A power distribution equipment remote diagnosis method based on edge computing

Abstract

The application discloses a power distribution equipment remote diagnosis method based on edge computing and particularly relates to the technical field of power equipment intelligent monitoring, and comprises the following steps: an edge computing node collects power distribution equipment operation state data in real time; a preliminary diagnosis result and key data are generated by performing diagnosis analysis locally on the node; structured data is uploaded to the cloud according to a preset strategy and integrity is verified; correlation analysis is performed on the data of multiple nodes on the cloud platform to identify common abnormalities, dynamically optimize a diagnosis algorithm, and automatically trigger alarm and work order distribution in stages, and quick closed-loop processing is realized in combination with the positions and skills of operation and maintenance personnel. Through the cooperation of the edge and the cloud, the application reduces communication bandwidth occupation, improves diagnosis accuracy and real-time performance, shortens fault response time, and is suitable for power distribution network line-level monitoring and operation and maintenance management.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power equipment, and more specifically, to a remote diagnostic method for power distribution equipment based on edge computing. Background Technology

[0002] Existing remote diagnostic technologies for power distribution equipment mostly adopt a centralized cloud processing model. They primarily collect operational status data from various sensors, then upload the collected data directly to a remote cloud platform. The cloud platform performs diagnostic analysis and anomaly detection on the data, and generates corresponding maintenance instructions or alarm information based on the analysis results. These instructions are then pushed to the terminals of maintenance personnel to guide equipment maintenance work.

[0003] However, in practical use, it still has some drawbacks, such as high data transmission pressure, with full data upload leading to excessive communication bandwidth consumption, which can easily cause network congestion and transmission delays in densely populated areas, affecting the real-time performance of diagnosis; over-reliance on cloud processing, with a lack of localized analysis capabilities at the edge, which can easily lead to interruption of the diagnostic process when the network is interrupted or the cloud load is too high, making it impossible to identify equipment anomalies in a timely manner; limited diagnostic accuracy, based on isolated analysis of data from a single node, making it difficult to correlate common anomaly trends across multiple devices and regions, and algorithm updates rely on manual intervention, resulting in high false alarm and false negative rates; and low emergency response efficiency, with cumbersome processes for judging emergency situations and dispatching work orders, lacking an automated hierarchical processing mechanism, which may delay fault handling and increase the risk of equipment damage or power outages. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides a remote diagnostic method for power distribution equipment based on edge computing, which solves the problems mentioned in the background art through the following scheme.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a remote diagnostic method for power distribution equipment based on edge computing, comprising:

[0006] S1: Data Acquisition and Transmission: Deploy edge computing nodes at the power distribution equipment site or in the vicinity to collect real-time operating status data of the power distribution equipment;

[0007] S2: Edge-side diagnostic analysis: Edge computing nodes execute diagnostic analysis algorithms on the collected operational status data to generate preliminary diagnostic results and key data; the key data includes: key indicators in the operational status data and feature data generated during the diagnostic analysis process;

[0008] S3: Upload Diagnostic Results: The edge computing node uploads structured data to a remote cloud platform via a communication network; the structured data includes: the preliminary diagnostic results and the screened key data;

[0009] S4: Cloud Processing and Feedback: The remote cloud platform performs the following processing based on the received structured data and system configuration policies:

[0010] When diagnostic data from multiple nodes is received: perform correlation analysis and comprehensive decision-making on the diagnostic results of multiple edge nodes;

[0011] When the algorithm optimization trigger condition is met: update the diagnostic algorithm based on the multi-node data analysis results, and push the optimized algorithm to the edge nodes;

[0012] When an abnormal situation requiring urgent handling is detected: generate an equipment maintenance work order or a real-time alarm, and push it to the relevant maintenance personnel's terminal.

[0013] Preferably, the operating status data includes: data collected by current sensors, voltage sensors, vibration sensors, and infrared temperature sensors; the collected analog signals are adjusted in amplitude and noise is suppressed by signal conditioning circuits, and then converted into digital signals; all digital signals are added with a unified time stamp with millisecond precision, and integrated into a standard data packet containing device ID, timestamp, and monitoring parameters according to a preset time window, and stored in the edge node cache area.

[0014] Preferably, the diagnostic analysis algorithm includes: cleaning and smoothing the operating status data; setting multi-level thresholds for temperature and humidity parameters to determine anomalies; calculating the rate of parameter change to identify abnormal trends; and extracting the proportion of specific harmonics as key features through harmonic analysis.

