Power distribution network fault locating method and system based on intelligent decision

Abstract

The application provides a distribution network fault positioning method and system based on intelligent decision-making, which obtains a target state data set collected by a global intelligent perception network of a distribution network, performs spatiotemporal semantic fusion processing to generate a multi-dimensional feature graph, uses a dynamic decision-making reasoning engine to perform context perception discrimination on the multi-dimensional feature graph, generates a fault risk discrimination result, traces a cause-effect link based on the fault risk discrimination result, generates a fault section positioning result, generates a fault positioning instruction according to the fault section positioning result and sends the fault positioning instruction to a distribution network intelligent operation and maintenance platform, and triggers a precise maintenance process, so that the global state data of the distribution network is comprehensively utilized, the precise evaluation of the fault risk of the distribution network, the rapid positioning of the fault section and the efficient triggering of the precise maintenance process are realized, and the operation reliability and safety of the distribution network are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network operation and maintenance, and in particular to a distribution network fault locating method and system based on intelligent decision-making. Background Art

[0002] In the operation and management of distribution networks, fault location is a key link in ensuring the safe and stable operation of the power grid. Traditional distribution network fault location methods mainly rely on manual inspections, empirical judgment, and simple fault indicators. These methods suffer from low positioning accuracy, slow response speed, and inability to fully perceive the operating status of the power grid. With the continuous expansion and increase in the complexity of distribution networks, as well as the large-scale integration of new energy sources and distributed power sources, the operating environment of distribution networks has become more complex and changeable. Traditional fault location methods can no longer meet the management needs of modern distribution networks. Although existing fault location methods based on intelligent technology have improved the accuracy and efficiency of positioning to a certain extent, most of them only focus on a single type of data or feature, lacking comprehensive analysis of the global status data of the distribution network and spatiotemporal semantic fusion processing. As a result, it is difficult to accurately reflect the dynamic operating status of the distribution network and the evolution of fault risks, resulting in inaccurate and incomplete fault location results. Summary of the Invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a distribution network fault location method based on intelligent decision-making, the method comprising:

[0004] Obtaining a target state data set collected by the global intelligent perception network of the distribution network, wherein the target state data set includes equipment operating condition data, network topology data, historical fault knowledge base, and environmental interference data;

[0005] Performing spatiotemporal semantic fusion processing on the target state data set to generate a multidimensional feature map reflecting the dynamic operation status of the distribution network, wherein the multidimensional feature map includes the state evolution trajectory in the time dimension and the equipment coordination relationship in the space dimension;

[0006] The multi-dimensional feature map is subjected to context-aware discrimination processing by a pre-built dynamic decision-making inference engine to generate a fault risk discrimination result of the distribution network, wherein the fault risk discrimination result includes a risk level identifier, a risk triggering period, and a risk impact range;

[0007] Performing causal link tracing processing on the target state data set based on the fault risk identification result to generate a fault section location result of the distribution network, wherein the fault section location result includes a risk propagation path, key impact nodes, and a quantitative value of the impact degree;

[0008] A fault location instruction including a path identifier, a node identifier and an impact degree is generated according to the fault section location result, and the fault location instruction is sent to the distribution network intelligent operation and maintenance platform to trigger a maintenance process.

[0009] On the other hand, an embodiment of the present invention also provides a distribution network fault location system based on intelligent decision-making, including a processor and a machine-readable storage medium, the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0010] Based on the above aspects, the embodiment of the present invention obtains a target state data set collected by the global intelligent perception network of the distribution network, covering multiple aspects such as equipment operating conditions, network topology, historical fault knowledge base, and environmental interference. The target state data set is subjected to spatiotemporal semantic fusion processing to generate a multidimensional feature map reflecting the dynamic operation status of the distribution network. The multidimensional feature map can accurately characterize the operating characteristics and equipment coordination relationship of the distribution network in different time and space dimensions. The multidimensional feature map is subjected to context-aware discrimination processing by a pre-built dynamic decision-making inference engine to generate a fault risk discrimination result including risk level identification, risk trigger period, and risk impact range, thereby achieving accurate assessment and early warning of distribution network fault risks. Based on the fault risk discrimination result, the target state data set is subjected to causal link tracking processing to generate a fault section location result including risk propagation path, key influencing nodes, and quantitative value of impact degree, further improving the accuracy and pertinence of fault location. Finally, a fault location instruction is generated based on the fault section location result and sent to the distribution network intelligent operation and maintenance platform, triggering a precise maintenance process, achieving rapid location and efficient processing of distribution network faults, and significantly improving the operational reliability and safety of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the execution flow of the distribution network fault location method based on intelligent decision-making provided by an embodiment of the present invention.

[0012] Figure 2 Schematic diagram of exemplary hardware and software components of a distribution network fault location system based on intelligent decision-making provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a distribution network fault location method based on intelligent decision-making provided by an embodiment of the present invention. The distribution network fault location method based on intelligent decision-making is introduced in detail below.

[0014] Step S110: Obtain a target state data set collected by the power distribution network global intelligent sensing network, the target state data set including device operation condition data, network topology structure data, historical fault knowledge base, and environmental interference data.

[0015] In this embodiment, the power distribution network global intelligent sensing network is a monitoring system widely distributed in various key positions of the power distribution network, which mainly collects various data related to the operation of the power distribution network to construct a target state data set. The target state data set covers device operation condition data, network topology structure data, historical fault knowledge base, and environmental interference data.

[0016] The collection of device operation condition data depends on sensors installed on various key devices in the power distribution network. For example, voltage sensors, current sensors, and temperature sensors are installed on transformers. Voltage sensors can monitor the voltage values at the input and output of the transformer in real time, current sensors are used to measure the current passing through the transformer, and temperature sensors can detect the temperature changes of the transformer during operation. The above sensors sample the operating parameters of the device according to the set sampling frequency to obtain a series of time series sampling values, such as voltage time series sampling values, current time series sampling values, and temperature time series sampling values. These sampling values constitute part of the device operation condition data, reflecting the operating state of the device at different times.

[0017] The network topology structure data is obtained by means of a network topology monitoring system. This system monitors and records the physical and logical connections between devices in the power distribution network to obtain the network topology structure information of the power distribution network. Specifically, it records the connection relationships between various device nodes (such as transformers, circuit breakers, switches, etc.), as well as the connection methods and paths. For example, it can record which circuit breakers a certain transformer is connected to and how these connections are achieved through cables or lines. Network topology structure data is crucial for analyzing the transmission path of electrical energy in the power distribution network and the cooperative relationship between devices.

[0018] The historical fault knowledge base is a database formed by detailed recording and organizing all past fault events in the power distribution network, which contains information such as the specific time of each fault occurrence, the location of the fault (i.e., the involved device node), the type of fault (such as short circuit fault, circuit breaking fault, etc.), and the cause of the fault.

[0019] Environmental interference data is collected using a variety of environmental monitoring equipment, such as weather stations and electromagnetic monitoring devices. Weather stations provide real-time monitoring of meteorological conditions in the distribution network area, including temperature, humidity, wind speed, air pressure, and other environmental factors. Electromagnetic monitoring equipment monitors the electromagnetic environment surrounding the distribution network, such as the intensity and frequency of electromagnetic interference. These environmental factors may affect the operation of distribution network equipment. For example, high temperatures may cause equipment to overheat, increasing the risk of equipment failure; strong electromagnetic interference may disrupt normal equipment communications and control signals.

[0020] Step S120: performing spatiotemporal semantic fusion processing on the target state data set to generate a multidimensional feature map reflecting the dynamic operation status of the distribution network, wherein the multidimensional feature map includes the state evolution trajectory in the time dimension and the equipment collaborative relationship in the space dimension.

[0021] In this embodiment, the purpose of performing spatiotemporal semantic fusion on the target state data set is to integrate data of different types and dimensions, explore the inherent connections between the data, and thus generate a multidimensional feature map that comprehensively reflects the dynamic operation status of the distribution network. This multidimensional feature map includes the state evolution trajectory in the time dimension and the device coordination relationship in the spatial dimension.

[0022] Step S121: performing time window alignment processing on the equipment operating condition data in the target state data set to obtain a condition data sequence with a time synchronization relationship, wherein the condition data sequence includes time series sampling values ​​of operating parameters in multiple dimensions.

[0023] In this embodiment, since the sampling frequencies of sensors of different devices may be different, the collected device operating condition data may have time differences. In order to effectively analyze and compare this data, it is necessary to perform time window alignment on the device operating condition data.

[0024] Specifically, we first need to determine a unified time window size and time step. The time window size determines the data range processed at each time, while the time step determines the interval between adjacent time windows. For example, we can set the time window size to a specific time period and the time step to a fraction of that time period. Then, using this unified time window as a benchmark, we align the time series sampling values ​​of the operating parameters of each device.

[0025] For the time series sampling values ​​of the operating parameters of each device, they are arranged in chronological order and then divided according to the time window and time step. If there are missing values ​​in the sampling values ​​of a certain device within a certain time window, the interpolation method can be used to supplement them. For example, the linear interpolation method can be used to estimate the missing values ​​based on the sampling values ​​of the device at adjacent time points. Through the above processing, a working condition data sequence with a time synchronization relationship is obtained. The working condition data sequence contains the time series sampling values ​​of the operating parameters of multiple dimensions, such as the time series sampling values ​​of voltage, the time series sampling values ​​of current, the time series sampling values ​​of power, etc. The sampling values ​​of each dimension are arranged on a unified time scale.

[0026] Step S122: performing graph model conversion processing on the network topology data in the target state data set to generate a distribution network topology graph including device nodes and connection edges, wherein the weight of the connection edge is determined by the power transmission capacity between devices.

[0027] In this embodiment, in order to more intuitively represent the network topology of the distribution network and facilitate subsequent analysis and processing, it is necessary to convert the network topology data into a graph model.

[0028] First, each device in the distribution network is abstracted as a node in a graph model. For example, transformers, circuit breakers, and switches can all be considered nodes in the graph. Then, edges are added to the graph based on the connection relationships between devices recorded in the network topology data. For example, if two devices are physically connected, an edge is added between the corresponding nodes.

