Intelligent inspection and fault positioning method and system for power distribution equipment

Abstract

The invention relates to the technical field of data processing, and discloses an intelligent inspection and fault positioning method and system for power distribution equipment. The method comprises the following steps: performing multi-type sensor deployment on key points around power distribution equipment to obtain an intelligent sensor network; performing multi-mode acquisition and multi-path transmission processing on data acquired by the intelligent sensor network to obtain an original monitoring data set; cleaning, standardizing and fusing the original monitoring data set to obtain preprocessed structured data; performing multi-dimensional spatial correlation analysis on the structured data to obtain health state evaluation data of the power distribution equipment; performing dynamic threshold calculation and intelligent diagnosis on the power distribution equipment to obtain early warning information and a preliminary diagnosis result; and performing expert knowledge base matching and multi-criterion decision analysis on the early warning information and the preliminary diagnosis result to obtain a maintenance decision suggestion scheme. According to the invention, the accuracy of intelligent inspection and fault positioning of the power distribution equipment is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method and system for intelligent inspection and fault location of power distribution equipment. Background Art

[0002] Distribution equipment is a key infrastructure in the power system, and its operating status directly affects the reliability and safety of power supply. Traditional inspection methods for distribution equipment mainly rely on regular manual inspections and simple sensor monitoring. This method has problems such as low inspection efficiency, untimely fault warning, and limited diagnostic accuracy. With the development of the Internet of Things and artificial intelligence technologies, intelligent sensor networks and data analysis technologies have been gradually applied to the monitoring and maintenance of distribution equipment, but they still face many challenges.

[0003] Existing intelligent inspection systems often have the following shortcomings: first, the deployment of sensors lacks systematicity and specificity, making it difficult to fully cover the key points of the equipment; second, the real-time and reliability of data collection and transmission are insufficient, affecting the quality of monitoring data; third, the data processing and analysis methods are relatively simple, making it difficult to effectively process multi-source heterogeneous data, resulting in low accuracy in equipment health status assessment; finally, the fault diagnosis and decision support systems lack flexibility and intelligence, making it difficult to adapt to complex and changeable operating environments and failure modes. Summary of the invention

[0004] The present application provides a method and system for intelligent inspection and fault location of power distribution equipment, which are used to improve the accuracy of intelligent inspection and fault location of power distribution equipment.

[0005] In the first aspect, the present application provides a method for intelligent inspection and fault location of distribution equipment, which includes: deploying multiple types of sensors at key points around the distribution equipment to obtain an intelligent sensor network; performing multi-mode acquisition and multi-path transmission processing on the data collected by the intelligent sensor network to obtain an original monitoring data set; cleaning, standardizing and fusing the original monitoring data set to obtain pre-processed structured data; performing multi-dimensional spatial correlation analysis on the structured data through a multi-scale time series analysis algorithm to obtain health status assessment data of the distribution equipment; based on the health status assessment data, performing dynamic threshold calculation and intelligent diagnosis on the distribution equipment to obtain early warning information and preliminary diagnosis results; performing expert knowledge base matching and multi-criteria decision analysis on the early warning information and preliminary diagnosis results to obtain maintenance decision recommendation plans.

[0006] In a second aspect, the present application provides an intelligent inspection and fault location system for power distribution equipment, the intelligent inspection and fault location system for power distribution equipment comprising: A deployment module is used to deploy multiple types of sensors at key points around the power distribution equipment to obtain an intelligent sensor network; A collection module, used for performing multi-mode collection and multi-path transmission processing on the data collected by the intelligent sensor network to obtain an original monitoring data set; A fusion module is used to clean, standardize and fuse the original monitoring data set to obtain pre-processed structured data; An analysis module, used to perform multi-dimensional spatial correlation analysis on the structured data by using a multi-scale time series analysis algorithm to obtain health status assessment data of the power distribution equipment; A diagnosis module, used to perform dynamic threshold calculation and intelligent diagnosis on the power distribution equipment based on the health status assessment data, and obtain early warning information and preliminary diagnosis results; The decision-making module is used to perform expert knowledge base matching and multi-criteria decision analysis on the warning information and the preliminary diagnosis results to obtain a maintenance decision recommendation plan.

[0007] In the technical solution provided by the present application, multi-type sensors are deployed at key points around the power distribution equipment to build an intelligent sensor network. This comprehensive data collection method provides rich and multi-dimensional raw data for subsequent analysis, greatly improving the comprehensiveness and accuracy of monitoring. Secondly, the data collected by the intelligent sensor network is processed by multi-mode acquisition and multi-path transmission to obtain the original monitoring data set. This method not only improves the reliability of data transmission, but also adapts to the characteristics of different types of sensors to ensure the integrity and real-time nature of the data. The original monitoring data set is cleaned, standardized and fused to obtain pre-processed structured data. This step effectively removes noise and outliers, improves data quality, and lays a solid foundation for subsequent analysis. The multi-dimensional spatial correlation analysis of structured data is performed through a multi-scale time series analysis algorithm to obtain the health status assessment data of the power distribution equipment. This advanced analysis method can capture subtle changes and long-term trends in the equipment status, greatly improving the accuracy and predictability of the health status assessment. Based on the health status assessment data, dynamic threshold calculation and intelligent diagnosis are performed on the power distribution equipment to obtain early warning information and preliminary diagnosis results. This dynamic adjustment method can adapt to changes in the equipment operating environment and load, reduce false alarms and missed alarms, and improve the accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0009] Figure 1 A schematic diagram of an embodiment of a method for intelligent inspection and fault location of power distribution equipment in an embodiment of the present application; Figure 2 It is a schematic diagram of an embodiment of the intelligent inspection and fault location system of the power distribution equipment in the embodiment of the present application. DETAILED DESCRIPTION

[0010] The embodiment of the present application provides a method and system for intelligent inspection and fault location of power distribution equipment. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0011] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the intelligent inspection and fault location method of the power distribution equipment in the embodiment of the present application includes: Step S101: deploy multiple types of sensors at key points around the power distribution equipment to obtain an intelligent sensor network; Step S102, performing multi-mode collection and multi-path transmission processing on the data collected by the intelligent sensor network to obtain an original monitoring data set; Step S103: cleaning, standardizing and fusing the original monitoring data set to obtain pre-processed structured data; Step S104: Perform multi-dimensional spatial correlation analysis on the structured data using a multi-scale time series analysis algorithm to obtain health status assessment data of the power distribution equipment; Step S105: Based on the health status assessment data, dynamic threshold calculation and intelligent diagnosis are performed on the power distribution equipment to obtain early warning information and preliminary diagnosis results; Step S106: Perform expert knowledge base matching and multi-criteria decision analysis on the warning information and the preliminary diagnosis results to obtain a maintenance decision recommendation plan.

[0012] It is understandable that the execution subject of the present application may be an intelligent inspection and fault location system for power distribution equipment, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0013] Specifically, multi-type sensors are deployed at key points around the power distribution equipment to build an intelligent sensor network. Through in-depth analysis of the physical structure and operating characteristics of the power distribution equipment, key monitoring points are determined and sensor types are matched. By calculating the sensor coverage, a coverage heat map is generated, and then the sensor layout is optimized using density clustering analysis. After that, the sensor communication protocol is configured, and the network topology and routing algorithm are designed to achieve efficient data transmission. The data collected by the intelligent sensor network is processed by multi-mode acquisition and multi-path transmission to obtain the original monitoring data set. This includes setting data acquisition trigger rules, compressing and encoding the original data stream, and formulating a multi-path transmission plan based on network topology analysis. By monitoring the transmission quality in real time and dynamically adjusting the transmission path, the reliability and real-time performance of data transmission are ensured. Finally, the aggregated data is timestamped and decoded to form a complete original monitoring data set.

