain a location response module, and extract an execution model of the location response module; A fault location target compensation parameter acquisition unit 30 is used to perform a difference identification of the fault location accuracy on the fault location evaluation result to obtain a fault location target compensation parameter; An execution model decomposition unit 40 is used to obtain weak parameters and missing parameters by decomposing the execution model with the fault location target compensation parameter as the target, wherein the weak parameters are parameters whose fault location accuracy does not meet the requirements; The execution model replacement unit 50 is used to extract the retention layer of the execution model according to the weak parameters and the missing parameters, build a transfer learning framework based on the retention layer and the weak parameters and the missing parameters, train the processing layer of the weak parameters and the missing parameters through the transfer learning framework, connect the training processing layer with the retention layer, and replace the execution model.

[0052] Furthermore, the tracing and parsing unit 20 further includes: A processing data chain acquisition unit is used to perform time series analysis on each fault location log and obtain a processing data chain of each fault location evaluation result; A deviation information acquisition unit, used to match the fault location evaluation result with the fault location accuracy requirement to obtain deviation information; The association mining unit is used to use the deviation information as an identification index, perform association mining according to the processing data chain, trace back the execution model, and obtain the positioning response module.

[0053] Furthermore, the association mining unit further includes: A correlation analysis unit, configured to perform correlation analysis on historical data based on the deviation parameters of the deviation information to obtain influencing factors and influencing relationships; A participating node acquisition unit, configured to use the influencing factor as an index to search in the processing data chain to obtain participating nodes; An error impact value acquisition unit, used to perform an impact calculation on the participating nodes according to the impact relationship to obtain an error impact value; The cumulative calculation unit is used to locate the participating nodes based on the processing data link timing relationship, and to perform cumulative calculation on the error impact value according to the timing relationship. When the cumulative amount reaches the deviation amount of the deviation information, the execution model is located.

[0054] Furthermore, the correlation analysis unit also includes: A first record library construction unit, configured to use the deviation parameter as a keyword, capture memory in historical data, and construct a first record library, wherein the first record library is a historical data set including the deviation parameter; A variable relationship fitting unit, used for performing variable relationship fitting on the deviation parameter according to the first record library to obtain a first record influencing factor and a first influencing relationship thereof; A second record library construction unit is used to extract a variable combination according to the first influence relationship in the first record library, and use the variable combination as a keyword to memorize and capture in a historical database to construct a second record library, wherein the second record library is a historical data set of variable combinations including deviation parameters; An association interaction unit, configured to perform association interaction on the deviation parameter according to the second record library to obtain a second record impact factor and a second impact relationship; The cross-fusion unit is used to perform cross-fusion using the first recorded influencing factor and its first influencing relationship, the second recorded influencing factor and its second influencing relationship to obtain the influencing factor and the influencing relationship.

[0055] Furthermore, the cumulative calculation unit also includes: A timing chain establishing unit, configured to establish a timing chain according to the timing corresponding positions of the participating nodes in the processing data chain timing relationship, wherein the timing chain nodes are participating nodes; A Markov chain model building unit, used to fit the influencing factors and influencing relationships corresponding to the participating nodes into the time series chain nodes to build a Markov chain model, wherein the Markov chain model describes the deviation dependency relationship among the participating nodes; A time series accumulation calculation unit is used to perform time series accumulation calculation on the impact results of the participating nodes according to the Markov chain model.

[0056] Furthermore, the time series accumulation calculation unit also includes: A deviation probability acquisition unit, used to acquire the deviation probability of all the preceding participating nodes through the Markov chain model based on the timing chain node to which the execution model belongs; A subsequent deviation impact probability acquisition unit is used to calculate the deviation probability of subsequent participating nodes by using the Markov chain model with the timing chain node to which the execution model belongs as a segmentation point to obtain the subsequent deviation impact probability; The screening and replacement unit is used to use the subsequent deviation impact probability as a screening threshold, screen and replace the deviation probabilities of all the previous participating nodes, obtain the participating node with the largest cumulative impact probability, and determine the execution model.

[0057] Furthermore, the execution model replacement unit 50 further includes: A migration training data set construction unit, used to collect compensation parameter data and construct a migration training data set according to the fault location target compensation parameter; An incremental learning unit, configured to perform incremental learning on the processing layer corresponding to the weak parameter and enhance the processing layer corresponding to the weak parameter by performing sample increment from the migration training data set according to the original training record based on the processing layer corresponding to the weak parameter; A missing processing layer construction unit, used to construct a missing processing layer according to the missing parameters, and to screen and establish a training set and a test set from the migration training data set based on the missing parameters; The missing processing layer acquisition unit is used to use the screening and constructed training set and test set to train and converge the missing processing layer to obtain the missing processing layer.

[0058] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0059] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0060] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A method for locating a fault in a high voltage transmission line, characterized in that: include: Connecting to a fault location system and extracting a fault location log, wherein the fault location log has a fault location evaluation result; According to the fault location evaluation result, the fault location log is retroactively analyzed to obtain a location response module, and an execution model of the location response module is extracted; Performing differential identification of fault location accuracy on the fault location evaluation result to obtain a fault location target compensation parameter; Taking the fault location target compensation parameter as a target, by decomposing the execution model, weak parameters and missing parameters are obtained, wherein the weak parameters are parameters whose fault location accuracy does not meet the requirements; The execution model is subjected to reserved layer extraction according to the weak parameters and the missing parameters, a transfer learning framework is constructed based on the reserved layer and the weak parameters and the missing parameters, a processing layer is trained on the weak parameters and the missing parameters through the transfer learning framework, the processing layer obtained by the training is connected to the reserved layer, and the execution model is replaced.