[0015] Preferably, the key data screening includes: retaining abnormal parameters directly related to the preliminary diagnosis results; screening the feature data generated by the diagnosis based on feature importance, retaining only features with importance higher than a preset threshold; and integrating the preliminary diagnosis results and the screened key data according to node identifier, timestamp, diagnosis results, and key data format.

[0016] Preferably, the feature importance includes: calculated by the information gain algorithm and verified based on historical datasets, the threshold can filter redundant features with a false alarm rate of less than 5%, and retain features with an importance higher than 0.3.

[0017] Preferably, the uploading of the structured data includes: when the parameters do not exceed the multi-level threshold, it is a normal state and is uploaded at a fixed period; otherwise, it is an abnormal state and triggers immediate uploading; a verification value is generated before transmission to verify data integrity; if no confirmation is received from the cloud platform, retransmission is performed according to the exponential backoff algorithm.

[0018] Preferably, the correlation analysis and comprehensive decision-making includes: determining the range of associated nodes based on the distribution network topology or geographical coordinates; calculating the similarity of the diagnostic results of associated nodes and identifying common abnormal trends; verifying the consistency of key data and distinguishing between systematic anomalies and independent anomalies; and generating regional or single-node decisions based on the verification results.

[0019] Preferably, the updated diagnostic algorithm includes: constructing an optimized dataset based on multi-node historical data and operation and maintenance feedback; iteratively adjusting algorithm parameters and verifying the false alarm rate and false negative rate through a test set; and pushing the accepted algorithm to edge nodes and updating it using a breakpoint resume mechanism.

[0020] Preferably, the generation of equipment maintenance work orders or real-time alarms includes: classifying the status level according to the severity of abnormal parameters; generating real-time alarms with risk descriptions and handling measures for high-level statuses; allocating work orders based on the location and skill matching of maintenance personnel, and tracking the processing progress.

[0021] The technical effects and advantages of this invention are as follows:

[0022] Reduce data transmission pressure and improve real-time diagnostics: By using edge computing nodes to filter and perform preliminary analysis of operational status data locally, only the structured preliminary diagnostic results and the filtered key data are uploaded to the cloud, which significantly reduces the amount of data uploaded, reduces communication bandwidth consumption, effectively avoids network congestion and transmission delays in densely populated areas, and ensures efficient and real-time diagnostics.

[0023] Enhance system stability and independence, and reduce dependence on the cloud: Edge computing nodes have localized diagnostic and analysis capabilities, enabling them to independently complete tasks such as data collection and preliminary diagnosis. In the event of network interruption or excessive cloud load, they can promptly identify device anomalies, ensuring the continuity and stability of the diagnostic process and improving system reliability.

[0024] Improve diagnostic accuracy and adaptability, and optimize algorithm iteration: The cloud can perform correlation analysis and comprehensive decision-making on the diagnostic results of multiple edge nodes, identify common abnormal trends of multiple devices and regions, distinguish between systematic abnormalities and independent abnormalities, and improve the accuracy of diagnosis; Based on the data analysis results of multiple nodes, the diagnostic algorithm can be dynamically updated, and the optimized algorithm can be pushed to edge nodes to realize automatic optimization and iteration of the algorithm, reducing the false alarm rate and false negative rate;

[0025] Improving emergency response efficiency and ensuring safe equipment operation: By classifying the severity of abnormal parameters into different levels, corresponding real-time alarms or equipment maintenance work orders are generated for different levels of conditions. Work orders are allocated based on the location and skill matching of maintenance personnel, and the processing progress is tracked. This forms an automated hierarchical processing mechanism, which speeds up emergency response, avoids delays in fault handling, reduces the risk of equipment damage and power outages, and ensures the safe and stable operation of power distribution equipment. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0027] Figure 2 This is a schematic diagram of the cloud-based correlation analysis structure of the present invention.