[0014] The original monitoring data set is then cleaned, standardized and fused to obtain preprocessed structured data. This process includes steps such as outlier detection and repair, data type identification and normalization, feature extraction and dimensionality reduction, and multi-source data alignment and fusion. Through these processes, the original data is converted into a standardized and structured form, laying the foundation for subsequent analysis. The multi-dimensional spatial correlation analysis of the structured data is performed using a multi-scale time series analysis algorithm to obtain the health status assessment data of the distribution equipment. First, the data is segmented into time windows and Fourier transformed to obtain frequency domain features. Then wavelet decomposition and principal component analysis are performed to extract the main characteristic components. Feature groups are identified by calculating spatial correlation and cluster analysis. Finally, time series pattern mining and trend analysis are performed to evaluate the health status of the equipment.

[0015] Based on the health status assessment data, dynamic threshold calculation and intelligent diagnosis are performed on the distribution equipment to obtain early warning information and preliminary diagnosis results. This process includes steps such as historical data comparison and analysis, dynamic threshold adjustment, early warning triggering and level classification, fault symptom feature extraction and pattern matching. Through these analyses, equipment abnormalities can be discovered in a timely manner and preliminary fault diagnosis can be performed. Finally, expert knowledge base matching and multi-criteria decision analysis are performed on the early warning information and preliminary diagnosis results to obtain maintenance decision recommendations. It involves semantic parsing, similar case retrieval, case similarity calculation, solution feasibility assessment, multi-criteria decision analysis and fuzzy comprehensive evaluation. Through these analyses, operational maintenance decision recommendations are finally generated.

[0016] For example, a main transformer in a substation uses this intelligent inspection and fault location method. The system deploys multiple types of sensors such as temperature, vibration, and partial discharge around the transformer to build an intelligent sensor network. During the data collection process, the system collects data every 5 minutes and ensures the real-time and reliability of the data through multi-path transmission. After preprocessing, the raw data is structured into multi-dimensional data with multi-dimensional features such as temperature, vibration, and partial discharge. Through multi-scale time series analysis, the system finds that the vibration characteristics of the transformer have shown an upward trend in the near future, and there is a certain correlation with temperature changes. Based on these analysis results, the system calculates the dynamic threshold and triggers an early warning when the vibration intensity exceeds the threshold. By comparing with the cases in the expert knowledge base, the system preliminarily diagnoses that the problem may be caused by the loosening of the fasteners inside the transformer. Finally, the system generates a maintenance decision plan that recommends further inspection and tightening. This process demonstrates the effectiveness of this method in practical applications, which can detect potential faults in a timely manner and provide targeted maintenance recommendations.

[0017] In the embodiment of the present application, multi-type sensors are deployed at key points around the power distribution equipment to build an intelligent sensor network. This comprehensive data collection method provides rich and multi-dimensional raw data for subsequent analysis, greatly improving the comprehensiveness and accuracy of monitoring. Secondly, the data collected by the intelligent sensor network is processed by multi-mode acquisition and multi-path transmission to obtain the original monitoring data set. This method not only improves the reliability of data transmission, but also adapts to the characteristics of different types of sensors to ensure the integrity and real-time nature of the data. The original monitoring data set is cleaned, standardized and fused to obtain pre-processed structured data. This step effectively removes noise and outliers, improves data quality, and lays a solid foundation for subsequent analysis. The structured data is analyzed by multi-dimensional spatial correlation through a multi-scale time series analysis algorithm to obtain the health status assessment data of the power distribution equipment. This advanced analysis method can capture subtle changes and long-term trends in the equipment status, greatly improving the accuracy and predictability of the health status assessment. Based on the health status assessment data, the power distribution equipment is dynamically calculated and intelligently diagnosed to obtain early warning information and preliminary diagnosis results. This dynamic adjustment method can adapt to changes in the equipment operating environment and load, reduce false alarms and missed reports, and improve the accuracy of diagnosis.

[0018] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Analyze the physical structure and operating characteristics of the power distribution equipment to obtain a distribution map of key monitoring points, match the sensor types on the distribution map of key monitoring points, and obtain a preliminary sensor deployment plan; (2) Calculate the sensor coverage range of the preliminary sensor deployment plan to obtain a sensor coverage heat map, and perform density clustering analysis on the sensor coverage heat map to obtain the sensor optimization layout plan; (3) Configure the communication protocol for each sensor in the sensor optimization layout scheme to obtain the sensor network topology, and design the routing algorithm for the sensor network topology to obtain the data transmission path diagram; (4) Bandwidth allocation and load balancing are performed on the data transmission path graph to obtain a network resource allocation strategy, and sensor parameters are configured according to the network resource allocation strategy to obtain a sensor configuration parameter set; (5) Optimize the data acquisition frequency of the sensor configuration parameter set to obtain a dynamic sampling strategy, and perform time synchronization on the sensor network based on the dynamic sampling strategy to obtain the synchronization signal generator configuration; (6) Perform hardware interface design for the synchronization signal generator configuration to obtain the sensor interface specification, and install and debug the sensor according to the sensor interface specification to obtain an intelligent sensor network.

[0019] Specifically, the physical structure and operating characteristics of the distribution equipment are analyzed in depth, including the geometric dimensions, material properties, electrical parameters, and operating environment of the equipment. Through these analyses, the key monitoring points of the equipment are determined, such as the windings, bushings, and oil tanks of the transformer, and the key components such as the contacts and insulators of the switchgear. These key points constitute the key monitoring point distribution map, which lays the foundation for the subsequent sensor deployment. Sensor types are matched according to the characteristics of the key monitoring points. For example, for the transformer oil temperature monitoring point, a temperature sensor is selected; for the switchgear contact resistance monitoring point, a resistance sensor is selected; for the cable partial discharge monitoring point, an ultrasonic sensor is selected. This matching process takes into account the characteristics of the monitoring parameters, the measurement range, the accuracy requirements, and other factors to form a preliminary sensor deployment plan. Based on the preliminary sensor deployment plan, the sensor coverage range is calculated. The effective coverage area of ​​each sensor is calculated using the sensor's sensing range and the geometric shape of the equipment. The calculation results are presented in the form of a heat map, in which the color depth indicates the strength of the sensor coverage. For example, for a rectangular switchgear, its surface can be divided into grids, and the color depth of each grid represents the degree of coverage of the location by the sensor. Density clustering analysis is performed on the sensor coverage heat map to optimize the sensor layout. Density clustering algorithms such as DBSCAN (density-based clustering algorithm) are applied to this process. The algorithm helps determine the best placement of sensors by identifying high-density and low-density areas to avoid coverage overlap or blind spots. Through this, the optimal sensor layout solution is obtained.

[0020] After the optimal layout of sensors is determined, the communication protocol is configured. According to the type, quantity, distribution location and data transmission requirements of sensors, a suitable communication protocol is selected, such as ZigBee, LoRa or NB-IoT. The selection of communication protocol takes into account factors such as transmission distance, power consumption, and bandwidth. After the configuration is completed, the sensor network topology is formed, which describes the connection relationship between sensors and between sensors and data aggregation nodes. Based on the formed network topology, a routing algorithm is designed. Common routing algorithms include the shortest path algorithm and the energy balance routing algorithm. For example, the Dijkstra algorithm is used to calculate the shortest path between any two points in the network, taking into account factors such as the distance between nodes and signal strength. The design result of the routing algorithm is the data transmission path diagram, which guides the flow direction of data in the network.