2. The method for locating a fault in a high-voltage transmission line according to claim 1, characterized in that: According to the fault location evaluation result, the fault location log is retroactively parsed to obtain a location response module, and an execution model of the location response module is extracted, including: Perform time series analysis on each fault location log to obtain the processing data chain of each fault location evaluation result; Matching the fault location evaluation result with the fault location accuracy requirement to obtain deviation information; The deviation information is used as an identification index, association mining is performed according to the processing data chain, and the execution model is traced back to obtain the positioning response module.

3. The method for locating a fault in a high-voltage transmission line according to claim 2, characterized in that: Using the deviation information as an identification index, performing association mining according to the processing data chain, and tracing back the execution model, including: Based on the deviation parameters of the deviation information, correlation analysis is performed through historical data to obtain influencing factors and influencing relationships; Using the impact factor as an index to search the processing data chain, and obtaining participating nodes; Performing influence calculation on the participating nodes according to the influence relationship to obtain an error influence value; The participating nodes are positioned based on the processing data link timing relationship, the error impact value is cumulatively calculated according to the timing relationship, and when the cumulative amount reaches the deviation amount of the deviation information, the execution model is positioned.

4. The method for locating a fault in a high-voltage transmission line according to claim 3, characterized in that: Based on the deviation parameters of the deviation information, correlation analysis is performed through historical data to obtain influencing factors and influencing relationships, including: The deviation parameter is used as a keyword to capture memory in historical data to construct a first record library, wherein the first record library is a historical data set including the deviation parameter; According to the first record library, performing variable relationship fitting on the deviation parameter to obtain a first record influencing factor and a first influencing relationship thereof; Extracting a variable combination according to the first influence relationship in the first record library, capturing the variable combination as a keyword in a historical database, and constructing a second record library, wherein the second record library is a historical data set of the variable combination including the deviation parameter; According to the second record library, the deviation parameters are associated and interacted to obtain a second record impact factor and a second impact relationship; The first recorded impact factor and its first impact relationship, the second recorded impact factor and its second impact relationship are cross-fused to obtain the impact factor and the impact relationship.

5. The method for locating a fault in a high-voltage transmission line according to claim 4, characterized in that: Positioning the participating nodes based on the processing data link timing relationship, and accumulating and calculating the error impact value according to the timing relationship, including: Establishing a timing chain according to the timing corresponding positions of the participating nodes in the processing data chain timing relationship, wherein the timing chain nodes are participating nodes; Fitting the influencing factors and influencing relationships corresponding to the participating nodes to the time series chain nodes to construct a Markov chain model, wherein the Markov chain model describes the deviation dependency relationship among the participating nodes; The influence results of the participating nodes are calculated by time series accumulation according to the Markov chain model.

6. The method for locating a fault in a high-voltage transmission line according to claim 5, characterized in that: Also includes: Based on the timing chain node to which the execution model belongs, obtaining the deviation probability of all the preceding participating nodes through the Markov chain model; Taking the timing chain node to which the execution model belongs as the segmentation point, the deviation probability of subsequent participating nodes is calculated through the Markov chain model to obtain the subsequent deviation impact probability; The subsequent deviation impact probability is used as a screening threshold, the deviation probabilities of all the previous participating nodes are screened and replaced, the participating node with the largest cumulative impact probability is obtained, and the execution model is determined.

7. The method for locating a fault in a high-voltage transmission line according to claim 1, characterized in that: The weak parameters and missing parameters are processed and trained through the transfer learning framework, including: According to the compensation parameters of the fault location target, the compensation parameter data is collected to build a migration training data set; Based on the processing layer corresponding to the weak parameter, sample increment is performed from the migration training data set according to the original training record, and incremental learning is performed on the processing layer corresponding to the weak parameter to enhance the processing layer corresponding to the weak parameter; Constructing a missing processing layer according to the missing parameters, and screening the migration training data set to establish a training set and a test set based on the missing parameters; The missing processing layer is trained and converged using the screening and constructed training set and test set to obtain the missing processing layer.

8. A fault location system for a high voltage transmission line, characterized in that: The system is used to implement the high-voltage transmission line fault location method according to any one of claims 1 to 7, and the system comprises: A fault location log extraction unit, used to connect to the fault location system and extract a fault location log, wherein the fault location log has a fault location evaluation result; A tracing and parsing unit, configured to perform tracing and parsing of the fault location log according to the fault location evaluation result, obtain a location response module, and extract an execution model of the location response module; A fault location target compensation parameter acquisition unit is used to perform a difference identification of the fault location accuracy on the fault location evaluation result to obtain a fault location target compensation parameter; An execution model decomposition unit is used to obtain weak parameters and missing parameters by decomposing the execution model with the fault location target compensation parameter as the target, wherein the weak parameters are parameters whose fault location accuracy does not meet the requirements; The execution model replacement unit is used to extract the reserved layer of the execution model according to the weak parameters and the missing parameters, build a transfer learning framework based on the reserved layer and the weak parameters and the missing parameters, perform processing layer training on the weak parameters and the missing parameters through the transfer learning framework, connect the processing layer obtained by training with the reserved layer, and replace the execution model.

Images