[0028] Figure 3 This is a schematic diagram of the alarm and work order processing structure of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] refer to Figures 1-3 The method for remote diagnostics of power distribution equipment based on edge computing, as shown, includes:

[0031] S1: Data Acquisition and Transmission: Edge computing nodes deployed at or near the power distribution equipment site collect real-time operating status data of the power distribution equipment;

[0032] S2: Edge-side diagnostic analysis: Edge computing nodes execute diagnostic analysis algorithms on the collected operational status data to generate preliminary diagnostic results and key data; the key data includes: key indicators in the operational status data and feature data generated during the diagnostic analysis process;

[0033] S3: Upload Diagnostic Results: The edge computing node uploads structured data to a remote cloud platform via a communication network; the structured data includes: the preliminary diagnostic results and the screened key data;

[0034] S4: Cloud Processing and Feedback: The remote cloud platform performs the following processing based on the received structured data and system configuration policies:

[0035] When diagnostic data from multiple nodes is received: perform correlation analysis and comprehensive decision-making on the diagnostic results of multiple edge nodes;

[0036] When the algorithm optimization trigger condition is met: update the diagnostic algorithm based on the multi-node data analysis results, and push the optimized algorithm to the edge nodes;

[0037] When an abnormal situation requiring urgent handling is detected: generate an equipment maintenance work order or a real-time alarm, and push it to the relevant maintenance personnel's terminal.

[0038] S1: Edge computing nodes deployed at the power distribution equipment site or in adjacent areas within 200 meters collect real-time operating status data of the power distribution equipment; raw electrical signals are captured in real time by current and voltage sensors directly installed on the electrical circuits, while physical motion is monitored by vibration sensors attached to mechanical parts, and the surface heat distribution of the equipment is scanned using non-contact infrared temperature sensors; the analog signals output by these sensors are first subjected to amplitude adjustment and noise suppression by signal conditioning circuits, and then converted into digital signals by analog-to-digital converters; all digital signals are integrated into structured data frames according to preset time windows with millisecond-level precision added based on the GPS clock, and finally output as standard data packets containing equipment ID, timestamp, current, voltage, mechanical vibration characteristics and temperature distribution, which are stored in the edge node buffer area.

[0039] It should be further explained that amplitude adjustment amplifies the weak analog signal output by the sensor to a range that the analog-to-digital converter can recognize; noise suppression removes environmental electromagnetic interference through filtering circuits.

[0040] It should be further explained that a unified time scale with millisecond-level precision is added to the GPS clock: signals collected by multiple sensors need to be aligned based on the same time base. If the time scale is not unified, it will lead to time mismatch between current anomalies and temperature anomalies, affecting subsequent diagnosis; millisecond-level precision can meet the time correlation requirements of transient signals of power distribution equipment.

[0041] It is necessary to further explain that the edge node cache is used for temporary data storage, buffering data traffic, and the basic mechanism of the cache.

[0042] Step S2: Clean the collected operating status data, remove outliers, and smooth high-frequency fluctuation data with a frequency greater than 100Hz using a moving average method. The formula is as follows:

[0043] ,in for Time-smoothed data, for The original data collected at that moment. To adjust the sliding window size;

[0044] For temperature ( ),humidity( ) and other parameters to set three levels of thresholds, such as:

[0045] Temperature: Normal Early warning ,abnormal ;

[0046] Humidity: Normal ,abnormal or ;

[0047] The slope of parameter change is calculated using linear regression, using the following formula: ,in For the rate of change, It is a time series. For parameter values, The number of time series data points, when (Temperature) or (Humidity) is considered an abnormal trend;

[0048] For current ( Harmonic analysis is performed, and the harmonic content is calculated using Fourier transform:

[0049] in, For harmonic order, for The proportion of subharmonics for RMS value of subharmonic current Extracting the effective value of the fundamental current. , As a key feature;

[0050] Preliminary diagnostic results: A structured description is generated based on threshold, trend, and harmonic characteristic anomalies, such as temperature anomalies (current). Heats up within 10 minutes ,rate );

[0051] Key data: Includes key indicators from the raw data and feature data generated from the diagnosis.

[0052] It needs to be further explained that taking , This is a key feature because the third and fifth harmonics are the most common and significant harmonic components in the power system, which are also the most prevalent in the operating characteristics of power distribution equipment.

[0053] Key indicators include temperature, peak current, humidity, and voltage-related parameters.

[0054] The characteristic data include the rate of temperature change, the proportion of the third harmonic, the proportion of the fifth harmonic, and the slope of parameter change.

