Differentiated operation and maintenance method and system based on real-time weak link of equipment

Abstract

The invention relates to the field of electrical power grid operation and maintenance, in particular to a differentiated operation and maintenance method and system based on real-time weak links of equipment. According to the method, online monitoring data and power supply reliability operation data of power distribution network equipment are collected in real time to form an equipment state multi-source data set; based on the data set, calculating an equipment health degree evaluation result by adopting a state evaluation model, and obtaining an importance degree evaluation result through an analytic hierarchy process in combination with the position and load importance of the equipment in the network topology; and inputting the two results into a preset operation and maintenance strategy matrix to match a corresponding operation and maintenance strategy type, and finally generating a differentiated operation and maintenance strategy instruction containing specific operation and maintenance measures, an execution time window and a resource allocation scheme. According to the method and the device, the problems of lack of equipment-level pertinence and unreasonable resource allocation of an operation and maintenance strategy in the prior art are effectively solved, and the operation and maintenance accuracy and the resource utilization efficiency are improved.

Description

A differentiated operation and maintenance method and system based on real-time weak links of equipment Technical Field

[0001] This invention relates to the field of electrical power grid operation and maintenance, and in particular to a differentiated operation and maintenance method and system based on real-time weak links of equipment. Background Technology

[0002] With the large-scale integration of distributed energy resources, the continuous increase in user electricity demand, and the frequent occurrence of extreme weather events, the operating environment of distribution networks is becoming increasingly complex, significantly increasing the risk of equipment failure and placing higher demands on the reliability of power supply. Distribution networks contain diverse types of equipment widely distributed, with significant differences in their operating status, aging levels, and roles within the network. This necessitates a shift in operation and maintenance (O&M) from the traditional "one-size-fits-all" approach to a more refined and differentiated approach, aiming to achieve optimal allocation of limited O&M resources and minimize the impact of failures on power supply. Existing technologies have attempted to improve the scientific nature of distribution network O&M through data analysis. For example, existing technology CN120069558A discloses a data-based method and system for assessing the reliability of distribution network power supply. This method integrates meteorological, GIS, and equipment monitoring data to train a meteorological-failure rate model, calculates the system risk index and line risk value to locate high-risk areas, and implements priority scheduling, load transfer, and tiered load shedding strategies based on scheduling priority coefficients, thereby improving the risk prevention and control capabilities of the distribution network to a certain extent. However, this solution primarily focuses on regional-level risk assessment and macro-level scheduling, failing to delve into the real-time health status characteristics of individual devices or fully integrate the device's location weight and load importance within the network topology for precise quantitative assessment. This results in a lack of device-level specificity in the formulated operation and maintenance strategies. Specifically, it only adopts uniform load transfer or shelving measures for high-risk areas, unable to develop personalized operation and maintenance plans based on differences in the degree of health degradation and the scope of fault impact of different devices. This often leads to untimely maintenance of critical equipment and excessive maintenance of ordinary equipment, resulting in the dual problems of wasted operation and maintenance resources and insufficient power supply reliability, making it difficult to meet the actual needs of refined operation and maintenance of distribution networks. Summary of the Invention

[0003] This invention provides a differentiated operation and maintenance method and system based on real-time weak links of equipment, aiming to solve the problems of lack of equipment-level targeting and unreasonable resource allocation in existing technologies.

[0004] To achieve the above objectives, the following technical solution is adopted.

[0005] A differentiated operation and maintenance method based on real-time weak links of equipment includes the following steps: real-time collection of online monitoring data and power supply reliability operation data of equipment in the distribution network to obtain a multi-source data set of equipment status; based on the multi-source data set of equipment status, a status evaluation model is used to calculate the current health status score of the equipment to obtain a health evaluation result; based on the power supply reliability operation data, combined with the location and load importance of the equipment in the network topology, the analytic hierarchy process (AHP) is used to evaluate the impact of equipment failures on power supply reliability to obtain an importance evaluation result; the health evaluation result and the importance evaluation result are input into a preset operation and maintenance strategy matrix for matching, wherein the operation and maintenance strategy matrix defines the operation and maintenance strategy types corresponding to different combinations of health level and importance level; based on the matching result, a differentiated operation and maintenance strategy instruction for the equipment is generated, wherein the instruction includes specific operation and maintenance measures, execution time windows, and resource allocation schemes.

[0006] Optionally, the step of acquiring online monitoring data and power supply reliability operation data of equipment in the distribution network in real time to obtain a multi-source dataset of equipment status specifically includes: synchronously accessing the data streams of the equipment online monitoring system, the distribution automation system, and the power supply service command system through a data interface to obtain raw multi-source data including equipment electrical quantity data, equipment operating condition data, and power supply interruption event records; performing time-scale alignment and data cleaning on the raw multi-source data, removing outliers and filling missing values ​​to obtain normalized multi-source time series data; for the normalized multi-source time series data, using a sliding time window mechanism to extract statistical features within each sliding time window, the statistical features including mean, variance, kurtosis, and trend slope to obtain a primary feature set; and associating and fusing the primary feature set with static attribute data from the equipment ledger, the static attribute data including equipment model, commissioning year, and family defect information to form the multi-source dataset of equipment status that integrates dynamic operating status and static background attributes.