[0029] The weight of a connection edge is determined based on the power transmission capacity between devices. The power transmission capacity reflects the maximum amount of electrical energy that can be transmitted between two devices. Specifically, for each connection edge, the greater the power transmission capacity between the corresponding devices, the higher the weight of the connection edge. For example, if the power transmission capacity between two devices is large, it means that the connection between them is crucial for the transmission of electrical energy in the distribution network. Therefore, the weight of the corresponding connection edge in the graph model will be set higher; conversely, if the power transmission capacity between the two devices is small, the weight of the connection edge will be lower. In this way, a distribution network topology graph containing device nodes and connection edges is generated. This distribution network topology graph intuitively reflects the connection relationship between devices in the distribution network and the importance of electrical energy transmission.

[0030] Step S123: performing spatial region division processing on the environmental interference data in the target state data set to generate an environmental parameter distribution matrix of the area where the device is located, wherein the environmental parameter distribution matrix includes spatial sampling values ​​of environmental parameters in multiple dimensions.

[0031] In this embodiment, since the environmental interference data is collected at different spatial locations of the distribution network, in order to better integrate it with the equipment operating condition data and network topology data, it is necessary to perform spatial region division processing on the environmental interference data.

[0032] First, the distribution network area is divided into multiple sub-areas based on the geographic scope and device distribution. The division principle can be determined based on factors such as geographic coordinates and device density. For example, the distribution network area can be divided into several small rectangular sub-areas according to a specific latitude and longitude grid.

[0033] Then, for each sub-area, sampled values ​​of environmental parameters are collected. Environmental parameters include temperature, humidity, wind speed, electromagnetic interference, and other dimensions. For each sub-area, these sampled values ​​are collated and statistically analyzed to obtain the environmental parameter characteristics of the sub-area. For example, the average temperature, maximum humidity, and standard deviation of wind speed within the sub-area can be calculated.

[0034] Finally, the environmental parameter characteristics of all sub-regions are arranged in a predetermined order to form an environmental parameter distribution matrix, where each row represents a sub-region and each column represents an environmental parameter dimension. For example, the first column of the matrix might represent temperature, the second column might represent humidity, and so on. This generates an environmental parameter distribution matrix for the region where the device is located. This environmental parameter distribution matrix reflects the environmental characteristics of different regions of the distribution network.

[0035] Step S124: input the operating condition data sequence, distribution network topology map and environmental parameter distribution matrix into the semantic fusion module, and extract the association rules between the equipment operating status and historical fault modes through semantic matching processing of the historical fault knowledge base.

[0036] In this embodiment, the main function of the semantic fusion module is to fuse the processed operating condition data sequence, distribution network topology map and environmental parameter distribution matrix, and use the historical fault knowledge base for semantic matching processing to extract the association rules between the equipment operating status and historical fault modes.

[0037] Step S1241: performing feature extraction processing on the operating condition data sequence to generate an operating condition feature vector containing abnormal operating condition events.

[0038] In this embodiment, the feature extraction process is performed on the operating condition data sequence in order to extract key features that can reflect the operating status of the equipment from a large number of time-series sampling values ​​of operating parameters.

[0039] Specifically, we first analyze the time-series sampling values ​​of the operating parameters for each dimension in the operating condition data sequence. For example, for the time-series sampling values ​​of voltage, we calculate its statistical characteristics such as mean, standard deviation, maximum value, and minimum value. These statistical characteristics can reflect the changing trend and fluctuation of voltage over a period of time.

[0040] Abnormal operating conditions are then identified using set thresholds. For example, for time-series voltage sampling, if the voltage value at a certain point in time exceeds the pre-set normal range (i.e., the voltage threshold), an abnormal operating condition is considered to have occurred at that point in time. The relevant information about these abnormal operating conditions (such as the time of occurrence and the severity of the abnormality) is organized and encoded to form a characteristic representation of the abnormal operating condition.

[0041] Finally, the statistical features of all dimensions and the characteristic representations of abnormal operating conditions are combined to form an operating condition feature vector. This operating condition feature vector contains information from multiple dimensions and can comprehensively reflect the operating status of the equipment.

[0042] Step S1242: performing neighborhood feature aggregation processing on the distribution network topology graph to generate a topology feature vector including node degrees, edge weights, and clustering coefficients.

[0043] In this embodiment, the neighborhood feature aggregation processing is performed on the distribution network topology graph in order to extract feature information of nodes and edges in the graph and generate a topological feature vector that can reflect the topological structure characteristics of the distribution network.

[0044] First, the node degree of each node in the distribution network topology is calculated. The node degree refers to the number of edges connected to the node, reflecting the density of connections within the distribution network topology. For example, a node with a high node degree indicates that it has many connections to other nodes and may be a relatively important node in the distribution network.

[0045] Then, for each edge in the graph, record its edge weight. The edge weight is determined based on the power transmission capacity between devices and reflects the importance of the edge in the distribution network.

[0046] Next, the clustering coefficient of each node is calculated. The clustering coefficient measures the density of connections between neighboring nodes. Specifically, for a node, the ratio of the number of actual edges connecting its neighboring nodes to the number of theoretically possible edges is calculated to obtain the clustering coefficient. A higher clustering coefficient indicates a denser connection between neighboring nodes.

[0047] Finally, the node degree, edge weight and clustering coefficient of each node are arranged in a set order to form a topological feature vector. This topological feature vector contains important feature information of nodes and edges in the distribution network topology graph.

[0048] Step S1243: performing statistical analysis on the environmental parameter distribution matrix to generate an environmental feature vector including the humidity mean, temperature variance, and electromagnetic interference peak value.

[0049] In this embodiment, the purpose of performing statistical analysis on the environmental parameter distribution matrix is ​​to extract key information that can reflect environmental characteristics from the environmental parameter data and generate an environmental characteristic vector.

[0050] First, calculate the mean of the humidity data in the environmental parameter distribution matrix. The mean humidity value reflects the overall humidity level in the distribution network area. For example, a high mean humidity value indicates that the air in the area is humid, which may affect the insulation performance of the equipment.

[0051] Next, calculate the variance of the temperature data. This variance reflects the fluctuations in temperature across different sub-regions. A large variance indicates significant temperature differences between regions, potentially leading to significant differences in the operating environment of the device in different regions.

[0052] Next, find the peak value of the electromagnetic interference data. This peak value reflects the maximum interference intensity of the electromagnetic environment surrounding the distribution network. If the peak value of electromagnetic interference is high, it may interfere with the communication and control signals of the equipment, affecting its normal operation.

[0053] Finally, the humidity mean, temperature variance, and electromagnetic interference peak are combined to form an environmental feature vector, which contains the key characteristic information of environmental parameters.

[0054] Step S1244: inputting the operating condition feature vector, topology feature vector, and environment feature vector into the semantic matching unit of the historical fault knowledge base, and calculating the cosine similarity between each feature vector and the historical fault mode feature vector.

[0055] In this embodiment, the main function of the semantic matching unit of the historical fault knowledge base is to compare the operating condition feature vector, the topology feature vector, and the environment feature vector with the historical fault mode feature vector and calculate the cosine similarity between them.

[0056] First, for each historical failure mode in the historical failure knowledge base, there is a corresponding historical failure mode feature vector. This historical failure mode feature vector is extracted based on information such as the equipment operating status, network topology, and environmental conditions when the failure mode occurred. It has the same dimension as the operating condition feature vector, topology feature vector, and environmental feature vector.

[0057] Next, the cosine similarity between the operating condition feature vector, topology feature vector, and environmental feature vector and each historical fault mode feature vector is calculated. Cosine similarity is a metric that measures the similarity between two vectors. It determines this similarity by calculating the cosine value of the angle between the two vectors. A cosine similarity value closer to 1 indicates a more similar vector; a cosine similarity value closer to 0 indicates a less similar vector.

[0058] Specifically, for two vectors A and B, the cosine similarity formula is: Cosine Similarity = (A·B) / (||A||*||B||), where A·B represents the dot product of vectors A and B, and ||A|| and ||B|| represent the moduli of vectors A and B, respectively. The calculated cosine similarity value can be used to measure the similarity between the current device operating status, network topology, and environmental conditions and historical failure modes.

[0059] Step S1245: historical fault patterns whose similarity exceeds a preset threshold are screened out according to the cosine similarity value, and their corresponding equipment state evolution rules and fault triggering conditions are extracted as association rules.

[0060] In this embodiment, based on the calculated cosine similarity value, historical fault patterns whose similarity with the current operating condition feature vector, topology feature vector, and environment feature vector exceeds a preset threshold are selected. The preset threshold is a similarity standard set based on actual conditions and experience. Only when the cosine similarity value exceeds the preset threshold is the current situation considered to have a high degree of similarity with the historical fault pattern.

[0061] For the selected historical failure patterns, the corresponding device state evolution patterns and fault triggering conditions are extracted. The device state evolution pattern describes how the device's operating state gradually changes before a failure occurs, for example, how operating parameters such as voltage, current, and temperature change over time. Fault triggering conditions refer to the specific causes and conditions that led to the failure, such as specific environmental conditions or abnormal device operation.

[0062] These device state evolution patterns and fault triggering conditions constitute association rules. These association rules can help predict potential faults in the current distribution network and take appropriate preventative measures. For example, if the current device operating state and environmental conditions are similar to a historical fault pattern, and the fault triggering conditions of that historical fault pattern are also partially met in the current situation, then a similar fault can be predicted, allowing for proactive equipment inspection and maintenance.

[0063] Step S125: Based on the association rules, feature cross-fusion processing is performed on the operating condition data sequence, the distribution network topology map and the environmental parameter distribution matrix to generate a multi-dimensional feature map containing time evolution trajectory and spatial collaborative relationship.

[0064] In this embodiment, based on the extracted association rules, feature cross-fusion processing is performed on the operating condition data sequence, distribution network topology map and environmental parameter distribution matrix, with the aim of generating a multidimensional feature map that can comprehensively reflect the dynamic operation status of the distribution network. The multidimensional feature map includes the state evolution trajectory in the time dimension and the equipment collaborative relationship in the spatial dimension.