[0021] Bandwidth allocation and load balancing are performed on the data transmission path graph. The data flow, transmission capacity and overall network load of each path are taken into account. For example, a weighted polling algorithm is used for load balancing, and the distribution ratio of data packets is dynamically adjusted according to the bandwidth and current load of each path. The processing results form a network resource allocation strategy to guide the efficient use of network resources. According to the network resource allocation strategy, the sensor parameters are configured. This includes setting parameters such as the sensor sampling rate, transmission power, and sleep cycle. For example, for sensors on heavily loaded paths, their sampling frequency may be reduced or the data compression ratio may be increased to reduce the network burden. These configurations form a sensor configuration parameter set.

[0022] The data acquisition frequency of the sensor configuration parameter set is optimized to obtain a dynamic sampling strategy. The sampling frequency is dynamically adjusted by taking into account factors such as the operating status of the device and environmental changes. For example, when an abnormal increase in the device temperature is detected, the sampling frequency of the relevant sensor is automatically increased to capture more detailed changes. Based on the dynamic sampling strategy, the sensor network is time synchronized. This ensures that the clocks of all sensors in the network remain consistent, providing a time reference for subsequent data fusion and analysis. Common time synchronization algorithms include the Network Time Protocol (NTP) or the Precision Time Protocol (PTP). The result of the synchronization process is the configuration scheme of the synchronization signal generator.

[0023] Finally, the hardware interface design of the synchronization signal generator configuration is carried out to form the sensor interface specification. This specification defines the physical connection method, signal type, electrical characteristics, etc. between the sensor and other hardware devices. According to this specification, the actual installation and debugging of the sensor is carried out, and finally a complete intelligent sensor network is established.

[0024] For example, a 35kV distribution transformer in a substation adopted this method for intelligent sensor network deployment. First, by analyzing the structure diagram and operating parameters of the transformer, 12 key monitoring points such as oil temperature, winding temperature, partial discharge, and vibration were determined. Then, according to the characteristics of these monitoring points, the corresponding sensor types were selected, such as PT100 temperature sensor, ultrasonic partial discharge sensor, etc. In the sensor coverage calculation, the transformer surface is taken as an example and divided into 100×100 grids. The coverage of each sensor is calculated by the Gaussian distribution model to obtain a 100×100 matrix, and each element value in the matrix represents the coverage strength of the location. The coverage matrices of multiple sensors are superimposed to form the final thermal map.

[0025] In the density clustering analysis, the DBSCAN algorithm was used, and the neighborhood radius ε=5 (grid unit) and the minimum number of points MinPts=4 were set. The algorithm identified 5 high-density areas and 3 low-density areas, and adjusted the location of the sensors accordingly to optimize the layout. ZigBee was selected as the communication protocol, considering its low power consumption and flexible networking. The network topology adopts a tree structure, and 12 sensor nodes are connected to 1 coordinator node through 3 routing nodes. The routing algorithm uses the improved AODV (Ad hoc On-Demand Distance Vector) algorithm, taking into account the node residual energy and link quality factors.

[0026] In bandwidth allocation, a weighted fair queuing algorithm is used. For example, the data of the partial discharge sensor has a higher priority and is allocated 50% of the bandwidth, while the temperature sensor is allocated 30% of the bandwidth, and other sensors share the remaining 20% ​​of the bandwidth. In the dynamic sampling strategy, the temperature sensor samples once every 5 minutes under normal circumstances. When the temperature exceeds the preset threshold (such as 80°C), the sampling frequency is automatically increased to once every 1 minute. The PTP protocol is used for time synchronization to ensure that the time error of all nodes in the network is within 1ms.

[0027] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Perform data collection and processing on the intelligent sensor network to obtain an initial collection plan, and configure rules for the initial collection plan to obtain a collection trigger rule set; (2) Monitor the collection trigger rule set in real time to obtain the original data stream, and perform data compression encoding on the original data stream to obtain a compressed data packet; (3) Perform network topology analysis on the compressed data packets to obtain a multi-path transmission scheme, and then route and distribute the data packets according to the multi-path transmission scheme to obtain a distributed data stream; (4) Monitor the transmission quality of distributed data streams to obtain link status information, and perform adaptive path adjustment based on the link status information to obtain an optimized transmission path; (5) The data on the optimized transmission path are merged in real time to obtain a merged data set, and the merged data set is timestamp aligned and data decoded to obtain the original monitoring data set.

[0028] Specifically, the intelligent sensor network is processed for data collection to obtain an initial collection plan. First, it is necessary to formulate corresponding data collection strategies according to the characteristics and monitoring requirements of different sensors. For example, for temperature sensors, parameters such as measurement range, accuracy and response time are considered; for partial discharge sensors, sampling frequency and trigger conditions need to be considered. The initial collection plan includes basic parameters such as sampling frequency, sampling accuracy, and sampling duration. The initial collection plan is configured with rules to obtain a collection trigger rule set. During the rule configuration process, different collection trigger conditions need to be set in combination with the equipment operation characteristics and monitoring objectives. For example, the sampling frequency for different time periods can be set according to the law of equipment load changes, or a dynamic trigger threshold can be set according to ambient temperature changes. Through such configuration, a complete collection trigger rule set containing multiple trigger rules is finally formed.

[0029] The collection trigger rule set is monitored in real time to obtain the original data stream. The collection trigger rule set includes multiple rules such as timed trigger, threshold trigger and event trigger, which are pre-set according to the operating characteristics and monitoring requirements of the distribution equipment. For example, for the transformer oil temperature, a timed trigger rule may be set to collect data every 5 minutes, and a threshold rule may be set to trigger the collection immediately when the temperature exceeds 80°C. During the real-time monitoring process, these rules are constantly checked to see if they meet the trigger conditions. Once they are met, the data collection of the corresponding sensor is immediately started to form the original data stream. The original data stream is compressed and encoded to obtain a compressed data packet. It aims to reduce the amount of data transmission and improve network efficiency. Commonly used compression algorithms include lossless compression algorithms (such as Huffman coding and LZW algorithm) and lossy compression algorithms (such as wavelet transform compression). The choice of compression algorithm depends on the characteristics and accuracy requirements of the data. For example, for slowly changing analog quantities such as transformer oil temperature, differential coding plus Huffman compression can be used; while for discrete quantities such as switch status, run-length coding compression can be used directly.

[0030] After the compressed data packet is generated, the network topology analysis is performed on it to obtain a multi-path transmission scheme. The network topology analysis takes into account the structural characteristics of the sensor network, such as node distribution, link connection status, etc. The purpose of the multi-path transmission scheme is to improve the reliability and efficiency of data transmission. Commonly used multi-path routing algorithms include disjoint multipath routing and braided multipath routing. These algorithms provide multiple transmission options for data packets by calculating multiple non-interfering or partially overlapping paths. According to the multi-path transmission scheme, the data packets are routed and distributed to obtain a distributed data stream. In the process of routing distribution, factors such as network load balancing and energy consumption balance are considered. For example, the weighted round robin algorithm can be used to dynamically adjust the distribution ratio of data packets on different paths according to parameters such as the bandwidth and current load of each path. This can avoid the situation where some paths are overloaded while other paths are idle, and improve the overall network utilization.

[0031] Monitor the transmission quality of distributed data streams to obtain link status information. Transmission quality monitoring includes multiple indicators such as delay, packet loss rate, and signal strength. These indicators can be calculated at the receiving end by embedding information such as timestamps and sequence numbers in data packets. For example, by comparing the sending timestamp and receiving timestamp, the end-to-end delay can be calculated; by checking the continuity of the sequence number, the packet loss rate can be estimated. Adaptive path adjustment is performed based on link status information to obtain the optimized transmission path. Use dynamic routing algorithms, such as an improved version of the Ad hoc On-Demand Distance Vector (AODV) algorithm. The algorithm dynamically adjusts the routing table based on the link status information monitored in real time and selects the current optimal transmission path. For example, when it is detected that the packet loss rate of a certain path suddenly increases, the algorithm will quickly transfer the traffic to other paths with better quality.