[0007] Optionally, the step of calculating the current health status score of the equipment using a state evaluation model to obtain a health evaluation result specifically includes: selecting a subset of features directly related to the physical degradation mechanism of the equipment from the multi-source dataset of equipment status to form a health evaluation feature vector; inputting the health evaluation feature vector into a pre-trained deep learning anomaly detection model, where the deep learning anomaly detection model maps the input features to a low-dimensional latent space through an encoder and reconstructs them through a decoder, calculating the reconstruction error between the input features and the reconstructed features to obtain an anomaly score representing the deviation of the equipment status from the normal operating mode; inputting the anomaly score into a dynamic failure rate prediction module based on the Weibull proportional hazards model, and calculating the probability of conditional failure of the equipment in a specific future time period by combining the equipment's years of operation with the average lifespan of similar equipment; fusing the anomaly score and the probability of conditional failure, and normalizing and mapping them to a preset health score range using a fuzzy comprehensive evaluation method, and finally outputting the quantified health evaluation result.

[0008] Optionally, the step of using the analytic hierarchy process (AHP) to assess the impact of equipment failures on power supply reliability and obtain an importance evaluation result specifically includes: constructing a hierarchical model with "impact on power supply reliability" as the top-level objective. The criterion layer of the hierarchical model includes the magnitude of load loss caused by equipment failure, the impact on the number of users during power outages, network topology reconfiguration capability, and the average time required to restore power; based on the location of the equipment in the network topology, quantifying the load loss caused by equipment failure and the number of users during power outages through power flow calculation and single-component failure simulation to obtain a first set of quantitative indicators; analyzing the network topology structure, calculating the equipment's ability to be supplied with load by other paths after a failure and the number of switches required to operate to obtain a second set of quantitative indicators; comparing each indicator in the criterion layer of the hierarchical model pairwise to construct a judgment matrix, and calculating the weight coefficient of each indicator after passing a consistency check; standardizing the first set of quantitative indicators and the second set of quantitative indicators, and performing a linear weighted sum based on the weight coefficients to calculate the comprehensive importance score for each device, thereby obtaining the importance evaluation result.

[0009] Optionally, the step of inputting the health evaluation results and importance evaluation results into a preset operation and maintenance strategy matrix for matching specifically includes: the operation and maintenance strategy matrix is ​​a two-dimensional decision table, whose horizontal axis is divided into multiple health level intervals based on the numerical range of the health evaluation results, and whose vertical axis is divided into multiple importance level intervals based on the numerical range of the importance evaluation results; according to the health level interval to which the current device's health evaluation result belongs, and the importance level interval to which the importance evaluation result belongs, the corresponding cell is located in the two-dimensional decision table, and the first operation and maintenance strategy type is predefined in this cell; after matching the first operation and maintenance strategy type, a resource constraint check process is triggered to check whether the currently available human resources, material resources, and time window resources meet the execution requirements of the first operation and maintenance strategy type; if the resources meet the requirements, the first operation and maintenance strategy type is directly confirmed as the final strategy; if the resources do not meet the requirements, a dynamic strategy adjustment mechanism is initiated to find a second operation and maintenance strategy type in the operation and maintenance strategy matrix that is adjacent to the current cell and whose resource requirements can be met as an alternative strategy.

[0010] Optionally, the step of generating differentiated operation and maintenance strategy instructions based on the matching results specifically includes: when the final strategy is a planned maintenance strategy, the strategy generation engine calls the maintenance knowledge base, matches the corresponding standard maintenance procedures, required spare parts list, and estimated working hours from the standard maintenance operation library according to the specific values ​​of the equipment type and health evaluation results, and generates a planned maintenance instruction containing specific work order content; when the final strategy is a condition monitoring enhancement strategy, the strategy generation engine adjusts the parameters of the online monitoring device related to the equipment, including increasing the monitoring data sampling frequency, enabling the diagnostic algorithm model, and setting stricter early warning thresholds, and generates a monitoring parameter adjustment instruction; when the final strategy is an immediate fault repair strategy, the strategy generation engine combines the geographic information system and the resource scheduling system to calculate the optimal dispatch path for maintenance personnel and the spare parts delivery path, and automatically generates a repair work order instruction containing personnel, materials, paths, and emergency operation steps; all generated instructions are encapsulated as structured task objects and include execution priority and time constraint information.

[0011] Optionally, it also includes a closed-loop feedback and optimization step after the execution of the operation and maintenance strategy: After executing the differentiated operation and maintenance strategy instruction, continuously collect new online monitoring data and power supply reliability operation data of the device after a complete operation and maintenance cycle to obtain feedback data; based on the feedback data, recalculate the health evaluation result of the device and compare it with the health evaluation result before the strategy execution to obtain the change in health status; analyze the correlation between the change in health status and the type of operation and maintenance strategy executed to evaluate the effectiveness of the strategy; store the strategy effectiveness evaluation result, device identification, executed strategy type and corresponding change in health status as a training sample in the strategy effect sample library; periodically use the updated strategy effect sample library to jointly optimize and iterate the status evaluation model and the operation and maintenance strategy matrix to dynamically correct the accuracy of the health evaluation and the applicability of the operation and maintenance strategy matching.