[0055] S3: The key data generated in S2 is simplified according to preset rules. For key indicators in the operational status data, only abnormal parameters directly related to the preliminary diagnosis result are retained. For example, if the preliminary diagnosis result is an abnormal temperature, only the temperature value and the duration of the abnormal temperature are retained, while normal data such as humidity and voltage are discarded. For feature data generated during the diagnostic analysis process, the feature importance is calculated using an information gain algorithm, with the following formula: ,in Features Information gain For dataset entropy, Features Given conditions The conditional entropy retains only high-importance features with information gain greater than 0.3, such as the rate of temperature change and the proportion of the third harmonic;

[0056] The preliminary diagnostic results and the selected key data are integrated in a unified format, including four core parts: node identifier, timestamp, diagnostic result, and key data. The node identifier is used to identify the edge computing node from which the data originates, the timestamp records the specific time the data was generated, and the diagnostic result presents temperature anomalies (current...). Exceeding the threshold The structured description of the data is summarized, while the key data is the filtered outlier parameters and high-importance features.

[0057] Under normal conditions, integrated data is uploaded via the 5G private network at fixed intervals, with TCP protocol ensuring reliable transmission. In case of abnormal situations requiring immediate handling, an instant upload mechanism is triggered, prioritizing communication resources and keeping transmission latency below 100ms. Before data transmission, integrity is verified using the SHA-256 algorithm, and a checksum is generated and sent along with the data frame. The formula is: The random salt value is dynamically generated by the edge nodes to ensure the uniqueness of the hash value transmitted each time;

[0058] The edge node receives the confirmation message from the cloud platform. If it does not receive it within 10 seconds, it automatically initiates retransmission (up to 3 times). The retransmission interval is calculated using the exponential backoff algorithm. ,in For the number of retransmissions and The initial interval is set to 1 second to avoid network congestion.

[0059] It should be further noted that the key data after screening are abnormal parameters that are directly related to the preliminary diagnostic results;

[0060] feature The feature representing the current importance to be evaluated is a specific feature data generated during the diagnostic analysis process, such as the rate of temperature change or the proportion of the third harmonic; dataset For features A collection of relevant power distribution equipment operating status data and diagnostic results data;

[0061] If the information gain is less than or equal to 0.3, the feature has a weak ability to distinguish abnormal device states and contributes little to the diagnostic results. Setting the feature selection threshold to 0.3 can filter out redundant features, reduce the amount of data uploaded to the cloud, avoid occupying too many communication resources, and ensure that key diagnostic information can be transmitted efficiently.

[0062] The SHA-256 algorithm is used to generate a check value, and the random salt value is dynamically generated by the edge node to ensure the uniqueness of the hash value for each transmission. When the data frame arrives at the remote cloud platform, the cloud platform can recalculate the check value of the received data using the same algorithm and compare it with the check value sent with the data frame. If the two are consistent, it means that the data has not been tampered with or damaged during transmission. If they are inconsistent, it can be determined that the data is abnormal, thereby realizing the verification of data integrity.

[0063] S4: When diagnostic data from multiple nodes is received:

[0064] Data association dimension determination: Based on the distribution network topology (pre-stored in the topology database of the cloud platform, including node locations, line connection relationships, etc.), determine the range of nodes to be associated; for example: if diagnostic data of 5 switchgear nodes on a certain 10kV line is received, these 5 nodes on the same line are given priority to be grouped into an association group; if node data of different lines but adjacent areas are received, nodes with a distance of less than or equal to 1km are included in the association analysis in combination with geographical coordinate information.

[0065] Diagnostic result similarity calculation: Feature extraction is performed on the preliminary diagnostic results of each node within the associated group (such as extracting core anomaly labels such as temperature anomaly, harmonic anomaly, and humidity anomaly), and the similarity of the results is calculated using the Jaccard similarity coefficient. The formula is as follows: ,in These are the sets of anomaly labels for the two nodes, The number of common anomaly labels, The total number of abnormal labels for two nodes; when there are 3 or more nodes in the associated group. At that time, it was determined that there was a common abnormal trend;

[0066] Cross-validation of key data: For key data corresponding to common abnormal trends (such as temperature values ​​and heating rates corresponding to temperature anomalies), calculate data consistency, for example: statistically analyze the mean and standard deviation of temperature anomaly values ​​within the associated group. ,in These are the abnormal temperature values ​​for each node. This represents the number of associated nodes. (i.e., low data dispersion) confirms that the anomaly is caused by the same type of reason (e.g., excessive line load); if If so, it is determined that each node is abnormal independently (such as a single node device aging).