[0032] The data on the optimized transmission path is aggregated in real time to obtain an aggregated data set. Data aggregation is an effective way to reduce network transmission load. Common data aggregation technologies include data aggregation and data fusion. For example, for the data of multiple temperature sensors, the average value can be calculated at the intermediate node, and only the average result is transmitted instead of all the original data. Finally, the aggregated data set is timestamped and decoded to obtain the original monitoring data set. Timestamp alignment solves the problem of data timing confusion caused by network transmission delay. Data decoding is the inverse operation of the compression process, which restores the compressed data to its original format.

[0033] For example, a group of switch cabinets in a 110kV substation adopted this data collection and transmission method. Ten sensors such as temperature, humidity, and partial discharge were deployed on the switch cabinet. The collection trigger rule was set as follows: temperature and humidity were collected every 10 minutes, and the collection was triggered immediately when the partial discharge signal strength exceeded the preset threshold. During one collection process, the temperature data was compressed using differential encoding plus Huffman compression to compress the original 16-bit floating point number to an average of 6 bits per data point, with a compression rate of 62.5%. After network topology analysis, three transmission paths were determined: the main path A was transmitted through the ZigBee network, and the backup paths B and C were transmitted through the 4G network. The route distribution adopted a weighted polling algorithm. Under normal circumstances, the main path A bears 70% of the traffic, and B and C each bear 15%. During a data transmission process, the average delay of the main path A increased from the normal 50ms to 150ms through transmission quality monitoring. The adaptive routing algorithm immediately adjusted the allocation strategy, reducing the traffic share of path A to 30%, while increasing the shares of paths B and C to 35% each. This adjustment keeps the overall average transmission delay below 80ms, ensuring the real-time nature of the data.

[0034] During the data aggregation process, the 3,600 raw data records generated by 10 sensors in 1 hour (one record every 6 minutes for each sensor) were reduced to 600 by calculating the 10-minute average. Finally, all data were timestamped in the control center to correct the maximum 2-second time deviation caused by network delay, and the original monitoring data set that accurately reflects the real-time status of the equipment was obtained.

[0035] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Perform preliminary detection and processing on the original monitoring data set to obtain a data detection report, and perform abnormal repair on the data detection report to obtain a repaired data set; (2) Identify the data type of the repaired data set to obtain a data type mapping table, and normalize the data according to the data type mapping table to obtain a standardized data set; (3) Extract features from the standardized data set to obtain a feature vector set, and perform dimensionality reduction on the feature vector set to obtain a feature matrix after dimensionality reduction; (4) Performing data alignment on the feature matrix after dimensionality reduction to obtain time-aligned multi-source data, and performing data fusion on the time-aligned multi-source data to obtain a fused feature tensor; (5) The data structure of the fused feature tensor is reorganized to obtain a structured data template, and the fused data is formatted according to the structured data template to obtain preprocessed structured data.

[0036] Specifically, the original monitoring data set is preliminarily tested and processed to obtain a data test report. The preliminarily testing and processing mainly includes three aspects: data integrity check, data consistency check and data validity check. The data integrity check mainly verifies whether there are missing data, such as whether the time series is continuous and whether the sensor data is complete; the data consistency check mainly verifies whether the data between different sensors conform to the physical relationship, such as whether the load current of the transformer matches the temperature change trend; the data validity check mainly verifies whether the data value is within a reasonable range, such as whether the temperature value is within the normal working temperature range of the equipment. The test results form a data test report, which contains information such as the type, location and severity of the abnormal data. The data test report is repaired to obtain the repaired data set. Different repair strategies are adopted according to the type of abnormality detected: for missing data, linear interpolation, polynomial interpolation or completion methods based on similar daily patterns can be used; for abnormal values, they can be corrected by moving average, median filtering or methods based on statistical features; for inconsistent data, it is necessary to combine physical models and expert experience for judgment and correction. For example, for a sudden change in the value of a temperature sensor during continuous monitoring, we can combine historical data and data from adjacent moments and use the sliding window median filtering method to repair it, ensuring that the repaired data not only maintains the original change trend but also eliminates unreasonable mutations.

[0037] The data type of the repaired data set is identified to obtain a data type mapping table. This involves classifying the data collected by different sensors, such as identifying temperature data as continuous values ​​and switch states as discrete Boolean values. Data type identification uses a combination of heuristic rules and statistical analysis. For example, the data type is determined by checking the value range, distribution characteristics, and change frequency of the data. The generated data type mapping table records the name, type, unit, and other information of each data field. The data is normalized according to the data type mapping table to obtain a standardized data set. The purpose of normalization is to eliminate the dimensional differences between different types of data so that all data fall within the same numerical range, usually [0, 1] or [-1, 1]. For continuous values, commonly used normalization methods include minimum-maximum normalization and Z-score normalization.

[0038] Feature extraction is performed on the standardized data set to obtain a feature vector set. The purpose of feature extraction is to extract features that can represent the essential characteristics of the data from the original data. For the monitoring data of distribution equipment, common feature extraction methods include time domain feature extraction (such as mean, variance, peak, etc.) and frequency domain feature extraction (such as Fourier transform, wavelet transform, etc.). For example, for transformer oil temperature data, features such as daily average temperature, temperature change rate, and temperature fluctuation amplitude can be extracted; for partial discharge signals, features such as discharge amplitude, discharge frequency, and discharge phase can be extracted. Dimensionality reduction is performed on the feature vector set to obtain a feature matrix after dimensionality reduction. The purpose of dimensionality reduction is to reduce data redundancy and improve the efficiency of subsequent analysis. Common dimensionality reduction methods include principal component analysis (PCA) and linear discriminant analysis (LDA). Taking PCA as an example, its basic idea is to project the original high-dimensional feature space into a new low-dimensional feature space so that the variance of the projected data is maximized. The core steps of PCA include calculating the covariance matrix, solving eigenvalues ​​and eigenvectors, and selecting principal components.

[0039] Data alignment is performed on the feature matrix after dimensionality reduction to obtain time-aligned multi-source data. The purpose of data alignment is to solve problems such as inconsistent data acquisition frequencies of different sensors and asynchronous timestamps. Common alignment methods include linear interpolation and spline interpolation. For example, for temperature data with a low sampling frequency, it can be extended to have the same time resolution as the high-frequency sampled current data through interpolation methods. Data fusion is performed on the time-aligned multi-source data to obtain a fused feature tensor. The purpose of data fusion is to combine information from different sensors to form a comprehensive description of the device status. Common data fusion methods include Kalman filtering, Bayesian fusion, etc. For example, a Kalman filter can be used to fuse multi-source data such as temperature, vibration, and partial discharge to obtain a more accurate device status estimate.

[0040] The fused feature tensor is restructured to obtain a structured data template. The purpose of data restructuring is to organize the fused data into a format that is convenient for subsequent analysis and storage. The structured data template defines how the data is organized, including field names, data types, and relationships between data. Finally, the fused data is formatted according to the structured data template to obtain preprocessed structured data. Formatting ensures that all data conforms to the predefined structure for subsequent analysis and processing.

[0041] For example, a main transformer in a 110kV substation uses this data preprocessing method. The transformer is equipped with a variety of sensors such as oil temperature, winding temperature, load current, and partial discharge. In the data type identification stage, the oil temperature and winding temperature are identified as continuous values ​​with a range of -20°C to 150°C; the load current is identified as a continuous value with a range of 0A to 1000A; and the partial discharge signal is identified as discrete event data.

[0042] In the normalization process, the minimum-maximum normalization method is used for temperature data. For example, the oil temperature measured at a certain moment is 75°C, and the normalized value calculated by the formula is (75 - (-20)) / (150 - (-20)) ≈ 0.5588.