[0065] Step S1251: Perform sliding window analysis on the operating condition data sequence in the time dimension, count the occurrence frequency of abnormal operating condition events within the sliding window, and generate a time evolution trajectory of the equipment status.

[0066] In this embodiment, the sliding window analysis processing of the time dimension is performed on the operating condition data sequence in order to observe the temporal changes of the equipment status and generate a temporal evolution trajectory of the equipment status.

[0067] First, define the size and step size of a sliding window. The size of a sliding window represents the time range covered by the window, and the step size represents the time interval between each window movement. For example, the size of a sliding window can be set to a specific time period, and the step size can be set to a fraction of that time period.

[0068] The sliding window is then slid across the operating condition data sequence, one step at a time. For the operating condition data within each sliding window, the frequency of abnormal operating condition events is calculated. This frequency reflects the degree of equipment instability within that time period. For example, a high frequency of abnormal operating condition events within a sliding window indicates that the equipment's operating status during that time period was unstable, potentially posing a risk of failure.

[0069] Finally, the frequency of abnormal operating conditions corresponding to each sliding window is arranged in chronological order to form a time evolution trajectory of the equipment status. This trajectory can intuitively show the changing trend of the equipment status over time.

[0070] Step S1252: Perform path search processing in the spatial dimension on the distribution network topology map, starting from the significant load device node, traverse all reachable nodes and record the edge weights on the path to generate the spatial collaborative relationship between devices.

[0071] In this embodiment, the path search process in the spatial dimension is performed on the distribution network topology map in order to analyze the spatial collaborative working relationship between devices and generate the spatial collaborative relationship between devices.

[0072] First, identify the significant load equipment nodes. Significant load equipment nodes refer to equipment nodes that bear a large load in the distribution network, such as large transformers. These nodes play an important role in the distribution network and may have a significant impact on the operating status of other equipment.

[0073] Then, starting with the node with the most significant load, the distribution network topology is traversed using a path search algorithm (such as depth-first search or breadth-first search). During the traversal, the edge weights along the path from the starting point to each reachable node are recorded. Edge weights reflect the power transfer capacity and importance of the connection between devices.

[0074] By traversing all reachable nodes and recording edge weights, we can obtain spatial coordination relationships between devices. This relationship describes the spatial connections and interactions between devices in the distribution network, such as which devices have direct power transmission relationships and which connections between devices are more important.

[0075] Step S1253: performing spatiotemporal joint analysis on the environmental parameter distribution matrix, counting the co-occurrence frequency of environmental parameters and equipment failures in the target area, and generating coupling characteristics of environmental interference and equipment status.

[0076] In this embodiment, the purpose of performing spatiotemporal joint analysis on the environmental parameter distribution matrix is ​​to deeply explore the intrinsic relationship between environmental factors and device states and generate coupling characteristics between environmental interference and device states.

[0077] First, define the target area. This area can be divided based on the actual conditions of the distribution network, such as geographic region or equipment distribution area. After determining the target area, analyze the environmental parameter distribution matrix within that area.

[0078] For each environmental parameter dimension in the environmental parameter distribution matrix (such as temperature, humidity, wind speed, and electromagnetic interference), combined with historical fault records, the co-occurrence frequency between this environmental parameter and the device fault is calculated. The co-occurrence frequency is calculated as follows: At each time point and spatial location, the environmental parameter is determined to be within a specified range and whether a device fault has occurred. If the environmental parameter is within the specified range and a device fault has occurred at the same time and spatial location, the environmental parameter is considered to co-occur with the device fault. The number of co-occurrences at all time points and spatial locations is counted and divided by the total number of time points and spatial locations to obtain the co-occurrence frequency between the environmental parameter and the device fault.

[0079] For example, for temperature parameters, we set a temperature range (e.g., above a certain temperature threshold), count the number of device failures within that temperature range, and divide the total number of counts by the total number of times to obtain the co-occurrence frequency of temperature and device failures. By performing the above statistical analysis on each environmental parameter dimension, we obtain a series of co-occurrence frequency values.

[0080] These co-occurrence frequency values ​​reflect the degree of correlation between environmental parameters and device failures. The higher the co-occurrence frequency, the greater the impact of the environmental parameter on the device failure. Arranging these co-occurrence frequency values ​​according to the environmental parameter dimensions creates a coupling signature between environmental interference and device status. This coupling signature reflects the interaction between environmental factors and device status.

[0081] Step S1254: Input the time evolution trajectory, spatial synergy relationship and coupling features into the feature map construction unit, and generate a multi-dimensional feature map with uniform time resolution and spatial resolution through feature dimension alignment processing.

[0082] In this embodiment, the function of the feature map construction unit is to integrate the time evolution trajectory, spatial synergy relationship and coupling features to generate a multi-dimensional feature map. Before integration, feature dimension alignment is required to ensure that these features have uniform temporal and spatial resolutions.

[0083] The time evolution trajectory records the changes in device status over time, with a certain temporal resolution, namely the time interval represented by each data point. The spatial coordination relationship describes the spatial connection and interaction between devices, with a certain spatial resolution, namely the spatial range represented by each node or region. Coupling characteristics are the correlation characteristics between environmental parameters and device failures, and they also need to be aligned with the time evolution trajectory and spatial coordination relationship in time and space.

[0084] The specific process of feature dimension alignment is as follows: First, determine a unified temporal and spatial resolution. The temporal resolution can be selected based on actual needs and data characteristics, for example, a fixed time interval can be used as the unified temporal resolution. The spatial resolution can be determined based on the geographic scope and device distribution of the distribution network, for example, by dividing the distribution network into several equal-sized regions as the unified spatial resolution.

[0085] Then, the temporal trajectory is resampled or interpolated according to a unified temporal resolution, so that the time interval of each data point is consistent with the unified temporal resolution. Spatial synergy relationships are divided or merged according to a unified spatial resolution, so that the spatial extent of each node or region is consistent with the unified spatial resolution. Coupling features are also adjusted according to the unified temporal and spatial resolutions to ensure their temporal and spatial alignment with the temporal trajectory and spatial synergy relationships.

[0086] Finally, the time evolution trajectories, spatial coordination relationships, and coupling features, which have undergone feature dimension alignment, are input into the feature map construction unit. This unit then stitches and integrates these features to generate a multidimensional feature map with uniform temporal and spatial resolution. This multidimensional feature map incorporates the state evolution trajectories in the temporal dimension, the device coordination relationships in the spatial dimension, and the coupling characteristics between environmental interference and device states, comprehensively reflecting the dynamic operation of the distribution network.

[0087] Step S1255: performing noise filtering on the multidimensional feature map, removing redundant features irrelevant to the historical failure mode, and retaining core features that are strongly correlated with the failure risk as the final multidimensional feature map.

[0088] In this embodiment, since the multidimensional feature map may contain some redundant features that are irrelevant to the historical failure mode, these features will increase the complexity of subsequent analysis and may affect the accuracy of fault risk identification. Therefore, it is necessary to perform noise filtering on the multidimensional feature map.

[0089] The specific process of noise filtering is as follows: First, based on the historical fault knowledge base and previously extracted association rules, core features strongly correlated with fault risk are identified. These core features are those that frequently appear in historical fault patterns and have a significant impact on fault risk assessment. Examples include specific equipment operating parameters, environmental parameters, or inter-equipment connectivity.

[0090] Each feature in the multidimensional feature map is then evaluated to determine its relevance to the core features. This can be done by calculating the correlation coefficient between features. If a feature's correlation coefficient with the core features is low, it indicates a weak correlation with the fault risk and may be redundant.

[0091] Features deemed redundant are removed from the multidimensional feature map. This process retains the core features strongly associated with fault risk, resulting in a final multidimensional feature map. This final multidimensional feature map is more concise and effective, better reflecting the fault risk profile of the distribution network.

[0092] Step S130: Perform context-aware discrimination processing on the multi-dimensional feature map through a pre-built dynamic decision-making inference engine to generate a fault risk discrimination result of the distribution network, wherein the fault risk discrimination result includes a risk level identifier, a risk triggering period, and a risk impact range.

[0093] In this embodiment, a pre-built dynamic decision-making inference engine is used to perform in-depth analysis and processing of the multi-dimensional feature map to generate a fault risk assessment result for the distribution network. The engine can extract useful contextual information from the multi-dimensional feature map and use this information to perform fault risk assessment.

[0094] Step S131: input the multidimensional feature map into the context perception layer of the dynamic decision-making inference engine, and extract the state mutation points in the time evolution trajectory and the abnormal connection edges in the spatial collaborative relationship as context features.

[0095] In this embodiment, the main function of the context perception layer of the dynamic decision-making inference engine is to extract context features from the multi-dimensional feature map. The above context features can reflect the abnormal situation of the distribution network operation status.

[0096] For time-evolution trajectories, a state mutation point refers to a point in time where the device state undergoes an abrupt change. For example, a sudden and significant change in a device's operating parameters, such as voltage, current, and power, at a certain point in time may indicate a potential device failure. State mutation points can be identified by performing differential calculations on the time-evolution trajectory or by setting thresholds. For example, the difference in operating parameters between adjacent time points can be calculated. If the difference exceeds a certain threshold, the time point is considered a state mutation point.

[0097] For spatial collaboration relationships, abnormal edges are those whose weight or status changes abnormally compared to normal. For example, a sudden drop in the power transmission capacity of certain edges, or a change in the status of an edge from normal to abnormal, could indicate a problem with the collaborative working relationship between devices. Abnormal edges can be identified by comparing current spatial collaboration relationships with historically normal spatial collaboration relationships or by setting a threshold for edge weight changes.

[0098] The identified state mutation points and abnormal connection edges are used as context features, which can reflect the abnormal conditions of the distribution network in the temporal and spatial dimensions.

[0099] Step S132: Input the context features into the risk assessment layer of the dynamic decision-making reasoning engine, and calculate the risk probability value of the device node through fuzzy logic reasoning processing. The risk probability value is determined by the severity of the state mutation point and the influence range of the abnormal connection edge.

[0100] In this embodiment, the risk assessment layer of the dynamic decision-making inference engine uses fuzzy logic to process contextual features and calculate the risk probability value of the device node. Fuzzy logic is a reasoning method that can handle uncertainty and ambiguity, making it suitable for problems with a certain degree of uncertainty, such as fault risk assessment.