[0067] Comprehensive decision generation: Decisions are generated by combining similarity and data consistency results. If there are common anomalies and the data is consistent, regional or line-level decisions are generated; if there are independent anomalies, single-node-specific decisions are generated.

[0068] Among them, when the algorithm optimization trigger condition is met:

[0069] Historical data filtering and annotation: Extract multi-node data related to the target algorithm from the historical database of the cloud platform, including the original running status data uploaded by edge nodes, preliminary diagnostic results, and actual abnormal or normal verification results reported by operation and maintenance personnel, and construct an algorithm optimization dataset;

[0070] Algorithm parameter iterative optimization: Iterative adjustments are made to the core parameters of the algorithm based on the dataset;

[0071] Optimization algorithm verification: Validate the optimized algorithm on the test set (30% of the samples are randomly selected from the dataset), calculate the false positive rate (number of false positives / total number of diagnoses) and false negative rate (number of false negatives / actual number of anomalies) of the new algorithm. If the false positive rate is less than or equal to 5% and the false negative rate is less than or equal to 3% (meeting the system configuration acceptance criteria), then the algorithm optimization is confirmed to be effective.

[0072] Algorithm push and node update: The cloud platform packages the optimized algorithm and sends it to the target edge node through targeted push and breakpoint resume mechanism. First, a push command is sent to the node. After the node returns a reception ready signal, the transmission begins. During the transmission, a check value is generated every 100KB. If the transmission is interrupted, it resumes from the breakpoint based on the check value. After receiving the algorithm, the node automatically replaces the old algorithm and returns a confirmation of successful update to the cloud platform after the replacement is completed.

[0073] When an abnormal situation requiring urgent handling is detected:

[0074] Status level determination: The status level is determined based on the preset level classification rules (e.g., Level 1: may cause equipment tripping or power outage; Level 2: requires handling within 24 hours; Level 3: requires handling within 72 hours) and the severity of abnormal parameters. For example, a temperature of 150℃ (exceeding the threshold of 80℃) is determined as Level 1, and a temperature of 95℃ (exceeding the threshold of 25℃) is determined as Level 2.

[0075] Real-time alarm generation and push: Real-time alarm information is generated for Level 1 and Level 2 conditions. The alarm information includes node location, abnormal parameters, risk description, and suggested handling measures. The alarms are pushed to the operation and maintenance personnel's terminals in parallel through multiple channels, including SMS, operation and maintenance APP push, and dispatch center large screen display.

[0076] Equipment maintenance work order generation and allocation: For level 2 and 3 situations, standardized work orders are generated. The standardized work order includes work order number, equipment information, anomaly description, planned processing time limit, and required tools or spare parts. The cloud platform allocates work orders based on the location and skill matching of maintenance personnel. The allocation of work orders is calculated by calculating the distance between the current location of the maintenance personnel and the equipment, and combining the personnel's skill tags to select personnel whose distance is less than or equal to 5km and whose skills match as the work order recipients. The work orders are pushed through the maintenance management system, and the recipients must confirm the acceptance of the work order within 10 minutes.

[0077] Processing progress tracking: The cloud platform receives real-time feedback on the processing progress from the terminals of operation and maintenance personnel. If the process is not completed within the planned time limit, an overdue reminder will be automatically pushed to the relevant person in charge to ensure closed-loop processing of the situation.

[0078] It should be further noted that when the overlap of abnormal labels among multiple nodes exceeds 60% (i.e., This indicates that these nodes not only have a single overlapping anomaly, but also share multiple related anomaly features. This strong correlation usually points to systemic causes such as line overload or sudden changes in the regional environment, rather than random failures of the equipment itself. If the threshold is 0.5, occasional occurrences with only a single identical label may be misjudged as system failures, significantly increasing the risk of false alarms.

[0079] Regional or line-level decision-making refers to the global control measures taken when the cloud determines that multiple nodes have systemic faults. These measures include adjusting the output power of upstream substations to reduce the load on the entire line, switching power supply circuits to isolate faulty sections, or issuing regional-level instructions. Single-node targeted decision-making, on the other hand, generates a handling plan that only applies to the fault of an individual device.

[0080] Algorithm optimization requires learning data patterns through the training set and verifying generalization ability through the test set. If the test set ratio is less than 10%, the insufficient sample size may prevent the coverage of all fault types, leading to a high degree of randomness in the verification results. If the ratio is too high as 50%, the training set sample size decreases, and the algorithm may not be able to fully learn the long-term operating characteristics of the equipment, easily resulting in overfitting. A ratio of 30% can strike a balance between the two, ensuring that the test set has enough samples to cover various operating conditions while reserving 70% of the samples in the training set to support the algorithm's full learning.