[0012] Optionally, it also includes a global operation and maintenance strategy adjustment step that considers network-level collaboration: periodically executing the differentiated operation and maintenance method based on real-time weak links of equipment as described in claim 1 on all equipment in the distribution network in batches to generate a preliminary differentiated operation and maintenance strategy instruction set for all equipment; performing conflict detection on all preliminary differentiated operation and maintenance strategy instructions to identify instruction combinations that conflict in time, space, or resources; establishing a global optimization model with the goal of maximizing the improvement of power supply reliability of the entire network and minimizing the total operation and maintenance cost, and inputting the preliminary differentiated operation and maintenance strategy instruction set as a decision variable into the global optimization model; using a multi-objective evolutionary algorithm to solve the global optimization model, and under the premise of satisfying global resource constraints, coordinating the operation and maintenance strategy type and execution time window of some equipment to output a global collaborative operation and maintenance plan that eliminates conflicts and achieves Pareto optimality.

[0013] A differentiated operation and maintenance system based on real-time equipment weaknesses includes: a data acquisition and fusion module, an equipment health evaluation module, an equipment importance assessment module, an operation and maintenance strategy matching module, and an operation and maintenance strategy generation module. The data acquisition and fusion module is configured to collect online monitoring data and power supply reliability operation data of equipment in the distribution network in real time, process and fuse the data to obtain a multi-source equipment status dataset, and send the multi-source equipment status dataset to the equipment health evaluation module and the equipment importance assessment module. The equipment health evaluation module is configured to calculate the current health status score of the equipment based on the received multi-source equipment status dataset using a status evaluation model, obtain a health evaluation result, and send the health evaluation result to the operation and maintenance strategy matching module. The importance assessment module is configured to evaluate the impact of equipment failures on power supply reliability based on the received power supply reliability operation data, combined with the equipment's location and load importance in the network topology, using the analytic hierarchy process (AHP) to obtain an importance evaluation result, and then send the importance evaluation result to the operation and maintenance strategy matching module. The operation and maintenance strategy matching module is configured to input the received health evaluation result and the importance evaluation result into a preset operation and maintenance strategy matrix for matching, determine the corresponding operation and maintenance strategy type, and send the matching result to the operation and maintenance strategy generation module. The operation and maintenance strategy generation module is configured to generate differentiated operation and maintenance strategy instructions for the equipment based on the received matching result, and the differentiated operation and maintenance strategy instructions include specific operation and maintenance measures, execution time windows, and resource allocation schemes.

[0014] Optionally, the equipment health evaluation module includes: a feature vector construction unit, a deep learning anomaly detection unit, a dynamic failure rate prediction unit, and a comprehensive evaluation mapping unit; the feature vector construction unit is configured to select a subset of features directly related to the physical degradation mechanism of the equipment from the multi-source dataset of equipment status, construct a health evaluation feature vector, and send the health evaluation feature vector to the deep learning anomaly detection unit; the deep learning anomaly detection unit is configured to input the received health evaluation feature vector into a pre-trained encoder-decoder structure anomaly detection model, and obtain a representation of the equipment status by calculating the reconstruction error between the input features and the model reconstruction features. An anomaly score is assigned to the device if its condition deviates from the normal operating mode. This anomaly score is then sent to the dynamic failure rate prediction unit and the comprehensive evaluation mapping unit. The dynamic failure rate prediction unit is configured to calculate the probability of conditional failure of the device within a specific future time period based on the received anomaly score, combined with the device's years of operation and the average lifespan of similar devices, using the Weibull proportional hazards model. This probability of conditional failure is then sent to the comprehensive evaluation mapping unit. The comprehensive evaluation mapping unit is configured to fuse the received anomaly score and the probability of conditional failure, and then normalize and map them to a preset health score range using a fuzzy comprehensive evaluation method. Finally, a quantitative health evaluation result is output.

[0015] Compared with existing technologies, this invention has the following advantages: This application collects online monitoring data and power supply reliability operation data of distribution network equipment in real time, integrates the dynamic operating status and static background attributes of the equipment to form a multi-source dataset, quantifies the health evaluation results and importance evaluation results of the equipment based on the dataset, and generates differentiated operation and maintenance strategy instructions by matching them with a preset operation and maintenance strategy matrix. This effectively solves the problems of lack of equipment-level targeting and unreasonable resource allocation in existing technologies, and realizes the transformation of operation and maintenance work from regional macro scheduling to precise and differentiated handling of equipment. It not only ensures that key and vulnerable equipment receives timely and effective operation and maintenance support, but also avoids excessive operation and maintenance of ordinary equipment, and significantly improves the utilization efficiency of operation and maintenance resources and the power supply reliability of the distribution network. In the data processing stage, time-stamp alignment, data cleaning, sliding time window feature extraction, and static attribute association fusion significantly improved the completeness and effectiveness of the multi-source dataset of equipment status, providing a solid data foundation for subsequent health and importance assessments. During health assessment, a deep learning anomaly detection model was used to capture equipment status anomalies, combined with a Weibull proportional hazards model to predict dynamic failure rates, and fuzzy comprehensive evaluation was used to achieve quantitative scoring, significantly improving the accuracy and foresight of equipment health status assessments. For importance assessment, a multi-criteria hierarchical model was constructed using the analytic hierarchy process (AHP) to quantify the impact of equipment failures on load loss, power outages, and other dimensions, ensuring the scientific rigor and comprehensiveness of equipment importance level classification. The system incorporates resource constraint checks and dynamic adjustment mechanisms during operation and maintenance strategy matching, enabling flexible optimization of operation and maintenance strategies based on actual resource conditions and ensuring their feasibility and adaptability. It generates structured instructions containing specific measures, time windows, and resource allocations for different types of operation and maintenance strategies, making operation and maintenance execution more standardized and efficient. Through closed-loop feedback and optimization steps, it continuously iterates and optimizes the state evaluation model and operation and maintenance strategy matrix using operation and maintenance effect data, achieving dynamic evolution of the operation and maintenance plan. The global collaborative adjustment steps, through conflict detection and multi-objective optimization, eliminate operation and maintenance conflicts between devices, achieving a balance between maximizing power supply reliability and minimizing operation and maintenance costs at the network-wide level, further improving the overall coordination and economy of distribution network operation and maintenance. Detailed Implementation