[0101] Step S1321: performing severity assessment processing on the state mutation points in the context features, and dividing the state mutation points into multiple state mutation levels according to the numerical values ​​of the abnormal working condition events.

[0102] In this embodiment, severity assessments are performed on state mutation points to more accurately measure the impact of these mutations on device operation. Each state mutation point is assessed based on the magnitude of the abnormal operating condition event. The magnitude of the abnormal operating condition event can be the magnitude of the change in a device operating parameter, such as voltage or current.

[0103] A series of thresholds are set, and based on the comparison of the numerical value of the abnormal operating condition event with these thresholds, the state mutation point is divided into multiple state mutation levels. For example, three thresholds are set. When the numerical value of the abnormal operating condition event is less than the first threshold, the state mutation point is classified as a low-level state mutation; when the numerical value of the abnormal operating condition event is between the first and second thresholds, the state mutation point is classified as a medium-level state mutation; when the numerical value of the abnormal operating condition event is greater than the second threshold, the state mutation point is classified as a high-level state mutation. Different state mutation levels reflect the different severity of the state mutation. A high-level state mutation indicates that the state mutation has a greater impact on equipment operation and may cause a higher risk of equipment failure.

[0104] Step S1322: perform an impact range assessment on the abnormal connection edges in the context features, and divide the abnormal connection edges into multiple abnormal connection edge levels according to the degree of reduction of the edge weights. The abnormal connection edge levels include local impact level, regional impact level and global impact level.

[0105] In this embodiment, the impact range assessment process for abnormal connection edges is performed to determine the extent and scope of their impact on the distribution network. Each abnormal connection edge is evaluated based on the degree of reduction in edge weight. This reduction in edge weight reflects the degradation of power transmission capacity between devices.

[0106] The different edge weight reduction thresholds are set, and the abnormal connection edges are divided into multiple abnormal connection edge levels according to the comparison results of the edge weight reduction degrees of the abnormal connection edges and the thresholds. For example, when the edge weight reduction degree is less than a lower threshold, the abnormal connection edge is divided into a local influence level, indicating that the abnormal connection edge only has an influence on the local device operation; when the edge weight reduction degree is between the lower threshold and a higher threshold, the abnormal connection edge is divided into a regional influence level, indicating that the abnormal connection edge has an influence on the device operation in a region; and when the edge weight reduction degree is greater than the higher threshold, the abnormal connection edge is divided into a global influence level, indicating that the abnormal connection edge has a great influence on the operation of the entire power distribution network.

[0107] Step S1323: A fuzzy reasoning rule base is constructed to define the influence relationship of the combination of the state mutation level and the abnormal connection edge level on the risk probability value.

[0108] In this embodiment, the fuzzy reasoning rule base is constructed to determine the influence relationship of the different combinations of the state mutation level and the abnormal connection edge level on the device node risk probability value. The fuzzy reasoning rule base is constructed based on expert knowledge and historical fault data.

[0109] According to the different combinations of the state mutation level and the abnormal connection edge level, a series of fuzzy rules are defined. For example, if the state mutation level is a high level and the abnormal connection edge level is a global influence level, the risk probability value of the device node is high; and if the state mutation level is a low level and the abnormal connection edge level is a local influence level, the risk probability value of the device node is low. These fuzzy rules are determined by analyzing and summarizing a large amount of historical fault data and combining the experience and judgment of experts.

[0110] Step S1324: The state mutation level and the abnormal connection edge level are input into the fuzzy reasoning rule base, and the risk probability value of the device node is calculated through fuzzification, rule matching and defuzzification processing.

[0111] In this embodiment, after the state mutation level and the abnormal connection edge level are input into the fuzzy reasoning rule base, fuzzification, rule matching and defuzzification processing are required to calculate the risk probability value of the device node.

[0112] First, fuzzification processing is performed. Since the state mutation level and the abnormal connection edge level are discrete level values, they need to be converted into membership values in fuzzy sets. For example, for the state mutation level being a high level, it can be converted into a membership value of 1 in the “high risk state mutation” fuzzy set and a membership value of 0 in other fuzzy sets.

[0113] Next, rule matching is performed. Based on the rules in the fuzzy inference rule base, the fuzzified state mutation level and the abnormal connection edge level are matched to identify applicable rules. For each applicable rule, its excitation strength is calculated based on the membership value in the rule's precondition.

[0114] Finally, defuzzification is performed. The excitation strengths and conclusions of all applicable rules are combined, and a defuzzification method (such as the centroid method) is used to calculate the risk probability value of the device node. This risk probability reflects the likelihood of a device node failure and is determined by the severity of the state mutation point and the impact range of the abnormal connection edge.

[0115] Step S133: inputting the risk probability value into the level classification layer of the dynamic decision-making inference engine, classifying the device node into multiple risk levels according to a preset risk level threshold, and generating a risk level identifier.

[0116] In this embodiment, the level classification layer of the dynamic decision-making inference engine classifies the risk probability values ​​of the device nodes according to a preset risk level threshold to determine the risk level of the device nodes.

[0117] Preset risk level thresholds are a series of numerical values ​​set based on historical failure data and actual experience. For example, three risk level thresholds can be set: low risk threshold, medium risk threshold, and high risk threshold. When the risk probability value of a device node is less than the low risk threshold, the device node is classified as low risk; when the risk probability value of a device node is between the low risk threshold and the medium risk threshold, the device node is classified as medium risk; and when the risk probability value of a device node is greater than the medium risk threshold, the device node is classified as high risk.

[0118] The risk level of each device node is represented by an identifier, such as letters or numbers, to form a risk level identifier. The risk level identifier can intuitively reflect the failure risk level of the device node.

[0119] Step S134: performing timestamp analysis on the state mutation points in the time evolution trajectory to determine the start time and end time of the risk event and generate a risk triggering period.

[0120] In this embodiment, by performing timestamp analysis on the state mutation points in the time evolution trajectory, the start time and end time of the risk event are determined, thereby generating a risk triggering period.

[0121] For each state mutation point, its corresponding timestamp records the time when the state mutation occurred. Analyze the timestamps of adjacent state mutation points to identify the starting and ending state mutation points of the risk event. The starting state mutation point is the time when the risk event begins to show abnormalities, and the ending state mutation point is the time when the abnormal state of the risk event ends.

[0122] The timestamp of the initial state mutation point is used as the start time of the risk event, and the timestamp of the final state mutation point is used as the end time of the risk event. These two time points constitute the risk trigger period. The risk trigger period can clearly define the time range in which the risk event occurs.

[0123] Step S135: performing regional coverage analysis on the abnormal connection edges in the spatial collaborative relationship, determining the range of device nodes affected by the risk event, and generating a risk impact range.

[0124] In this embodiment, the region coverage analysis and processing of abnormal connection edges in the spatial collaborative relationship is performed in order to determine the range of device nodes affected by the risk event and generate the risk impact range.

[0125] First, the area covered by the abnormal edge is determined based on its location and connection relationships. For each abnormal edge, the device nodes connected to it are identified. Then, with these device nodes as the center, the affected device nodes are determined based on predefined rules (such as distance thresholds or connection relationships).

[0126] For example, a distance threshold may be set, and for a device node connected to an abnormal connection edge, all other device nodes within the distance threshold range from the node are found. These device nodes constitute the device node range affected by the risk event.

[0127] The affected device node ranges are organized and identified to generate a risk impact range, which can intuitively demonstrate the spatial impact and scope of the risk event.

[0128] Step S136: combining the risk level identifier, the risk triggering period, and the risk impact range to form a fault risk identification result including the risk level identifier, the risk triggering period, and the risk impact range.

[0129] In this embodiment, the generated risk level identifier, risk trigger period, and risk impact range are combined to form a complete fault risk identification result. This result can comprehensively reflect the fault risk situation of device nodes in the distribution network, including the fault risk level, occurrence time range, and impact spatial range.

[0130] Step S140: performing a causal link tracking process on the target state data set based on the fault risk discrimination result, to generate a fault section positioning result of the power distribution network, the fault section positioning result including a risk propagation path, a key influence node, and an influence degree quantization value.

[0131] In this embodiment, the causal link tracking process is performed on the target state data set based on the fault risk discrimination result, aiming to find out the cause and propagation path of the fault, determine the key influence node and the influence degree quantization value, and generate the fault section positioning result.

[0132] Step S141: if the risk level corresponding to the risk level identifier in the fault risk discrimination result is higher than a set risk level, extracting a state data subset in the target state data set that overlaps with a risk triggering time period.

[0133] In this embodiment, it is first determined whether the risk level corresponding to the risk level identifier in the fault risk discrimination result is higher than a set risk level. The set risk level is a threshold value set according to actual requirements and safety standards of the power distribution network. When the risk level is higher than the threshold value, it indicates that the equipment has a high fault risk, and further analysis and processing are needed.

[0134] If the risk level is higher than the set risk level, a state data subset that overlaps with a risk triggering time period is extracted from the target state data set. The target state data set includes equipment operating condition data, network topology structure data, historical fault knowledge base, and environmental interference data. According to the risk triggering time period, the equipment operating condition data, network topology structure data, and environmental interference data recorded in the time period are filtered out to form the state data subset. The state data subset can reflect the actual operating state of the power distribution network during the risk event.

[0135] Step S142: performing a causal relationship mining process on the state data subset to construct a fault causal graph including equipment nodes and causal relationship edges, a weight of the causal relationship edge being determined by a mutual information value of operating parameters between equipment nodes.

[0136] In this embodiment, the causal relationship mining process is performed on the state data subset to find out the causal relationship between the equipment nodes and construct the fault causal graph.

[0137] Step S1421: performing a Granger causality test process on the equipment operating parameters in the state data subset to determine a causal relationship direction between the equipment nodes, the causal relationship direction indicating whether a state change of one node will cause a state change of another node.

[0138] In this embodiment, Granger causality testing is a key step in determining the direction of causal relationships between device nodes. In the state data subset, each device node has operating parameters in multiple dimensions, such as voltage, current, and power. These parameters change over time, forming time series data.