[0043] In the feature extraction stage, the daily average temperature, temperature change rate and other features were extracted from the temperature data; the mean value, peak factor and other features were extracted from the current data; the discharge amplitude, discharge frequency and other features were extracted from the partial discharge data. Finally, a feature vector set containing 20 features was obtained. In the dimensionality reduction process, the PCA method was used to reduce the 20-dimensional features to 8 dimensions, retaining 95% of the data variance. In the data alignment process, all data were unified to a frequency of one sampling point per minute. In the data fusion stage, the Kalman filter was used to fuse the temperature, current and partial discharge data to obtain a more accurate transformer state estimate.

[0044] Finally, the preprocessed structured data includes fields such as timestamp, fused temperature estimation, load estimation, and partial discharge intensity estimation, providing a high-quality data foundation for subsequent equipment status assessment and fault diagnosis. Through this set of preprocessing methods, the original heterogeneous and misaligned monitoring data is converted into a standardized and structured data set, which greatly improves the efficiency and accuracy of subsequent analysis.

[0045] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Perform data analysis on the structured data to obtain an analysis feature set, and perform frequency domain conversion on the analysis feature set to obtain a frequency domain feature matrix; (2) Perform wavelet decomposition on the frequency domain feature matrix to obtain a time-frequency feature map, and perform principal component analysis on the time-frequency feature map to obtain the main feature components; (3) Calculate the spatial correlation of the main characteristic components to obtain the correlation matrix, and perform cluster analysis on the correlation matrix to obtain the characteristic groups; (4) Mining the time series patterns of feature groups to obtain the equipment operation status sequence, and performing trend analysis on the equipment operation status sequence to obtain the health status change trend; (5) Perform a health score on the equipment based on the health status change trend to obtain health status assessment data.

[0046] Specifically, data analysis and processing are performed on structured data to obtain an analysis feature set. First, feature extraction is performed on structured data from a time domain perspective, including statistical features and morphological features. Statistical features include statistical quantities such as mean, standard deviation, kurtosis, and skewness, which are used to describe the overall distribution characteristics of the data; morphological features include waveform factor, pulse factor, margin factor, etc., which are used to describe the fluctuation characteristics of the data. At the same time, it is also necessary to extract the trend characteristics of the data, such as using polynomial fitting to obtain the change trend of the data and extract key feature points such as inflection points and extreme points. These extracted features together constitute the analysis feature set, which provides a basis for subsequent frequency domain analysis. The analysis feature set is then transformed into the frequency domain to obtain a frequency domain feature matrix. Frequency domain transformation mainly uses methods such as fast Fourier transform (FFT), short-time Fourier transform (STFT) or wavelet transform to convert time domain signals into frequency domain space. For example, for the vibration signal of the power distribution equipment, the amplitude and phase information of each frequency component can be obtained through FFT to form the spectrum characteristics; the time-frequency distribution characteristics of the signal can be obtained through STFT to reflect the frequency characteristics at different times; and the frequency characteristics at multiple scales can be obtained through wavelet transform. These frequency domain features are organized into a matrix form, where the rows of the matrix represent different feature dimensions and the columns represent different frequency components or time scales, thus forming a complete frequency domain feature matrix.

[0047] The frequency domain feature matrix is ​​decomposed by wavelet to obtain the time-frequency feature map. Wavelet decomposition is a multi-resolution analysis method that can perform localized analysis on the signal in both the time domain and the frequency domain. For the monitoring data of power distribution equipment, the commonly used wavelet basis functions include Daubechies wavelet, Morlet wavelet, etc. In the wavelet decomposition process, the original signal is decomposed into wavelet coefficients of different scales. These coefficients constitute the time-frequency feature map, which reflects the energy distribution of the signal at different time and frequency scales. The time-frequency feature map is subjected to principal component analysis to obtain the main characteristic components. Principal component analysis (PCA) is a dimensionality reduction technique that aims to find the main direction of change in the data. In the PCA process, the covariance matrix of the time-frequency feature map is first calculated, and then the eigenvalues ​​and eigenvectors of the matrix are solved. The eigenvectors are sorted according to the size of the corresponding eigenvalues, and the first few eigenvectors constitute the main characteristic components. These main characteristic components retain most of the information in the original data while greatly reducing the dimension of the data.

[0048] The spatial correlation of the main feature components is calculated to obtain the correlation matrix. The spatial correlation calculation aims to reveal the correlation between different monitoring points or different types of data. Common correlation measurement methods include Pearson correlation coefficient, Spearman rank correlation coefficient, etc. In the calculated correlation matrix, each element represents the degree of correlation between two features. Cluster analysis is performed on the correlation matrix to obtain feature groups. The purpose of cluster analysis is to group similar features together to facilitate subsequent pattern recognition and status assessment. Common clustering algorithms include K-means, hierarchical clustering, etc. In the clustering process, it is necessary to define the distance measure between features based on the correlation matrix, and then divide the features into several groups based on this distance. Each feature group represents an aspect of the equipment status or a class of related parameters. Time series pattern mining is performed on the feature groups to obtain the equipment operation status sequence. The purpose of time series pattern mining is to discover recurring patterns or laws from time series data. Common time series pattern mining methods include dynamic time warping (DTW), hidden Markov model (HMM), etc. Through these methods, the typical operating status of the equipment at different times can be identified, such as normal operation, mild abnormality, severe failure, etc. Trend analysis is performed on the equipment operating status sequence to obtain the trend of health status changes. Trend analysis aims to predict the future health status changes of the equipment. Common trend analysis methods include linear regression and exponential smoothing. By analyzing the change pattern of the equipment status sequence, it is possible to predict whether the equipment has a trend of gradual deterioration or whether there are periodic changes.

[0049] Finally, the health score of the equipment is calculated based on the trend of health status changes to obtain health status assessment data. Health scoring is a process that comprehensively considers multiple factors, usually using weighted scoring or fuzzy comprehensive evaluation methods. The scoring result can be a continuous value or a discrete level, such as "excellent", "good", "average", "poor", etc.

[0050] For example, an oil-immersed transformer in a 110kV substation uses this multi-scale time series analysis method for health status assessment. The transformer is equipped with a variety of sensors such as oil temperature, winding temperature, load current, and dissolved gas in oil. First, the data collected within a week were decomposed by wavelet, and the Daubechies 4th-order wavelet was used to obtain wavelet coefficients of 5 scales. These coefficients constitute the time-frequency characteristic diagram, which clearly shows the changing characteristics of transformer parameters at different time scales. PCA analysis was performed on the time-frequency characteristic diagram, and the first 5 principal components were selected, which cumulatively explained 95% of the variance of the data. These 5 principal components correspond to the main features such as load change, temperature change, and gas content change. Then the correlation matrix between these principal components was calculated, and it was found that the correlation coefficient between load change and temperature change was 0.85, indicating that the two are highly correlated.

[0051] The K-means algorithm was used to cluster the correlation matrix and divide the features into three groups: temperature-load group, gas group, and vibration group. Time series pattern mining was performed on each group of features, and the DTW algorithm was used to identify four typical operating states: normal operation, mild overload, insulation degradation, and severe fault. Through trend analysis of the operating state sequence for one month, it was found that the frequency of insulation degradation of the transformer had a gradual increasing trend. Finally, the transformer was scored for health based on the weight of each indicator (temperature 30%, load 20%, gas 40%, vibration 10%). The score was 78 points (out of 100 points), corresponding to the "good" level, but with a slight deterioration trend.