[0016] The present invention will now be described in detail with reference to embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of this application can be combined with each other.

[0017] The following detailed descriptions are exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. Example 1

[0018] This embodiment details the implementation process of a differentiated operation and maintenance method based on real-time weak links of equipment. It covers the entire process, including data collection and fusion, health assessment, importance evaluation, operation and maintenance strategy matching, differentiated instruction generation, closed-loop feedback optimization, and global collaborative adjustment. All technical details are designed around the real-time status of equipment and the actual needs of power distribution network operation to ensure the operability and practicality of the method.

[0019] In the process of acquiring real-time online monitoring data and power supply reliability operation data of equipment in the distribution network to obtain multi-source datasets of equipment status, the data streams of various systems are first synchronously accessed through standardized data interfaces. The data interfaces support multiple communication protocols such as RESTAPI, MQTT, and OPCUA, and are adaptable to different data output formats of online equipment monitoring systems, distribution automation systems, and power supply service command systems. Specifically, the online equipment monitoring data includes electrical quantity data and operating condition data of various equipment such as transformers, circuit breakers, and lines. Electrical quantity data covers three-phase voltage, three-phase current, active power, reactive power, power factor, and harmonic content. Operating condition data includes equipment casing temperature, oil temperature, winding temperature (for transformers), number of operations, opening and closing status, and insulation status monitoring data. Power supply reliability operation data includes raw multi-source data such as power outage event records, outage duration, outage range, fault occurrence time and location, fault type, load rate change curves, and voltage sag / boost records.

[0020] After acquiring the raw multi-source data, time-stamp alignment and data cleaning are performed. Time-stamp alignment uses the unified timestamp of the power distribution automation system as a benchmark, employing linear interpolation to adjust the time nodes of data with different sampling frequencies, ensuring consistency across all data in the time dimension. For example, sampling data from the online equipment monitoring system (10 seconds / time) and interruption records from the power supply service command system (1 minute / time) are aligned to the same time granularity. During data cleaning, the 3σ principle and box plot method are used to identify and remove outliers. Data exceeding the physical range of normal equipment operation (e.g., voltage exceeding rated voltage ±10%) is directly marked as invalid and removed. Random outliers are corrected using a local weighted regression algorithm. Missing value imputation selects the appropriate method based on the data type. Short-term continuous missing data is imputed using linear interpolation or neighborhood mean, while long-term missing data or data lacking key parameters are imputed using an LSTM-based time series prediction model, ensuring data integrity.

[0021] After data cleaning, the resulting multi-source time series data is regularized, and statistical features are extracted using a sliding time window mechanism. The size of the sliding time window can be flexibly configured according to the equipment type and data characteristics. For example, a 5-minute window is used for line current data, and a 10-minute window is used for transformer oil temperature data. The window step size is set to half the window size to ensure the continuity of features. Statistical feature calculations include mean, variance, kurtosis, and trend slope. Supplementary features such as skewness, maximum, minimum, range, and energy value can also be extracted to form a primary feature set. Subsequently, the primary feature set is correlated and fused with the static attribute data of the equipment ledger. In addition to equipment model, years of operation, and family defect information, the static attribute data of the equipment ledger also includes equipment rated parameters (rated voltage, rated current, rated capacity), installation location, manufacturer information, maintenance records, and technical upgrade information. By using the unique identifier of the equipment, dynamic operating features are associated with static background attributes, forming a complete and multi-dimensional multi-source dataset of equipment status, providing comprehensive data support for subsequent evaluation.

[0022] To calculate the current health status score of equipment based on a multi-source equipment status dataset, the first step in obtaining the health evaluation result is to select a subset of features directly related to the physical degradation mechanism of the equipment from the multi-source equipment status dataset. Feature selection employs a combination of mutual information and random forest feature importance assessment to eliminate redundant and irrelevant features. For example, for transformers, features related to oil temperature and winding temperature, oil chromatography data features, insulation resistance features, and partial discharge features are selected to construct the health evaluation feature vector, ensuring that the feature vector accurately reflects the health status of the equipment, such as aging, wear, and potential faults.