[0139] First, consider the time series of operating parameters of any two device nodes, denoted as parameter sequence A and parameter sequence B. The core concept of Granger causality testing is based on the predictive nature of time series. If the past values ​​of parameter sequence A can help predict the future values ​​of parameter sequence B, while the past values ​​of parameter sequence B cannot help predict the future values ​​of parameter sequence A, then parameter sequence A can be considered to Granger cause parameter sequence B. This reflects the direction of the possible causal relationship between the device nodes at the parameter level.

[0140] In the specific operation, two regression models are constructed. The first regression model uses only the past values ​​of parameter sequence B to predict the current value of parameter sequence B, denoted as model M1. The second regression model uses the past values ​​of parameter sequence A and parameter sequence B to predict the current value of parameter sequence B, denoted as model M2.

[0141] For model M1, the model parameters are estimated using methods such as least squares to minimize the model's fitting error to parameter sequence B. For model M2, the parameters are also estimated using least squares. The residual sums of squares for the two models are then calculated and recorded as RSS1 and RSS2, respectively.

[0142] Next, we use the F-test to determine whether Model M2 is significantly better than Model M1. The F-test statistic is calculated by dividing (RSS1 - RSS2) by the combined value of RSS2 and the degrees of freedom. The degrees of freedom here are related to the number of parameters in the model and the number of samples. The calculated F-test statistic is compared to the critical value of the F-distribution for a given significance level.

[0143] If the F-test statistic is greater than the critical value, it means that model M2 is significantly better than model M1, that is, the past values ​​of parameter sequence A have a significant contribution to the prediction of the current value of parameter sequence B. At this time, it can be considered that parameter sequence A is the Granger cause of parameter sequence B, indicating that at the device node level, the state change of the device node corresponding to parameter sequence A may cause the state change of the device node corresponding to parameter sequence B, thereby determining a causal relationship direction between the two device nodes.

[0144] Such Granger causality tests are performed on the operating parameters of all device nodes in the status data subset, and complete causal relationship direction information between device nodes is obtained.

[0145] Step S1422: Calculate the mutual information value of the operating parameters of the device node pairs with a causal relationship, where the mutual information value represents the correlation strength of the state changes between the device nodes.

[0146] After determining the direction of the causal relationship between device nodes, it is necessary to calculate the mutual information value of the operating parameters of the device node pairs with causal relationship to measure the correlation strength of the state changes between the device nodes.

[0147] For the time series of operating parameters of two causally related device nodes, let them be parameter sequence X and parameter sequence Y. Mutual information is a metric that measures the correlation between two random variables. Based on the concepts of information theory, it reflects the extent to which the information of one random variable can reduce the uncertainty of another random variable.

[0148] First, the parameter sequences X and Y are discretized. Because continuous operating parameters are not conducive to directly calculating mutual information, the range of the parameter sequences is divided into several intervals, each corresponding to a discrete state. For example, for the voltage parameter, its range can be divided into low voltage interval, normal voltage interval, and high voltage interval.

[0149] Then, the joint probability distribution and marginal probability distribution of parameter sequence X and parameter sequence Y in each discrete state are statistically analyzed. The joint probability distribution represents the probability that parameter sequence X and parameter sequence Y are in two discrete states at the same time, while the marginal probability distribution represents the probability that parameter sequence X and parameter sequence Y are in a discrete state respectively.

[0150] The mutual information value is calculated based on the joint probability distribution and the marginal probability distribution. The mutual information value is calculated by multiplying the joint probability of each state combination in the joint probability distribution by the ratio of the logarithm of the product of the joint probability and the marginal probability, and then summing the values ​​for all state combinations. The result is the mutual information value of the parameter sequence X and the parameter sequence Y.

[0151] The larger the mutual information value, the stronger the correlation between the operating parameters of the two device nodes, that is, the closer the association between the state change of one device node and the state change of another device node.

[0152] Step S1423: generating directed causal relationship edges between device nodes according to the causal relationship direction and the mutual information value, wherein the direction of the edge is consistent with the causal relationship direction, and the weight of the edge is the mutual information value.

[0153] According to the causal relationship direction and mutual information value between the device nodes obtained above, a directed causal relationship edge between the device nodes is generated.

[0154] For each pair of device nodes with a causal relationship, the direction of the directed causal edge is determined according to the direction of the causal relationship. For example, if the Granger causality test determines that the state change of device node A will cause the state change of device node B, then the direction of the directed causal edge is from device node A to device node B.

[0155] At the same time, the mutual information value of the operating parameters of the two device nodes is used as the weight of the directed causal relationship edge. The larger the mutual information value, the stronger the causal relationship between the two device nodes, which is reflected as a larger edge weight in the directed causal relationship edge.

[0156] In the above manner, a directed causal edge is generated for each pair of device nodes with a causal relationship. These edges reflect the causal relationship between the device nodes and the strength of the relationship.

[0157] Step S1424: All device nodes and directed causal relationship edges are assembled into a graph structure to generate a fault causal graph that reflects the causal relationship between the device nodes.

[0158] After all directed causal edges are generated, all device nodes and directed causal edges are assembled into a graph structure to generate a fault causal graph that reflects the causal relationship between device nodes.

[0159] First, each device node is considered as a vertex in the graph, and directed causal edges are considered as directed edges in the graph. Then, the vertices and directed edges are connected and laid out according to the actual connections and causal relationships of the device nodes.

[0160] During assembly, ensure that the graph structure accurately reflects the causal relationships between device nodes. For example, if multiple device nodes have causal relationships with a central device node, these connections should be correctly reflected in the graph. Furthermore, ensure that the directions and weights of directed edges are consistent with the causal relationship directions and mutual information values ​​calculated previously.

[0161] Through the above graph structure assembly processing, a complete fault causal graph is obtained, which intuitively shows the causal relationship and the strength of the relationship between device nodes in the distribution network.

[0162] Step S1425: performing a sparse processing on the fault causal graph, and removing weak causal relationship edges whose mutual information values ​​are lower than a preset threshold.

[0163] In order to make the fault causal graph more concise and effective, it is necessary to perform sparse processing on it and remove weak causal edges whose mutual information values ​​are lower than the preset threshold.

[0164] The preset threshold is a standard set based on actual conditions and experience to distinguish strong causal relationships from weak causal relationships. For each directed causal edge in the fault causal graph, its edge weight (i.e., mutual information value) is checked.

[0165] If the mutual information value of a directed causal edge is lower than a preset threshold, it indicates that the causal relationship between the device nodes represented by this edge is weak and has little impact on fault analysis. Such weak causal edges are removed from the fault causal graph.

[0166] Through the sparsification process, strong causal edges with high mutual information values ​​are retained, making the fault causal graph clearer and highlighting the main causal relationships between device nodes.

[0167] Step S143: starting from the significant risk device node in the fault causality graph, traverse all reachable nodes using a depth-first search algorithm to generate a set of risk propagation path candidates.

[0168] After obtaining the sparse fault causal graph, starting from the significant risk device node, the depth-first search algorithm is used to traverse all reachable nodes to generate a candidate set of risk propagation paths.

[0169] Significant risk device nodes are determined based on the previous fault risk identification results. These nodes have a higher risk level and may be the source of the fault or the node that is most affected by the fault.

[0170] The depth-first search algorithm is used to traverse a graph. Starting from a node with a significant risk, it first visits that node, then selects an unvisited directed edge connected to it and follows that edge to visit the next node. This process is repeated for the next node until no further progress is possible or all reachable nodes have been visited.

[0171] During the traversal process, the path from the starting point to each reachable node is recorded. Each recorded path is a candidate risk propagation path.

[0172] By performing such a depth-first search traversal on all significant risk device nodes, a set of candidate risk propagation paths is obtained. The candidate risk propagation path set includes possible risk propagation paths starting from the significant risk device nodes.

[0173] Step S144: performing weighted screening on the risk propagation path candidate set, and retaining risk propagation paths with mutual information values ​​exceeding a preset threshold as valid risk propagation paths.

[0174] After obtaining the candidate set of risk propagation paths, it is necessary to perform weight screening on them to determine the effective risk propagation paths.

[0175] For each path in the candidate set of risk propagation paths, calculate the mutual information value of all directed causal edges on the path. The mutual information value on the path can be synthesized by summing or averaging.

[0176] The calculated comprehensive mutual information value is then compared with a preset threshold. This threshold is a criterion for distinguishing valid paths from invalid ones. If the comprehensive mutual information value of a path exceeds the preset threshold, it indicates that the risk propagation relationship represented by this path is strong and may be a true risk propagation path. It is retained as a valid risk propagation path.

[0177] Through the above-mentioned weight screening process, effective risk propagation paths are screened out from the risk propagation path candidate set, and these effective risk propagation paths are more likely to be actual risk propagation paths.

[0178] Step S145: Identify the node with the highest degree of abnormality of the operating parameters on the effective risk propagation path as the key influencing node, where the degree of abnormality is determined by the magnitude of deviation of the current, voltage, and temperature from the reference value.

[0179] After obtaining effective risk propagation paths, it is necessary to identify the nodes with the highest degree of abnormality in operating parameters on these paths as key influencing nodes.

[0180] For each node on the effective risk propagation path, its operating parameters such as current, voltage, and temperature are monitored. Each operating parameter has a baseline value, which can be the rated parameter of the equipment or the average parameter during historical normal operation.

[0181] Calculate the deviation of each node's operating parameters, such as current, voltage, and temperature, from their baseline values. For example, for current, subtract the baseline current value from the current value and take the absolute value to determine the deviation. Repeat the same process for voltage and temperature.

[0182] Then, the deviation magnitudes of these operating parameters are comprehensively evaluated. A weighted summation approach can be used to assign different weights to different operating parameters, as they may have varying degrees of impact on device failures. For example, current parameters may have a greater impact on device overload failures and therefore be assigned a higher weight; temperature parameters may have a greater impact on the device's heat dissipation and insulation performance and therefore also be assigned a set weight.

[0183] Through comprehensive evaluation, the abnormality level of each node's operating parameters is determined. On the effective risk propagation path, the node with the highest abnormality level is identified and designated as the key impact node. This key impact node may be the key source of the fault or the node with the greatest impact on fault propagation.