[0052] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Perform parameter analysis on the health status assessment data to obtain a parameter feature set, and perform threshold calculation on the parameter feature set to obtain a dynamic threshold sequence; (2) Adaptively adjust the dynamic threshold sequence to obtain the threshold adjustment rule, and perform threshold judgment on the real-time health status data according to the threshold adjustment rule to obtain the early warning trigger signal; (3) Perform frequency analysis on the warning trigger signal to obtain the warning level classification standard, and classify the warning information according to the warning level classification standard to obtain graded warning information; (4) Extract symptom features from the graded warning information to obtain a fault symptom feature set, and perform pattern matching on the fault symptom feature set to obtain candidate fault types; (5) Perform probability evaluation on candidate fault types to obtain fault probability distribution, and generate preliminary diagnosis results based on the fault probability distribution.

[0053] Specifically, the health status assessment data is subjected to parameter analysis and processing to obtain a parameter feature set. Parameter analysis and processing mainly include three aspects: static feature analysis, dynamic feature analysis and correlation feature analysis. Static feature analysis mainly examines the absolute value level of various indicators of the equipment, such as the numerical value of parameters such as temperature, vibration, and noise; dynamic feature analysis focuses on the change characteristics of parameters, including the change rate, change acceleration, and change periodicity; correlation feature analysis studies the mutual influence relationship between different parameters, such as the correspondence between temperature and load, and the correlation between vibration and noise. Through these analyses, key parameter features that can fully reflect the health status of the equipment are extracted to form a parameter feature set. Threshold calculation is performed on the parameter feature set to obtain a dynamic threshold sequence. The threshold calculation process takes into account the dynamic characteristics of equipment operation and uses an adaptive method to determine the alarm threshold. For example, for transformer oil temperature, a dynamic threshold model can be established using multiple regression or machine learning methods based on factors such as load level and ambient temperature; for partial discharge signals of switchgear, dynamic alarm thresholds can be calculated using probability statistics methods in combination with factors such as equipment operation time and ambient humidity. These thresholds calculated for different parameters are arranged in time series to form a complete dynamic threshold sequence.

[0054] Adaptively adjust the dynamic threshold sequence to obtain the threshold adjustment rules. In the analysis of historical data and real-time data, factors such as the equipment operating environment and load changes are taken into account. Adaptive adjustment uses algorithms such as exponentially weighted moving average (EWMA) to enable the threshold to change dynamically over time and operating conditions. The adjusted threshold is more in line with the actual operating status of the equipment, improving the accuracy of the early warning. According to the threshold adjustment rules, the threshold of the real-time health status data is judged to obtain the early warning trigger signal. The real-time monitoring data is compared with the dynamically adjusted threshold. When a certain indicator exceeds the corresponding threshold, the early warning signal is triggered. The early warning trigger signal contains information such as the trigger time, the trigger reason, and the degree of overlimit.

[0055] The frequency analysis of the warning trigger signal is carried out to obtain the warning level classification standard. The frequency analysis takes into account the frequency, duration and severity of the warning signal. Through the statistical analysis of the historical warning data, the correlation model between the warning frequency and the severity of the equipment failure is established. This model helps to determine the classification thresholds of different warning levels. The warning information is classified according to the warning level classification standard to obtain the graded warning information. The warning level is usually divided into multiple levels, such as minor, medium, severe and emergency. Each level corresponds to different processing strategies and response time requirements. The classification process takes into account multiple factors such as the frequency, duration and importance of the equipment involved of the warning signal.

[0056] Symptom feature extraction is performed on the graded warning information to obtain a fault symptom feature set. Symptom feature extraction is to extract key features that can reflect the abnormal state of the equipment from the warning information. These features may include the rate of abnormal temperature rise, sudden changes in vibration amplitude, frequency distribution of partial discharge signals, etc. The feature extraction process utilizes multiple methods such as time domain analysis, frequency domain analysis, and time-frequency analysis. Pattern matching is performed on the fault symptom feature set to obtain candidate fault types. The pattern matching process is to compare the extracted symptom features with the pre-established fault mode library. The fault mode library contains typical feature patterns of various known faults. The matching process uses a fuzzy matching algorithm that can handle the uncertainty and ambiguity of the features. The matching result is a set of possible fault types, each of which has a matching score.

[0057] The probability of candidate fault types is evaluated to obtain the fault probability distribution. The probability evaluation takes into account multiple factors, including the results of pattern matching, the historical fault records of the equipment, the current operating environment, etc. The evaluation process uses a Bayesian network model, which can comprehensively consider multiple sources of evidence to obtain the probability of each candidate fault type. Finally, a preliminary diagnosis result is generated based on the fault probability distribution. The diagnosis result includes the most likely fault type, fault probability, possible fault cause, and recommended treatment measures.

[0058] For example, an oil-immersed transformer in a 110kV substation uses this intelligent inspection and fault location method. The transformer is equipped with multiple sensors such as oil temperature, winding temperature, partial discharge, and dissolved gas in oil. During one monitoring, the system detected that the transformer oil temperature had an abnormally high trend. First, the dynamic threshold adjustment algorithm dynamically adjusted the oil temperature alarm threshold from the usual 85°C to 90°C, taking into account that it is currently a high-load period in summer. When the real-time oil temperature reached 88°C, the early warning signal was triggered. The system conducted a frequency analysis of recent early warning signals and found that in the past 24 hours, the oil temperature exceeded 85°C 5 times, each time lasting more than 30 minutes. According to the pre-set grading criteria, this situation is classified as a "medium" level early warning.

[0059] During the symptom feature extraction process, the system extracted the following features: the oil temperature rise rate was 2°C / hour and lasted for more than 6 hours; at the same time, the dissolved gas analysis showed that the acetylene content increased from the normal 5ppm to 15ppm. These features have a high degree of match with the "winding overheating" and "local overheating" modes in the fault mode library. Through the Bayesian network model for probability evaluation, considering the service life of the transformer, historical fault records and current load conditions, the system gave the following probability distribution: winding overheating 65%, local overheating 30%, and other faults 5%.

[0060] Based on this probability distribution, the system generated a preliminary diagnosis result: the most likely fault type is winding overheating, and the recommended measures include reducing the load, increasing cooling, and conducting further winding insulation tests. This diagnosis result provides maintenance personnel with clear fault location and handling suggestions, greatly improving the efficiency and accuracy of fault handling.

[0061] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Perform feature analysis on the warning information and preliminary diagnosis results to obtain a feature mapping set, and perform case matching on the feature mapping set to obtain a similar case set; (2) Calculate the case similarity of the similar case set to obtain the case matching ranking, and select the top N most similar cases according to the case matching ranking to obtain the candidate solution set, where N is a positive integer greater than 1; (3) Perform feasibility assessment on the candidate solution set to obtain solution evaluation indicators, and construct a multi-criteria decision matrix based on the solution evaluation indicators to obtain a decision weight vector; (4) Perform fuzzy comprehensive evaluation on the decision weight vector to obtain the comprehensive score of the scheme, and sort the candidate schemes according to the comprehensive score to obtain the optimal maintenance scheme; (5) Conduct a cost-benefit analysis on the optimal maintenance plan, obtain implementation suggestions for the plan, and generate a maintenance decision-making recommendation plan based on the implementation suggestions.

[0062] Specifically, feature analysis is performed on the warning information and preliminary diagnosis results to obtain a feature mapping set. Feature analysis is mainly performed from three dimensions: warning feature dimension, fault feature dimension, and time series feature dimension. The warning feature dimension analysis includes features such as warning level, warning frequency, and warning duration; the fault feature dimension analysis includes features such as fault type, fault symptom, and fault development trend; the time series feature dimension focuses on the temporal correlation between warnings and faults, such as the evolution process from warning to fault, the temporal relationship between different warnings, etc. Through multi-dimensional feature analysis, the key features in the warning information and diagnosis results are extracted, and the mapping relationship between features and fault modes is established to form a feature mapping set. Case matching is performed on the feature mapping set to obtain a similar case set. The case matching process adopts a multi-level matching strategy. First, a rough matching is performed based on the fault type to filter out historical cases of the same type; then, a precise matching is performed based on specific features, such as comparing the similarity of warning modes and the matching degree of fault symptoms; finally, the matching degree of background information such as equipment type and operating environment is also considered. Through this multi-level matching process, the cases most similar to the current situation are selected from the historical case library to form a similar case set.