[0023] The health assessment feature vector is input into a pre-trained deep learning anomaly detection model. This model employs an autoencoder (AE) or variational autoencoder (VAE) architecture. The encoder consists of 3-5 fully connected or convolutional layers, which progressively reduce the dimensionality of the input features to a low-dimensional latent space. The decoder uses a network structure symmetrical to the encoder to reconstruct the low-dimensional latent space vector into an output vector with the same dimension as the input features. The model training process aims to minimize the reconstruction error, using the Adam optimizer to iteratively update the parameters. The training data comes from historical normal operation data of power distribution network equipment, ensuring that the model can learn the feature representation of the equipment's normal operation mode. The mean squared error (MSE) or mean absolute error (MAE) between the input features and the reconstructed features is calculated as the reconstruction error. This error value is the anomaly score characterizing the deviation of the equipment's state from the normal operation mode; a higher anomaly score indicates a greater deviation of the equipment's state from normal.

[0024] The anomaly score is input into a dynamic failure rate prediction module based on the Weibull proportional hazards model. This module first uses maximum likelihood estimation to fit the shape and scale parameters of the Weibull distribution based on historical failure data and operational data of similar equipment, thus determining the basic parameters of the model. Combining the current equipment's years of operation, anomaly score, and average lifespan data of similar equipment, the module calculates the conditional failure probability of the equipment within a specific future time period (e.g., 1 month, 3 months, 6 months) using the cumulative distribution function of the Weibull proportional hazards model. This probability reflects the failure risk trend of the equipment in its current health state.

[0025] The health assessment results are obtained by integrating anomaly scores and conditional failure probabilities, and a fuzzy comprehensive evaluation method is used for normalization mapping. First, a fuzzy comprehensive evaluation system is established, with the factor set consisting of two evaluation indicators: anomaly scores and conditional failure probabilities. The weight set is determined using the analytic hierarchy process (AHP), assigning corresponding weights based on the degree of influence of the two indicators on the equipment's health status. The evaluation criteria are set into five levels: "Excellent," "Good," "Average," "Poor," and "Severe," corresponding to a preset health score range (0-100 points). The membership degree of each evaluation level is calculated using a fuzzy relation matrix, and a weighted average method is used to obtain the comprehensive membership degree. Finally, the comprehensive membership degree is normalized and mapped to the corresponding health score range, outputting a quantitative health assessment result, thus achieving a precise quantitative representation of the equipment's health status.

[0026] Based on power supply reliability operation data, and considering the location and load importance of equipment in the network topology, the analytic hierarchy process (AHP) is used to assess the impact of equipment failures on power supply reliability. To obtain the importance evaluation results, a hierarchical model is first constructed with "impact on power supply reliability" as the top-level objective. This model includes an objective layer, a criterion layer, and a solution layer. The criterion layer comprises four indicators: the magnitude of load loss caused by equipment failure, the impact on the number of users affected by power outages, network topology reconfiguration capability, and the average time required to restore power. The solution layer includes all equipment in the distribution network to be evaluated.

[0027] The quantification process of the criteria-level indicators is as follows: Load loss magnitude is determined through power flow calculation and single-component fault simulation. Power flow calculation uses the Newton-Raphson method or PQ decomposition method, calculating the power flow distribution under normal equipment operation based on the current distribution network topology and load data. Single-component fault simulation recalculates the power flow distribution by artificially setting the equipment to an outage state, obtaining the load interruption caused by the fault. The load loss magnitude is the proportion of the interrupted load to the total load within the equipment's power supply range. The impact on the number of users during power outages is calculated by statistically analyzing the number of users experiencing power outages due to equipment failure, combined with the estimated outage duration: Number of users experiencing power outages = Number of users experiencing power outages × Estimated outage duration. Network topology reconfiguration capability is quantified by analyzing the number of load transfer paths, transfer load capacity, and number of switches required for transfer after equipment failure. The more transfer paths, the larger the transfer capacity, and the fewer the number of switch operations, the stronger the network topology reconfiguration capability, and the higher the corresponding quantification index value. The average time required to restore power is estimated based on a comprehensive consideration of factors such as historical repair data of similar equipment failures, equipment failure type, availability of spare parts, and arrival time of repair personnel.

[0028] Each indicator at the criterion level is compared pairwise, and a judgment matrix is ​​constructed using a 1-9 scale. Scale 1 indicates that two indicators are equally important, scale 3 indicates that the former is slightly more important than the latter, scale 5 indicates that the former is significantly more important than the latter, scale 7 indicates that the former is strongly more important than the latter, and scale 9 indicates that the former is extremely more important than the latter. Scales 2, 4, 6, and 8 are the median values ​​of the above adjacent scales. The elements of the judgment matrix satisfy reciprocity. After construction, a consistency check is performed, calculating the consistency index CI and the random consistency ratio CR. When CR < 0.1, the judgment matrix meets the consistency requirements; otherwise, the elements of the judgment matrix need to be adjusted until consistency is satisfied. The weight coefficients of each criterion level indicator are calculated using the eigenvalue method to ensure that the weight allocation reflects the relative importance of each indicator to the power supply reliability.