[0184] Step S146: Calculate the quantitative value of the degree of influence of the key influencing node on the risk propagation path, and form a fault section positioning result including the risk propagation path, the key influencing node and the quantitative value of the degree of influence. The quantitative value of the degree of influence is jointly determined by the position of the key influencing node in the risk propagation path and the degree of abnormality of the operating parameters.

[0185] After determining the key influencing nodes, it is necessary to calculate the quantitative value of the impact of the key influencing nodes on the risk propagation path to form a complete fault section location result.

[0186] The quantitative impact value is determined by the location of the key impact node in the risk propagation path and the degree of abnormality in the operating parameters. Regarding the location of the key impact node in the risk propagation path, nodes closer to the starting point of the risk propagation path may have a greater impact on the entire path because they may be the source of the fault or the first node to be affected by the fault. The location impact factor can be determined based on the distance from the key impact node to the starting point of the risk propagation path. For example, the closer the distance to the starting point, the greater the location impact factor.

[0187] The degree of abnormality of the operating parameters of the key influencing nodes has been calculated previously to obtain an abnormality value. The location impact factor and the abnormality value of the operating parameters are combined, for example, using multiplication or weighted summation, to obtain a quantitative value of the impact of the key influencing node on the risk propagation path.

[0188] The risk propagation path, key impact nodes, and calculated impact quantification values ​​are combined to form a fault segment location result that includes the above information. This fault segment location result accurately identifies the segment where the fault may occur, the key impact nodes, and the degree of impact of the key impact nodes on fault propagation.

[0189] Step S150: Generate a fault location instruction including a path identifier, a node identifier, and an impact degree according to the fault section location result, and send the fault location instruction to the distribution network intelligent operation and maintenance platform to trigger a precise maintenance process.

[0190] After obtaining the fault section location result, a fault location instruction including path identification, node identification and impact degree is generated based on the result, and the fault location instruction is sent to the distribution network intelligent operation and maintenance platform to trigger the precise maintenance process.

[0191] Step S151: extract the risk propagation path from the fault section location result, and assign a unique path identifier to each risk propagation path according to the path coding rule preset in the distribution network. The path identifier includes a starting node number and an end node number.

[0192] First, the risk propagation path is extracted from the fault section location results. Then, a unique path identifier is assigned to each risk propagation path according to the path encoding rules preset in the distribution network.

[0193] The distribution network's preset path coding rules are designed to accurately identify and distinguish different risk propagation paths. Path identifiers include the start and end node numbers, which uniquely identify the starting and ending locations of a risk propagation path.

[0194] For example, the starting node number can be a combination of the device node's geographic location code and device type code, and the same applies to the ending node number. In this way, each risk propagation path is given a unique path identifier, which facilitates subsequent management and query.

[0195] Step S152: extracting key influencing nodes from the fault section location result, and assigning a unique node identifier to each key influencing node according to the node encoding rule preset in the distribution network, wherein the node identifier includes a node geographic location code and a device type code.

[0196] Next, the key influencing nodes are extracted from the fault section location results. According to the node coding rules preset in the distribution network, a unique node identifier is assigned to each key influencing node.

[0197] Node coding rules are developed based on the actual conditions of the distribution network to accurately identify each device node. Node identification includes both the node's geographic location code and the device type code. The geographic location code can use latitude and longitude information or regional division codes. The device type code distinguishes devices based on their function and type, such as transformers and circuit breakers.

[0198] By assigning a unique node identifier to each key influencing node, the key influencing node can be accurately located, making it easier for maintenance personnel to quickly find the faulty node.

[0199] Step S153: extracting the impact degree quantified value from the fault section location result, and converting the quantified value into an impact degree grade identifier according to a preset impact degree grading standard.

[0200] The impact degree quantified value is extracted from the fault section location result, and then converted into the impact degree grade identifier according to the preset impact degree grading standard.

[0201] The preset impact level classification standards are set based on actual experience and troubleshooting needs. They are used to classify continuous impact level values ​​into different levels. For example, the impact level values ​​can be divided into several intervals, each corresponding to an impact level, such as mild impact, moderate impact, and severe impact.

[0202] The extracted impact degree quantified value is compared with the grading standard to determine the grade interval to which it belongs, thereby obtaining the corresponding impact degree level identifier. This impact degree level identifier can intuitively reflect the impact degree of the key influencing node on the fault propagation.

[0203] Step S154: Slice the risk propagation path in the time dimension to generate a segmentation identifier for the risk propagation path within the risk triggering period, wherein the segmentation identifier includes the start time and end time of the time period.

[0204] In order to describe the risk propagation path more accurately, the risk propagation path is sliced ​​in the time dimension to generate segmented identification of the risk propagation path within the risk triggering period.

[0205] The risk trigger period is the time range within which the risk event determined in the previous fault risk identification results occurs. Based on the risk trigger period, the risk propagation path is divided according to time.

[0206] For example, the risk trigger period can be divided into several time periods according to the set time interval, and each time period corresponds to a segment of the risk propagation path. For each segment, its start time and end time are recorded to form a segment identifier.

[0207] Segment identification can show in detail the risk propagation path in different time periods, helping maintenance personnel analyze the development process and trend of faults.

[0208] Step S155: performing coordinate mapping processing on the key influencing node in a spatial dimension to generate a coordinate identifier of the key influencing node in the distribution network geographic information system, wherein the coordinate identifier includes longitude and latitude information.

[0209] The key influencing nodes are subjected to spatial coordinate mapping processing to generate coordinate identifications of the key influencing nodes in the distribution network geographic information system.

[0210] The distribution network GIS records the geographical location information of each device node in the distribution network. By querying the GIS, the longitude and latitude information of key influencing nodes are obtained and combined into coordinate identification.

[0211] Coordinate identification can accurately locate the position of key influencing nodes in geographic space, making it convenient for maintenance personnel to quickly find fault nodes during actual maintenance.

[0212] Step S156: performing association and binding processing on the path identifier, node identifier, impact level identifier, segment identifier and coordinate identifier to generate an identifier set containing multi-dimensional positioning information.

[0213] The previously generated path identifiers, node identifiers, impact level identifiers, segment identifiers, and coordinate identifiers are associated and bound to generate an identifier set containing multi-dimensional positioning information.

[0214] Binding is the process of integrating these different types of identification information to establish a corresponding relationship between them. For example, the path identifier of a risk propagation path is associated with the node identifier, impact level identifier, segment identifier, and coordinate identifier of the key impact node on the path.

[0215] Through the association and binding process, a complete identification set is obtained, which contains the multi-dimensional information required for fault location and can comprehensively and accurately describe the location, impact and development process of the fault.

[0216] Step S157: input the identification set into the instruction generation module, perform data encapsulation processing according to the communication protocol format of the distribution network intelligent operation and maintenance platform, and generate a fault location instruction including a path identification field, a node identification field and an impact degree field.

[0217] The generated identification set is input into the instruction generation module, and data encapsulation processing is performed according to the communication protocol format of the distribution network intelligent operation and maintenance platform to generate a fault location instruction.

[0218] The distribution network intelligent operation and maintenance platform has a compatible communication protocol format for receiving and processing externally sent commands. The command generation module organizes and encapsulates the information in the identification set according to the communication protocol format.

[0219] Place information such as path identifiers, node identifiers, and impact level identifiers into corresponding fields, such as the path identifier field, node identifier field, and impact level field. Through data encapsulation processing, the identifier set is converted into a fault location instruction that conforms to the communication protocol format of the distribution network intelligent operation and maintenance platform.

[0220] Step S158: Send the fault location instruction to the distribution network intelligent operation and maintenance platform to trigger a precise maintenance process.

[0221] After the generation of the identification set containing multi-dimensional positioning information is completed, the fault location instruction encapsulated by the identification set needs to be sent to the distribution network intelligent operation and maintenance platform to trigger the precise maintenance process.

[0222] First, the communication interface and protocol requirements of the distribution network intelligent operation and maintenance platform must be clearly defined. Different distribution network intelligent operation and maintenance platforms may have different communication interface types, such as serial communication and network communication, and each has its own specific communication protocol. These protocols specify data transmission formats, encoding methods, verification rules, and other details. Therefore, before sending fault location commands, they must be formatted and encoded accordingly to ensure they meet the communication protocol requirements of the distribution network intelligent operation and maintenance platform.

[0223] The data transmission format requires that information such as path identifiers, node identifiers, impact level identifiers, segment identifiers, and coordinate identifiers be arranged in the order and field length specified by the protocol. For example, the protocol may specify that the path identifier be placed at the beginning of the instruction, followed by the node identifier, and that each identifier has a fixed length. If the path identifier requires a code with a set number of bits, the previously generated path identifier must be padded or truncated to ensure that its length meets the protocol requirements.

[0224] In terms of encoding, it may be necessary to convert identification information into a specific encoding format, such as ASCII or binary. For example, if the protocol uses ASCII for data transmission, then the characters in information such as path identifiers and node identifiers must be converted to their corresponding ASCII code values. Furthermore, to ensure the accuracy of data transmission, a checksum must be added according to the protocol's verification rules. Common checksum methods include parity check and cyclic redundancy check (CRC). For example, a CRC checksum is calculated on the data portion of the fault location instruction to generate a checksum, which is then added to the end of the instruction.

[0225] After format conversion and encoding, the fault location command is sent to the distribution network intelligent operation and maintenance platform via the corresponding communication interface. If serial communication is used, the serial port parameters such as baud rate, data bits, and stop bits must be configured, and then the command is sent using the serial communication program. During this process, ensure the stability of the communication line to avoid data loss or transmission errors. If network communication is used, the command must be sent to the designated IP address and port number of the distribution network intelligent operation and maintenance platform via a network connection.

[0226] When the distribution network intelligent operation and maintenance platform receives a fault location instruction, it can parse it. This parsing process decodes the received coded data according to the rules of the communication protocol, restoring information such as path identifiers, node identifiers, impact level identifiers, segment identifiers, and coordinate identifiers. The platform then triggers a precise maintenance process based on this multi-dimensional location information.

[0227] The platform determines the path through which the fault may propagate according to the path identifier, which helps the maintenance personnel understand the scope and direction of the fault propagation. For example, through the path identifier, it can be known that the fault is propagated from which device node, through which intermediate nodes, and finally affects which terminal nodes.