[0063] The case similarity is calculated for the similar case set to obtain the case matching ranking. The similarity calculation uses the weighted Euclidean distance method, taking into account multiple features such as fault type, equipment type, and operating environment. The calculation formula is: ; in, Show case and The similarity between represents the weight of the kth feature, and Respectively represent cases and The value on the kth feature, where m is the total number of features. The smaller the calculated similarity, the more similar the cases are. Select the top N most similar cases based on the case matching degree to obtain the candidate solution set. The choice of N value depends on the size of the case library and the similarity distribution. Usually 3 to 5 most similar cases are selected. These cases constitute the candidate solution set, which provides a basis for subsequent solution evaluation.

[0064] The feasibility of the candidate solution set is evaluated to obtain the solution evaluation index. The feasibility evaluation takes into account multiple aspects such as technical feasibility, economic feasibility, and time feasibility. The evaluation adopts the fuzzy comprehensive evaluation method to convert qualitative indicators into quantitative indicators. The evaluation indicators include implementation difficulty, required resources, expected effects, etc. A multi-criteria decision matrix is ​​constructed based on the solution evaluation index to obtain the decision weight vector. The elements of the multi-criteria decision matrix A are It represents the score of the i-th solution on the j-th evaluation index. The decision weight vector W is obtained by the analytic hierarchy process (AHP), which reflects the relative importance of each evaluation index. The calculation process of AHP includes the steps of constructing the judgment matrix, calculating the eigenvector and consistency test.

[0065] Perform fuzzy comprehensive evaluation on the decision weight vector to obtain the comprehensive score of the scheme. The calculation formula of fuzzy comprehensive evaluation is: ; Among them, B represents the comprehensive evaluation result vector, W is the weight vector, and R is the fuzzy relationship matrix. Represents fuzzy synthesis operation. The candidate solutions are sorted according to their comprehensive scores to obtain the optimal maintenance solution. The sorting process takes into account the numerical value and distribution characteristics of the comprehensive score, and selects the solution with the highest score that is significantly better than other solutions as the preferred maintenance solution.

[0066] Finally, a cost-benefit analysis is conducted on the optimal maintenance plan to obtain the implementation recommendations. The cost-benefit analysis uses the net present value method (NPV), and the calculation formula is: ; in, represents the income in year t, represents the cost in year t, r is the discount rate, and T is the analysis period. The scheme with NPV greater than 0 is considered to be economically feasible. Based on the results of cost-benefit analysis, a maintenance decision recommendation is generated. The recommendation includes specific maintenance measures, implementation schedule, required resources, and expected results.

[0067] For example, an oil-immersed transformer in a 110kV substation experienced abnormal vibration and local overheating. The system first retrieved five similar cases from the case library, and the matching degrees were calculated to be 0.92, 0.87, 0.85, 0.83, and 0.79, respectively. The first three most similar cases were selected as the candidate solution set. The feasibility of these three candidate solutions was evaluated, considering four indicators: technical difficulty, required time, resource consumption, and expected effect. The evaluation results formed a 3×4 decision matrix. The weight vector of these four indicators calculated by the AHP method was [0.3, 0.2, 0.25, 0.25].

[0068] After fuzzy comprehensive evaluation, the comprehensive scores of the three schemes were 0.85, 0.78 and 0.72 respectively. According to the score ranking, the highest score of 0.85 was selected as the preferred maintenance scheme. This scheme recommends partial discharge detection and thermal image analysis of the transformer, and decides whether local repairs or replacement of parts are needed based on the results. Cost-benefit analysis shows that the initial cost of implementing this scheme is 500,000 yuan, which is expected to extend the service life of the transformer by 5 years and save 200,000 yuan in operation and maintenance costs each year. Assuming a discount rate of 5%, the calculated NPV is 346,000 yuan, which is greater than 0, proving that the scheme is economically feasible.

[0069] The maintenance decision proposal generated in the end includes: 1) conducting detailed partial discharge detection and thermal image analysis; 2) based on the detection results, it may be necessary to replace damaged insulation or cooling system components; 3) it is recommended to be implemented during the next planned power outage, which is expected to take 3 days; 4) the required professionals include 1 high-voltage test engineer and 1 thermal imager operator. This case shows how to extract useful information from a large number of historical cases through a systematic approach and combine multiple decision-making methods to generate scientific and reasonable maintenance decision proposals.

[0070] The above describes the intelligent inspection and fault location method of the power distribution equipment in the embodiment of the present application. The following describes the intelligent inspection and fault location system of the power distribution equipment in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the intelligent inspection and fault location system of the power distribution equipment includes: A deployment module 201 is used to deploy multiple types of sensors at key points around the power distribution equipment to obtain an intelligent sensor network; The acquisition module 202 is used to perform multi-mode acquisition and multi-path transmission processing on the data collected by the smart sensor network to obtain an original monitoring data set; The fusion module 203 is used to clean, standardize and fuse the original monitoring data set to obtain pre-processed structured data; The analysis module 204 is used to perform multi-dimensional spatial correlation analysis on the structured data by using a multi-scale time series analysis algorithm to obtain health status assessment data of the power distribution equipment; The diagnosis module 205 is used to perform dynamic threshold calculation and intelligent diagnosis on the power distribution equipment based on the health status assessment data to obtain early warning information and preliminary diagnosis results; The decision module 206 is used to perform expert knowledge base matching and multi-criteria decision analysis on the warning information and the preliminary diagnosis results to obtain a maintenance decision recommendation plan.

[0071] Through the collaboration of the above components, multi-type sensors are deployed at key points around the distribution equipment to build an intelligent sensor network. This comprehensive data collection method provides rich and multi-dimensional raw data for subsequent analysis, greatly improving the comprehensiveness and accuracy of monitoring. Secondly, the data collected by the intelligent sensor network is processed by multi-mode acquisition and multi-path transmission to obtain the original monitoring data set. This method not only improves the reliability of data transmission, but also adapts to the characteristics of different types of sensors to ensure the integrity and real-time nature of the data. The original monitoring data set is cleaned, standardized and fused to obtain pre-processed structured data. This step effectively removes noise and outliers, improves data quality, and lays a solid foundation for subsequent analysis. The multi-dimensional spatial correlation analysis of structured data is performed through a multi-scale time series analysis algorithm to obtain the health status assessment data of the distribution equipment. This advanced analysis method can capture subtle changes and long-term trends in the equipment status, greatly improving the accuracy and predictability of the health status assessment. Based on the health status assessment data, dynamic threshold calculation and intelligent diagnosis are performed on the distribution equipment to obtain early warning information and preliminary diagnosis results. This dynamic adjustment method can adapt to changes in the equipment operating environment and load, reduce false alarms and missed alarms, and improve the accuracy of diagnosis.

[0072] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent inspection and fault location method for power distribution equipment, characterized in that: The intelligent inspection and fault location method of the power distribution equipment includes: Deploy multiple types of sensors at key points around the power distribution equipment to obtain an intelligent sensor network; Performing multi-mode collection and multi-path transmission processing on the data collected by the intelligent sensor network to obtain an original monitoring data set; Cleaning, standardizing and fusing the original monitoring data set to obtain preprocessed structured data; Performing multi-dimensional spatial correlation analysis on the structured data by using a multi-scale time series analysis algorithm to obtain health status assessment data of the power distribution equipment; Based on the health status assessment data, dynamic threshold calculation and intelligent diagnosis are performed on the power distribution equipment to obtain early warning information and preliminary diagnosis results; Expert knowledge base matching and multi-criteria decision analysis are performed on the warning information and preliminary diagnosis results to obtain a maintenance decision recommendation plan.