[0029] The first set of quantitative indicators (load loss magnitude, impact on the number of users during power outages) and the second set of quantitative indicators (network topology reconfiguration capability, average time required to restore power) are standardized. A min-max normalization method is used to map all indicator values ​​to the [0,1] interval, eliminating the influence of different indicator dimensions and numerical ranges. The standardized indicator values ​​are then linearly weighted and summed based on the calculated weight coefficients to obtain a comprehensive importance score for each device. Importance levels are then assigned based on the scores, resulting in an importance evaluation. Importance levels are categorized into four levels: critical, important, moderate, and minor, each corresponding to a different scoring range, facilitating subsequent matching of operation and maintenance strategies.

[0030] When the health evaluation results and importance evaluation results are input into the preset operation and maintenance strategy matrix for matching, the operation and maintenance strategy matrix is ​​a two-dimensional decision table. Its horizontal axis is divided into five health level intervals according to the numerical range of the health evaluation results, namely Excellent (90-100 points), Good (80-89 points), Average (60-79 points), Poor (40-59 points), and Serious (0-39 points); the vertical axis is divided into four importance level intervals according to the numerical range of the importance evaluation results, namely Critical (80-100 points), Important (60-79 points), Ordinary (40-59 points), and Minor (0-39 points). Each cell in the two-dimensional decision table predefines a corresponding first maintenance strategy type. For example, the first maintenance strategy type corresponding to critical importance + severe health is immediate fault repair; the first maintenance strategy type corresponding to critical importance + poor health is emergency planned maintenance; the first maintenance strategy type corresponding to important importance + average health is routine planned maintenance; the first maintenance strategy type corresponding to ordinary importance + good health is enhanced status monitoring; and the first maintenance strategy type corresponding to minor importance + excellent health is routine monitoring.

[0031] Once the first maintenance strategy type is matched, a resource constraint check process is triggered to comprehensively verify whether the currently available human resources, material resources, and time window resources meet the execution requirements of the first maintenance strategy type. Human resource verification includes the number of personnel with the corresponding equipment maintenance skills, their current work status (whether they are en route or idle), and the distance between the personnel's location and the equipment location. Material resource verification includes the inventory quantity of spare parts required by the first maintenance strategy, the storage location of spare parts, the quality status of spare parts, transportation accessibility, and the availability of required maintenance tools and instruments. Time window resource verification includes periods of low power grid load, user-permitted power outage periods, and periods of severe weather avoidance, ensuring that maintenance operations will not have an excessive impact on power supply reliability.

[0032] If resources meet the execution requirements of the first maintenance strategy type, then that first maintenance strategy type is directly confirmed as the final strategy. If resources do not meet the requirements, such as insufficient spare parts inventory, all maintenance personnel being busy, or no suitable time window, then the dynamic strategy adjustment mechanism is activated. The dynamic strategy adjustment mechanism follows the principle of "minimizing impact and maximizing resource adaptation." It searches for adjacent cells in the maintenance strategy matrix, including horizontally adjacent cells (with health levels fluctuating by one level) and vertically adjacent cells (with importance levels fluctuating by one level). It prioritizes selecting the second maintenance strategy type corresponding to the cell with the least impact on power supply reliability and whose resource requirements are closest to the currently available resources as the alternative strategy. For example, if the first maintenance strategy corresponding to critical importance + severe health is immediate fault repair, and there are currently insufficient repair personnel, it can be adjusted to emergency planned repair corresponding to critical importance + poor health as the alternative strategy, and the execution time window is updated synchronously to adapt to the resource situation.

[0033] When generating differentiated operation and maintenance strategy instructions for the device based on the matching results, the engine generates corresponding structured instructions according to the type of the final strategy, ensuring that the instructions include specific operation and maintenance measures, execution time windows, and resource allocation schemes. When the final strategy is a planned maintenance strategy, the strategy generation engine calls a preset maintenance knowledge base, which stores information such as standard maintenance procedures, technical specifications, and safety requirements for different device types and health states. Based on the equipment type (such as transformer, circuit breaker, line) and the specific values ​​of the health evaluation results, corresponding standard maintenance procedures are matched from the standard maintenance operation library. For example, when the transformer health is poor, the matched maintenance procedures include oil chromatography analysis, winding insulation resistance testing, core grounding current testing, tap changer inspection, etc. At the same time, based on the maintenance procedures and equipment model, a list of required spare parts is generated, which clearly specifies the spare parts name, specifications, quantity, quality requirements, etc. Combining historical maintenance data and process complexity, the estimated working hours for each maintenance procedure are calculated, and the total estimated working hours are summarized. Finally, a planned maintenance instruction is generated, which includes specific work order content (equipment identification, maintenance content, execution steps, safety precautions), execution time window (prioritizing off-peak periods), and resource allocation plan (maintenance personnel list, skill requirements, spare parts allocation, tool and instrument configuration).