[0228] The node identifier enables the maintenance personnel to accurately find the location of the key affected node. Combined with the coordinate identifier, the maintenance personnel can intuitively see the specific geographic location of the key affected node in the power distribution network geographic information system and quickly reach the fault site. At the same time, the impact degree identifier provides the maintenance personnel with the impact degree information of the key affected node on the fault propagation, enabling them to reasonably arrange the maintenance resources and priorities according to the impact degree.

[0229] The segmentation identifier records the segmentation of the risk propagation path during the risk triggering period, which helps the maintenance personnel analyze the development process of the fault. For example, by viewing the start time and end time in the segmentation identifier, the maintenance personnel can understand the propagation speed and change trend of the fault in different time periods, thereby formulating more effective maintenance strategies.

[0230] After receiving the fault location instruction, the smart operation and maintenance platform of the distribution network can automatically generate a maintenance task sheet containing detailed information and maintenance requirements of the fault according to the above information. The maintenance personnel can obtain the maintenance task sheet through a mobile terminal device or the operation interface of the platform and perform accurate maintenance according to the requirements of the task sheet, thereby improving the maintenance efficiency and reducing the impact of the fault on the operation of the power distribution network.

[0231] In the above embodiments, the processing of the target state data set collected by the global intelligent sensing network of the power distribution network is involved, which may contain some privacy-sensitive data, such as special operating parameters of some devices, environmental data of specific areas, etc. In order to protect these privacy-sensitive data, anonymization processing technology is adopted during data collection. For the collected device operating condition data and environmental interference data, the information that can directly or indirectly identify individuals or specific devices is removed. For example, for the operating parameters of the device, the specific number or the name of the unit to which the device belongs is not recorded, but a randomly generated identifier is used instead. In this way, during the subsequent data processing and analysis process, even if the data is accidentally obtained, it is also impossible to trace back to the specific device or individual through these data.

[0232] During data transmission, data is encrypted using either symmetric or asymmetric encryption algorithms. In symmetric encryption algorithms, the same key is used for both encryption and decryption. For example, with the AES encryption algorithm, a single key is used at the data sender to encrypt data into ciphertext, and the same key is used at the receiver to decrypt the ciphertext into plaintext. Asymmetric encryption algorithms use a pair of keys: a public key and a private key. The data sender encrypts data using the receiver's public key, and the receiver decrypts it using its own private key. This ensures data security during transmission and prevents data theft or tampering.

[0233] Regarding data storage, privacy-sensitive data is stored in a dedicated database with strict access controls. Only authorized personnel can access this data, and access rights are carefully divided based on personnel responsibilities and needs. For example, maintenance personnel can only access equipment operating parameters related to maintenance tasks and are not allowed to access other sensitive environmental data. Furthermore, the database is regularly backed up to prevent data loss or corruption.

[0234] During data processing and analysis, differential privacy technology protects data privacy by adding noise to the data. During data analysis, a certain amount of noise is added to the input data, preventing the analysis results from revealing the specific information of individual data points. For example, when calculating the statistical values ​​of device operating parameters, random noise is added to each data point before the statistical calculations are performed. This prevents an attacker from inferring the specific content of the original data, even if they gain access to the analysis results.

[0235] The construction and training process of the dynamic decision-making inference engine is as follows:

[0236] The dynamic decision-making reasoning engine mainly includes three core modules: context perception layer, risk assessment layer and level classification layer.

[0237] The context-aware layer is the input layer of the dynamic decision-making inference engine. Its primary function is to extract contextual features from the multidimensional feature map. This layer can be composed of multiple submodules, such as the state mutation point extraction submodule and the abnormal connection edge extraction submodule. The state mutation point extraction submodule is responsible for identifying state mutation points from the time evolution trajectory. This is achieved by performing operations such as differential calculation and threshold judgment on time series data. The abnormal connection edge extraction submodule identifies abnormal connection edges from spatial collaborative relationships and determines abnormal connection edges by comparing changes in edge weights. The outputs of these two submodules are passed as contextual features to the risk assessment layer.

[0238] The risk assessment layer is the core computing layer of the dynamic decision-making inference engine. It receives contextual features output by the context perception layer and calculates the risk probability value of the device node through fuzzy logic reasoning. This layer includes a state mutation point severity assessment submodule, an abnormal connection edge impact range assessment submodule, a fuzzy reasoning rule base, and a fuzzy reasoning calculation submodule. The state mutation point severity assessment submodule classifies state mutation points according to the numerical value of the abnormal operating condition event, and the abnormal connection edge impact range assessment submodule classifies abnormal connection edges according to the degree of reduction in edge weight. The fuzzy reasoning rule base stores the relationship between the combination of state mutation level and abnormal connection edge level and the risk probability value. The fuzzy reasoning calculation submodule calculates the risk probability value of the device node based on the input state mutation level and abnormal connection edge level through fuzzification, rule matching, and defuzzification.

[0239] The grading layer is the output layer of the dynamic decision-making inference engine. It divides the risk probability values ​​output by the risk assessment layer according to preset risk level thresholds to determine the risk level of the device node. This layer can be composed of a simple threshold comparison submodule that compares the risk probability value with different thresholds to determine the risk level of the device node.

[0240] The above modules are connected in sequence according to the hierarchical structure. The output of the context perception layer serves as the input of the risk assessment layer, and the output of the risk assessment layer serves as the input of the classification layer, forming a complete reasoning process.

[0241] Before training the dynamic decision-making inference engine, a large amount of training data is required. This training data consists of multidimensional feature graph samples and corresponding fault risk identification result labels. Multidimensional feature graph samples are obtained by performing spatiotemporal semantic fusion processing on historical target state data sets. Fault risk identification result labels can be annotated based on historical fault records and expert experience.

[0242] The first step is to initialize the parameters of the dynamic decision-making inference engine. For the submodules of the context perception layer, initialize the threshold parameters for the state mutation point extraction submodule and the abnormal connection edge extraction submodule. For example, set the differential threshold for state mutation point extraction and the edge weight change threshold for abnormal connection edge extraction. For the fuzzy inference rule base of the risk assessment layer, initialize the rule premise and conclusion parameters. For example, set the thresholds for the state mutation level and abnormal connection edge level, as well as the parameters for calculating the rule's excitation strength. For the level classification layer, initialize the risk level threshold.

[0243] Second, input the multi-dimensional feature graph samples in the training data into the context-aware layer to extract context features. According to the initialized threshold parameters, the state mutation point extraction submodule and the abnormal connection edge extraction submodule process the time evolution trajectory and the spatial coordination relationship in the multi-dimensional feature graph respectively to obtain the context features.

[0244] Third, input the context features into the risk assessment layer to calculate the risk probability value. The state mutation severity assessment submodule and the abnormal connection edge influence range assessment submodule evaluate the context features respectively to obtain the state mutation level and the abnormal connection edge level. Then, the fuzzy reasoning calculation submodule performs fuzzification, rule matching and defuzzification on the state mutation level and the abnormal connection edge level according to the rules in the fuzzy reasoning rule base to calculate the risk probability value of the device node.

[0245] Fourth, input the risk probability value into the level division layer to divide the risk level. The threshold comparison submodule compares the risk probability value with the preset risk level threshold to determine the risk level identifier of the device node.

[0246] Fifth, calculate the training error. Compare the risk level identifier output by the level division layer with the fault risk judgment result label in the training data to calculate the error. The error can be calculated using methods such as cross-entropy loss function, mean square error, etc.

[0247] Sixth, update the parameters according to the error. Use optimization algorithms such as stochastic gradient descent algorithm, Adagrad algorithm, etc. to update the parameters of the dynamic decision-making reasoning engine. For example, adjust the threshold parameters of the context-aware layer, the fuzzy reasoning rule base parameters of the risk assessment layer and the risk level threshold of the level division layer according to the size and direction of the error, so that the error gradually decreases.

[0248] Repeat the process of the second to sixth steps until the training error reaches the preset threshold or the number of training times reaches the preset upper limit.

[0249] During the training process, some training parameters need to be set, such as learning rate, training batch size, training rounds, etc. The learning rate controls the step size of parameter update. Too large learning rate may cause unstable parameter update, and too small learning rate will cause slow training speed. The training batch size determines the number of samples used for each training. A suitable batch size can improve training efficiency. The number of training rounds indicates the number of times the entire training data is trained. Enough training rounds can ensure that the model learns the features and rules in the data sufficiently.

[0250] With this design, the dynamic decision-making reasoning engine can learn the intrinsic correlation between the multi-dimensional feature map and the fault risk judgment results, so as to accurately judge the fault risk of the distribution network in practical applications.

[0251] Figure 2 A schematic diagram illustrates exemplary hardware and software components of a distribution network fault location system 100 based on intelligent decision-making, which can implement the concepts of the present application, provided in some embodiments of the present application. For example, a processor 120 can be used in the distribution network fault location system 100 based on intelligent decision-making and perform the functions of the present application.

[0252] The intelligent decision-making-based distribution network fault location system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the intelligent decision-making-based distribution network fault location method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0253] For example, the intelligent decision-making-based distribution network fault location system 100 may include a network port 110 connected to the network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the intelligent decision-making-based distribution network fault location system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application may be implemented based on these program instructions. The intelligent decision-making-based distribution network fault location system 100 also includes an I / O interface 150 between the computer and other input and output devices.

[0254] For ease of explanation, only one processor is described in the intelligent decision-making-based distribution network fault location system 100. However, it should be noted that the intelligent decision-making-based distribution network fault location system 100 in this application may also include multiple processors, so the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the intelligent decision-making-based distribution network fault location system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually in one processor. For example, the first processor performs step A and the second processor performs step B, or the first processor and the second processor perform steps A and B together.

[0255] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the above-mentioned distribution network fault location method based on intelligent decision-making is implemented.