[0081] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0082] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An edge-computing-based power distribution equipment remote diagnosis method, characterized in that, Comprise: S1: data acquisition and transmission: deploy edge computing nodes in the power distribution equipment field or adjacent area, real-time acquisition of operation state data of power distribution equipment; S2: edge side diagnostic analysis: the edge computing node executes diagnostic analysis algorithm on the collected operation state data to generate preliminary diagnostic results and key data; The key data includes: key indicators in the operation state data and feature data generated by the diagnostic analysis process; The diagnostic analysis algorithm includes: cleaning and smoothing processing of operation state data; setting multiple threshold values for temperature and humidity parameters to determine abnormalities; calculating parameter change rate to identify trend anomalies; extracting specific harmonic proportion as key features through harmonic analysis; S3: diagnostic result upload: the edge computing node uploads structured data to a remote cloud platform through a communication network; the structured data includes: the preliminary diagnostic results and the filtered key data; The filtering of the key data includes: retaining abnormal parameters directly related to the preliminary diagnostic results; filtering feature data generated by the diagnostic analysis based on feature importance, only retaining features with importance higher than the feature filtering threshold; integrating the preliminary diagnostic results and the filtered key data according to node identifier, timestamp, diagnostic result and key data format; S4: cloud processing and feedback: the remote cloud platform executes the following processing according to the received structured data and system configuration strategy: When receiving diagnostic data from multiple nodes: perform correlation analysis and comprehensive decision-making on the diagnostic results of multiple edge nodes; When the algorithm optimization trigger condition is met: update the diagnostic algorithm based on the analysis results of multiple node data, and push the optimized algorithm to the edge nodes; When detecting abnormal conditions that need emergency handling: generate a device maintenance work order or real-time alarm and push it to the terminal of the relevant operation and maintenance personnel. 2.The power distribution equipment remote diagnosis method based on edge computing of claim 1, wherein, The operation state data includes: collected by current sensor, voltage sensor, vibration sensor and infrared temperature sensor; the analog signals are adjusted in amplitude and noise suppressed by a signal conditioning circuit, and then converted into digital signals; all digital signals are added with a unified timestamp of millisecond level precision, integrated into standard data packets containing device ID, timestamp and monitoring parameters according to the preset time window, and stored in the edge node cache area. 3.The power distribution equipment remote diagnosis method based on edge computing of claim 1, wherein, The feature importance includes: calculated by information gain algorithm, verified based on historical data set, the feature filtering threshold can filter redundant features with a false positive rate less than 5%, and retain features with importance higher than 0.

3.

4. The power distribution equipment remote diagnosis method based on edge computing according to claim 1, characterized in that, The uploading of structured data includes: when the parameters do not exceed the multi-level threshold, it is in normal state and uploaded at a fixed period; otherwise, it is in abnormal state and triggers immediate upload; a check value is generated before transmission for data integrity verification; if no confirmation is received from the cloud platform, retransmission is performed according to the exponential backoff algorithm.

5. The edge computing-based power distribution equipment remote diagnosis method according to claim 1, characterized in that, The correlation analysis and comprehensive decision-making include: determining the range of associated nodes based on the topology relationship or geographical coordinates of the power distribution network; calculating the similarity of the diagnostic results of the associated nodes to identify common abnormal trends; verifying the consistency of the key data to distinguish between systematic and independent abnormalities; generating regional or single node decisions based on the verification results.

6. The power distribution equipment remote diagnosis method based on edge computing according to claim 1, characterized in that, The updating diagnosis algorithm comprises: constructing an optimization dataset based on multi-node historical data and operation and maintenance feedback; iteratively adjusting algorithm parameters and verifying false positive rate and missed detection rate through a test set; and pushing the accepted algorithm to the edge node and updating it using a breakpoint resume mechanism.

7. The power distribution equipment remote diagnosis method based on edge computing according to claim 1, characterized in that, The generating device maintenance work order or real-time alarm comprises: dividing a condition level according to the severity of an abnormal parameter; generating a real-time alarm containing a risk description and a treatment measure for a high-level condition; assigning a work order based on the location and skill matching degree of an operation and maintenance personnel, and tracking the processing progress.

Images