2. The intelligent inspection and fault location method for power distribution equipment according to claim 1, characterized in that: The multi-type sensors are deployed at key points around the power distribution equipment to obtain an intelligent sensor network, including: Analyze the physical structure and operating characteristics of the power distribution equipment to obtain a distribution map of key monitoring points, and match the sensor types on the distribution map of key monitoring points to obtain a preliminary sensor deployment plan; Calculating the sensor coverage range of the preliminary sensor deployment plan to obtain a sensor coverage heat map, and performing density clustering analysis on the sensor coverage heat map to obtain a sensor optimization layout plan; Performing communication protocol configuration on each sensor in the sensor optimization layout scheme to obtain a sensor network topology structure, and performing routing algorithm design on the sensor network topology structure to obtain a data transmission path diagram; Performing bandwidth allocation and load balancing processing on the data transmission path graph to obtain a network resource allocation strategy, and performing parameter configuration on the sensor according to the network resource allocation strategy to obtain a sensor configuration parameter set; Optimizing the data acquisition frequency of the sensor configuration parameter set to obtain a dynamic sampling strategy, and performing time synchronization processing on the sensor network according to the dynamic sampling strategy to obtain a synchronization signal generator configuration; A hardware interface design is performed on the synchronization signal generator configuration to obtain a sensor interface specification, and sensors are installed and debugged according to the sensor interface specification to obtain an intelligent sensor network.

3. The intelligent inspection and fault location method for power distribution equipment according to claim 1, characterized in that: The data collected by the intelligent sensor network is subjected to multi-mode collection and multi-path transmission processing to obtain an original monitoring data set, including: Performing data collection processing on the intelligent sensor network to obtain an initial collection plan, and performing rule configuration on the initial collection plan to obtain a collection trigger rule set; The collection trigger rule set is monitored in real time to obtain an original data stream, and the original data stream is compressed and encoded to obtain a compressed data packet; Performing network topology analysis on the compressed data packet to obtain a multi-path transmission scheme, and performing routing distribution on the data packet according to the multi-path transmission scheme to obtain a distributed data stream; Performing transmission quality monitoring on the distributed data stream to obtain link status information, and performing adaptive path adjustment according to the link status information to obtain an optimized transmission path; The data on the optimized transmission path are merged in real time to obtain a merged data set, and the merged data set is timestamp aligned and data decoded to obtain an original monitoring data set.

4. The intelligent inspection and fault location method for power distribution equipment according to claim 1, characterized in that: The cleaning, standardization and fusion processing of the original monitoring data set to obtain pre-processed structured data includes: Performing preliminary detection processing on the original monitoring data set to obtain a data detection report, and performing abnormal repair on the data detection report to obtain a repaired data set; Performing data type identification on the repaired data set to obtain a data type mapping table, and normalizing the data according to the data type mapping table to obtain a standardized data set; Performing feature extraction on the standardized data set to obtain a feature vector set, and performing dimensionality reduction processing on the feature vector set to obtain a feature matrix after dimensionality reduction; Performing data alignment on the feature matrix after dimension reduction to obtain time-aligned multi-source data, and performing data fusion on the time-aligned multi-source data to obtain a fused feature tensor; The fused feature tensor is reorganized into a data structure to obtain a structured data template, and the fused data is formatted according to the structured data template to obtain preprocessed structured data.

5. The intelligent inspection and fault location method for power distribution equipment according to claim 1, characterized in that: The multi-dimensional spatial correlation analysis of the structured data is performed by a multi-scale time series analysis algorithm to obtain health status assessment data of the power distribution equipment, including: Performing data analysis on the structured data to obtain an analysis feature set, and performing frequency domain conversion on the analysis feature set to obtain a frequency domain feature matrix; Performing wavelet decomposition on the frequency domain feature matrix to obtain a time-frequency feature graph, and performing principal component analysis on the time-frequency feature graph to obtain main feature components; Performing spatial correlation calculation on the main characteristic components to obtain a correlation matrix, and performing cluster analysis on the correlation matrix to obtain characteristic groups; Performing time series pattern mining on the feature group to obtain a device operation state sequence, and performing trend analysis on the device operation state sequence to obtain a health state change trend; The health status of the equipment is scored according to the health status change trend to obtain health status assessment data.

6. The intelligent inspection and fault location method for power distribution equipment according to claim 1, characterized in that: The method of performing dynamic threshold calculation and intelligent diagnosis on the power distribution equipment based on the health status assessment data to obtain early warning information and preliminary diagnosis results includes: Performing parameter analysis on the health status assessment data to obtain a parameter feature set, and performing threshold calculation on the parameter feature set to obtain a dynamic threshold sequence; Adaptively adjusting the dynamic threshold sequence to obtain a threshold adjustment rule, and performing threshold judgment on the real-time health status data according to the threshold adjustment rule to obtain an early warning trigger signal; Performing frequency analysis on the warning trigger signal to obtain a warning level classification standard, and classifying the warning information according to the warning level classification standard to obtain graded warning information; Extracting symptom features from the graded warning information to obtain a fault symptom feature set, and performing pattern matching on the fault symptom feature set to obtain a candidate fault type; A probability evaluation is performed on the candidate fault types to obtain a fault probability distribution, and a preliminary diagnosis result is generated according to the fault probability distribution.

7. The intelligent inspection and fault location method for power distribution equipment according to claim 1, characterized in that: The expert knowledge base matching and multi-criteria decision analysis are performed on the warning information and the preliminary diagnosis results to obtain a maintenance decision suggestion plan, including: Performing feature analysis on the warning information and the preliminary diagnosis results to obtain a feature mapping set, and performing case matching on the feature mapping set to obtain a similar case set; Calculating case similarity for the similar case set to obtain a case matching ranking, and selecting the top N most similar cases according to the case matching ranking to obtain a candidate solution set, where N is a positive integer greater than 1; Performing feasibility evaluation on the candidate solution set to obtain solution evaluation indicators, and constructing a multi-criteria decision matrix based on the solution evaluation indicators to obtain a decision weight vector; Performing fuzzy comprehensive evaluation on the decision weight vector to obtain a comprehensive score of the scheme, and sorting the candidate schemes according to the comprehensive score of the scheme to obtain an optimal maintenance scheme; A cost-benefit analysis is performed on the preferred maintenance plan to obtain a plan execution suggestion, and a maintenance decision suggestion plan is generated based on the plan execution suggestion.

8. An intelligent inspection and fault location system for power distribution equipment, used to implement the intelligent inspection and fault location method for power distribution equipment as described in any one of claims 1 to 7, characterized in that: The intelligent inspection and fault location system of the power distribution equipment includes: A deployment module is used to deploy multiple types of sensors at key points around the power distribution equipment to obtain an intelligent sensor network; A collection module, used for performing multi-mode collection and multi-path transmission processing on the data collected by the intelligent sensor network to obtain an original monitoring data set; A fusion module is used to clean, standardize and fuse the original monitoring data set to obtain pre-processed structured data; An analysis module, used to perform multi-dimensional spatial correlation analysis on the structured data by using a multi-scale time series analysis algorithm to obtain health status assessment data of the power distribution equipment; A diagnosis module, used to perform dynamic threshold calculation and intelligent diagnosis on the power distribution equipment based on the health status assessment data, and obtain early warning information and preliminary diagnosis results; The decision-making module is used to perform expert knowledge base matching and multi-criteria decision analysis on the warning information and the preliminary diagnosis results to obtain a maintenance decision recommendation plan.

Images