[0256] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A distribution network fault location method based on intelligent decision-making, characterized in that: The method comprises: Obtaining a target state data set collected by the global intelligent perception network of the distribution network, wherein the target state data set includes equipment operating condition data, network topology data, historical fault knowledge base, and environmental interference data; Performing spatiotemporal semantic fusion processing on the target state data set to generate a multidimensional feature map reflecting the dynamic operation status of the distribution network, wherein the multidimensional feature map includes the state evolution trajectory in the time dimension and the equipment coordination relationship in the space dimension; The multi-dimensional feature map is subjected to context-aware discrimination processing by a pre-built dynamic decision-making inference engine to generate a fault risk discrimination result of the distribution network, wherein the fault risk discrimination result includes a risk level identifier, a risk triggering period, and a risk impact range; Performing causal link tracing processing on the target state data set based on the fault risk identification result to generate a fault section location result of the distribution network, wherein the fault section location result includes a risk propagation path, key impact nodes, and a quantitative value of the impact degree; Generate a fault location instruction including a path identifier, a node identifier, and an impact degree according to the fault section location result, and send the fault location instruction to the distribution network intelligent operation and maintenance platform to trigger a maintenance process; The performing spatiotemporal semantic fusion processing on the target state data set to generate a multi-dimensional feature map reflecting the dynamic operation status of the distribution network includes: Performing time window alignment processing on the equipment operating condition data in the target state data set to obtain a condition data sequence with a time synchronization relationship, wherein the condition data sequence includes time-series sampling values ​​of operating parameters in multiple dimensions; Performing graph model conversion processing on the network topology data in the target state data set to generate a distribution network topology graph including device nodes and connection edges, wherein the weights of the connection edges are determined by the power transmission capacity between devices; Performing spatial region division processing on the environmental interference data in the target state data set to generate an environmental parameter distribution matrix for the area where the device is located, the environmental parameter distribution matrix including spatial sampling values ​​of environmental parameters in multiple dimensions; The operating condition data sequence, distribution network topology map and environmental parameter distribution matrix are input into the semantic fusion module, and the association rules between the equipment operating status and historical fault modes are extracted through semantic matching processing of the historical fault knowledge base; Based on the association rules, the operating condition data sequence, the distribution network topology map and the environmental parameter distribution matrix are subjected to feature cross-fusion processing to generate a multi-dimensional feature map containing time evolution trajectory and spatial collaborative relationship; specifically, the process includes: Performing sliding window analysis on the operating condition data sequence in the time dimension, counting the frequency of abnormal operating condition events within the sliding window, and generating a time evolution trajectory of the equipment status; Performing a spatial dimension path search process on the distribution network topology map, starting from a significant load device node, traversing all reachable nodes and recording edge weights on the path to generate a spatial collaborative relationship between devices; Performing spatiotemporal joint analysis on the environmental parameter distribution matrix, counting the co-occurrence frequency of environmental parameters and equipment faults in the target area, and generating coupling characteristics of environmental interference and equipment status; Inputting the time evolution trajectory, spatial synergy relationship and coupling features into a feature map construction unit, and generating a multi-dimensional feature map with uniform time resolution and spatial resolution through feature dimension alignment processing; The multidimensional feature map is subjected to noise filtering processing to remove redundant features that are irrelevant to historical failure modes, and the core features that are strongly correlated with failure risks are retained as the final multidimensional feature map.

2. The distribution network fault location method based on intelligent decision-making according to claim 1 is characterized in that: The operating condition data sequence, distribution network topology map and environmental parameter distribution matrix are input into the semantic fusion module, and the association rules between the equipment operating status and historical failure modes are extracted through semantic matching processing of the historical failure knowledge base, including: Performing feature extraction processing on the operating condition data sequence to generate an operating condition feature vector containing abnormal operating condition events; Performing neighborhood feature aggregation processing on the distribution network topology graph to generate a topological feature vector including node degree, edge weight, and clustering coefficient; Performing statistical analysis on the environmental parameter distribution matrix to generate an environmental feature vector including humidity mean, temperature variance, and electromagnetic interference peak value; Inputting the working condition feature vector, topology feature vector and environment feature vector into the semantic matching unit of the historical fault knowledge base, and calculating the cosine similarity between each feature vector and the historical fault mode feature vector; According to the cosine similarity value, historical fault patterns whose similarity exceeds a preset threshold are screened out, and their corresponding equipment state evolution laws and fault triggering conditions are extracted as association rules.

3. The distribution network fault location method based on intelligent decision-making according to claim 1 is characterized in that: The pre-built dynamic decision-making inference engine performs context-aware discrimination processing on the multi-dimensional feature map to generate a fault risk discrimination result of the distribution network, including: Inputting the multidimensional feature map into the context perception layer of the dynamic decision-making inference engine, extracting the state mutation points in the time evolution trajectory and the abnormal connection edges in the spatial collaborative relationship as context features; Input the context features into the risk assessment layer of the dynamic decision-making inference engine, and calculate the risk probability value of the device node through fuzzy logic reasoning. The risk probability value is determined by the severity of the state mutation point and the influence range of the abnormal connection edge; Inputting the risk probability value into the level classification layer of the dynamic decision-making inference engine, classifying the device node into multiple risk levels according to a preset risk level threshold, and generating a risk level identifier; Performing timestamp analysis on the state mutation points in the time evolution trajectory to determine the start and end times of the risk event and generate a risk trigger period; Perform regional coverage analysis on the abnormal connection edges in the spatial collaborative relationship to determine the range of device nodes affected by the risk event, generate the risk impact range, and form a fault risk judgment result including risk level identification, risk triggering period and risk impact range.

4. The distribution network fault location method based on intelligent decision-making according to claim 3 is characterized in that: The step of inputting the context features into the risk assessment layer of the dynamic decision-making inference engine and calculating the risk probability value of the device node through fuzzy logic inference processing includes: Performing severity assessment on the state mutation points in the context features, and dividing the state mutation points into multiple state mutation levels according to the numerical value of the abnormal working condition event; Performing an impact range assessment process on the abnormal connection edges in the context features, and classifying the abnormal connection edges into a plurality of abnormal connection edge levels according to a degree of reduction in edge weights, wherein the abnormal connection edge levels include a local impact level, a regional impact level, and a global impact level; Construct a fuzzy inference rule base to define the impact of the combination of state mutation level and abnormal connection edge level on the risk probability value; The state mutation level and the abnormal connection edge level are input into the fuzzy inference rule library, and the risk probability value of the device node is generated through fuzzification, rule matching and defuzzification processing.

5. The distribution network fault location method based on intelligent decision-making according to claim 1 is characterized in that: The performing causal link tracing processing on the target state data set based on the fault risk identification result to generate a fault section location result of the distribution network includes: If the risk level corresponding to the risk level identifier in the fault risk identification result is higher than the set risk level, extracting a status data subset overlapping with the risk trigger period in the target status data set; Performing causal relationship mining on the status data subset to construct a fault causal graph comprising device nodes and causal relationship edges, wherein the weights of the causal relationship edges are determined by mutual information values ​​of operating parameters between the device nodes; In the fault causality graph, starting from the significant risk device node, all reachable nodes are traversed by a depth-first search algorithm to generate a candidate set of risk propagation paths; Performing weighted screening on the candidate set of risk propagation paths, and retaining risk propagation paths whose mutual information values ​​exceed a preset threshold as valid risk propagation paths; Identifying, on the effective risk propagation path, a node with the highest degree of abnormality in operating parameters as a key influencing node, wherein the degree of abnormality is determined by the magnitude of deviation of current, voltage, and temperature from a baseline value; Calculate the quantitative value of the degree of influence of the key influencing node on the risk propagation path to form a fault section positioning result including the risk propagation path, the key influencing node and the quantitative value of the degree of influence. The quantitative value of the degree of influence is jointly determined by the position of the key influencing node in the risk propagation path and the degree of abnormality of the operating parameters.

6. The distribution network fault location method based on intelligent decision-making according to claim 5 is characterized in that: The performing causal relationship mining on the status data subset to construct a fault causal graph including device nodes and causal relationship edges includes: Performing Granger causality test on the device operating parameters in the status data subset to determine the direction of causal relationships between device nodes, where the direction of causal relationships indicates whether a state change of one node will cause a state change of another node; Calculating mutual information values ​​of operating parameters of a pair of device nodes having a causal relationship, wherein the mutual information values ​​represent the strength of correlation between state changes between the device nodes; generating directed causal relationship edges between device nodes according to the causal relationship direction and the mutual information value, wherein the direction of the edge is consistent with the causal relationship direction, and the weight of the edge is the mutual information value; Assemble all device nodes and directed causal relationship edges into a graph structure to generate a fault causal graph that reflects the causal relationship between device nodes; The fault causal graph is subjected to sparse processing to remove weak causal relationship edges whose mutual information values ​​are lower than a preset threshold.

7. The distribution network fault location method based on intelligent decision-making according to claim 1 is characterized in that: Generating a fault location instruction including a path identifier, a node identifier, and an impact degree according to the fault section location result includes: Extract the risk propagation path from the fault section location result, and assign a unique path identifier to each risk propagation path according to the path coding rules preset by the distribution network, wherein the path identifier includes a starting node number and an end node number; Extracting key influencing nodes from the fault section location result, and assigning a unique node identifier to each key influencing node according to a node encoding rule preset in the distribution network, wherein the node identifier includes a node geographic location code and a device type code; Extracting the impact degree quantified value from the fault section location result, and converting the quantified value into an impact degree grade identifier according to a preset impact degree grading standard; Slicing the risk propagation path in the time dimension to generate a segment identifier of the risk propagation path within the risk triggering period, wherein the segment identifier includes a start time and an end time of the time period; Performing spatial coordinate mapping processing on the key influencing node to generate a coordinate identifier of the key influencing node in a distribution network geographic information system, wherein the coordinate identifier includes longitude and latitude information; Associating and binding the path identifier, node identifier, impact level identifier, segment identifier, and coordinate identifier to generate an identifier set containing multi-dimensional positioning information; The identification set is input into the instruction generation module, and data encapsulation processing is performed according to the communication protocol format of the distribution network intelligent operation and maintenance platform to generate a fault location instruction including a path identification field, a node identification field and an impact degree field.

8. A distribution network fault location system based on intelligent decision-making, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the distribution network fault location method based on intelligent decision-making as described in any one of claims 1 to 7.

Images