[0034] When the final strategy is a condition monitoring enhancement strategy, the strategy generation engine adjusts the parameters of the online monitoring devices associated with the equipment. The sampling frequency is adjusted based on the current health level; for example, the sampling frequency for devices with average health is increased from 1 time / minute to 5 times / minute, and for devices with poor health, the sampling frequency is increased to 10 times / minute, ensuring timely capture of equipment status changes. Targeted diagnostic algorithm models are enabled; these models can be random forests, XGBoost, LightGBM, etc., trained based on historical fault and normal operation data, enabling real-time analysis of abnormal features in the monitoring data. Stricter warning thresholds are set, determined based on statistical analysis of historical normal operation data; for example, the 95th percentile is used as the standard warning threshold, but adjusted to the 90th percentile during condition monitoring enhancement to improve the sensitivity of anomaly warnings. Finally, a monitoring parameter adjustment instruction is generated, including the sampling frequency adjustment value, the diagnostic algorithm model identifier, and the warning threshold parameters, ensuring the monitoring device can accurately capture potential equipment fault signals.

[0035] When the final strategy is immediate fault repair, the strategy generation engine combines the Geographic Information System (GIS) and the resource scheduling system to generate emergency repair work orders. The GIS accurately locates the physical position of the equipment and displays information such as the road distribution, traffic conditions, and emergency repair site locations around the equipment. The resource scheduling system obtains real-time information on currently available maintenance personnel (including skill level, location, and tools they are carrying) and spare parts inventory information (including spare parts location and available quantity). It uses a genetic algorithm or simulated annealing algorithm to calculate the optimal dispatch route for maintenance personnel and the spare parts delivery route. The route planning comprehensively considers factors such as distance, traffic congestion, and delivery time to ensure that maintenance personnel and spare parts can reach the site as quickly as possible. The emergency repair work order includes equipment fault information (fault type, fault location, fault impact range), maintenance personnel information (name, contact information, skill level), material allocation information (spare parts list, tool and instrument list), optimal route information, and emergency operation steps (power outage operation sequence, fault isolation steps, emergency repair construction steps, and power restoration steps). All generated instructions are encapsulated as structured task objects, stored and transmitted in JSON format, and include execution priorities (e.g., immediate fault repair has the highest priority, and planned repair has the medium priority) and time constraints (e.g., must respond within 2 hours, repair must be completed within 4 hours) to facilitate rapid execution by maintenance personnel.

[0036] After the operation and maintenance strategy is implemented, a closed-loop feedback and optimization process is initiated to ensure continuous iteration and optimization of the operation and maintenance methods. Following the execution of differentiated operation and maintenance strategy instructions, new online monitoring data and power supply reliability operation data of the equipment are continuously collected after a complete operation and maintenance cycle. The complete operation and maintenance cycle is determined based on the equipment type and operation and maintenance strategy type. For example, the complete operation and maintenance cycle for immediate fault repair is 3 months, for planned maintenance it is 6 months, and for enhanced condition monitoring it is 1 month. The collected feedback data is consistent with the scope of previous data collection to ensure data comparability.

[0037] Based on the feedback data, the health evaluation result of the device is recalculated following the same health evaluation process as before. The new health evaluation result is compared with the health evaluation result before the strategy implementation to obtain the change in health status. Change in health status = Health score after strategy implementation - Health score before strategy implementation. The correlation between the change in health status and the type of operation and maintenance strategy implemented is analyzed to evaluate the effectiveness of the strategy. If the change in health status is positive and reaches a preset threshold (e.g., an increase of more than 10 points), the strategy is considered effective; if the change in health status is positive but does not reach the preset threshold, the strategy is considered partially effective; if the change in health status is zero or negative, the strategy is considered invalid.

[0038] Information such as strategy effectiveness assessment results, equipment identification, executed strategy type, corresponding health status changes, execution time, and resource consumption are used as training samples and stored in the strategy effectiveness sample library. The strategy effectiveness sample library employs a combined time-series and relational database storage method. The time-series database stores the time-series sequence of feedback data, while the relational database stores the structured information of the samples, facilitating subsequent queries and retrieval. The updated strategy effectiveness sample library is used periodically to jointly optimize and iterate the state evaluation model and the operation and maintenance strategy matrix. The optimization cycle can be set to once a month. During the optimization process, the gradient descent algorithm is used to adjust the parameters of the state evaluation model to improve the accuracy of health assessment. Based on the effectiveness data of different strategy types for different health statuses and equipment of different importance in the sample library, the strategy type corresponding to each cell in the operation and maintenance strategy matrix is ​​adjusted, or the division interval between health level and importance level is optimized, dynamically correcting the applicability of operation and maintenance strategy matching to ensure that the operation and maintenance methods can adapt to changes in equipment status and the distribution network operating environment.

[0039] To achieve optimized operation and maintenance across the entire network, a global operation and maintenance strategy adjustment step considering network-level collaboration is also included. The aforementioned differentiated operation and maintenance methods are periodically executed in batches on all equipment in the distribution network. The batch execution cycle can be set to once a week, generating a preliminary set of differentiated operation and maintenance strategy instructions for all equipment. This set includes information such as the operation and maintenance strategy type, execution time window, and resource requirements for each equipment. Conflict detection is performed on all preliminary differentiated operation and maintenance strategy instructions to identify instruction combinations that conflict in time, space, or resources. Time conflicts refer to operation and maintenance strategy instructions for multiple devices requiring execution within the same time period, which cannot be avoided by time adjustment; space conflicts refer to operation and maintenance strategy instructions for multiple devices being executed in the same area, potentially leading to traffic congestion, on-site work interference, and other problems; resource conflicts refer to operation and maintenance strategy instructions for multiple devices requiring a total of human and material resources exceeding the current available resources, such as multiple instructions requiring the same batch of maintenance personnel at the same time, or the total amount of spare parts required exceeding the inventory.

[0040] A global optimization model is established with the objectives of maximizing the overall power supply reliability improvement and minimizing the total operation and maintenance cost. The objective function of the global optimization model is max(overall power supply reliability improvement rate) - λ×(total operation and maintenance cost), where λ is a weighting coefficient used to balance the relationship between power supply reliability improvement and operation and maintenance cost control, and can be adjusted according to the needs of distribution network operation and management. The constraints of the model include human resource constraints (number of maintenance personnel, working hours limits), material resource constraints (spare parts inventory limits, transportation capacity limits), time window constraints (allowed operation and maintenance time limits for each device), and network security constraints (distribution network power flow must not exceed safety limits during operation and maintenance operations, and large-scale power outages must not be caused). The initial differentiated operation and maintenance strategy instruction set is used as the decision variable input into the global optimization model.

[0041] A multi-objective evolutionary algorithm is used to solve the global optimization model. NSGA-II or MOEA / D can be selected as the multi-objective evolutionary algorithm. Through operations such as population initialization, fast non-dominated sorting, congestion calculation, selection, crossover, and mutation, the algorithm searches for a Pareto optimal solution set while satisfying all constraints. Each solution in the Pareto optimal solution set corresponds to a set of adjusted operation and maintenance strategy instructions, achieving an optimal balance between improving the overall power supply reliability and the total operation and maintenance cost. Based on the actual operational needs and decision-making preferences of the distribution network, the final optimization scheme is selected from the Pareto optimal solution set. The operation and maintenance strategy types and execution time windows of some equipment are adjusted collaboratively. For example, the planned maintenance strategy of some non-critical equipment is adjusted to a condition monitoring enhancement strategy, or the execution time window of some equipment is adjusted from the early stage of load trough to the later stage, avoiding resource and time conflicts. Finally, a globally coordinated operation and maintenance plan that eliminates conflicts and achieves Pareto optimality is output to guide the orderly implementation of operation and maintenance work across the entire network. Example 2

[0042] The differentiated operation and maintenance system based on real-time weak points of equipment in this embodiment adopts a layered architecture design, including a hardware layer, a data layer, a service layer, and an application layer. Data interaction and function calls are realized between each layer through standardized interfaces. The system as a whole has the characteristics of high reliability, high scalability, and easy maintenance, and can accurately support the entire process execution of the above-mentioned differentiated operation and maintenance methods.

[0043] The system includes a data acquisition and fusion module, an equipment health assessment module, an equipment importance assessment module, an operation and maintenance strategy matching module, and an operation and maintenance strategy generation module. Each module is independent yet works collaboratively, and data sharing and command transmission are achieved through message queues and databases to ensure the efficiency and stability of the system operation.

[0044] The hardware component of the data acquisition and fusion module includes a data acquisition terminal, an edge gateway, and a communication module. The data acquisition terminal is deployed on various devices in the power distribution network, including voltage sensors, current sensors, temperature sensors, humidity sensors, and partial discharge sensors, enabling real-time acquisition of online monitoring data. The edge gateway is deployed in the power distribution network substation or distribution room, responsible for aggregating data from surrounding data acquisition terminals and performing preliminary preprocessing. The communication module supports multiple communication methods such as 5G, 4G, WiFi, LoRa, and fiber optics, ensuring the stability and real-time performance of data transmission. The software component includes a data access interface, a data processing engine, and a data fusion unit. The data access interface supports multiple protocols such as RESTAPI, MQTT, and OPCUA, enabling seamless integration with online equipment monitoring systems, power distribution automation systems, and power supply service command systems to acquire raw, multi-source data in real time.

[0045] The data processing engine is responsible for performing operations such as time-stamp alignment, data cleaning, missing value imputation, and outlier removal on the raw multi-source data. The time-stamp alignment module uses a time synchronization protocol (such as NTP) to ensure that the timestamps of all data are consistent. The data cleaning module integrates algorithms such as the 3σ principle, box plot method, and locally weighted regression. The data imputation module supports multiple methods such as linear interpolation, mean imputation, and LSTM prediction imputation, and can automatically select the optimal imputation strategy according to the data type. The data fusion unit uses a sliding time window mechanism to extract statistical features. The window size and step size can be flexibly adjusted through the system configuration interface. After feature extraction, it is associated and fused with the static attribute data of the equipment ledger. The equipment ledger data is stored in a relational database. The dynamic data and static data are accurately associated through the unique equipment identifier. Finally, a multi-source dataset of equipment status is generated and sent to the equipment health evaluation module and the equipment importance assessment module. Data transmission uses an encryption protocol to ensure data security.

[0046] The equipment health assessment module is deployed on a server cluster, supporting distributed computing to improve assessment efficiency. This module includes a feature vector construction unit, a deep learning anomaly detection unit, a dynamic failure rate prediction unit, and a comprehensive evaluation mapping unit. Each unit achieves data flow through internal interfaces. The feature vector construction unit receives multi-source datasets of equipment status from the data acquisition and fusion module. It employs a feature selection algorithm combining mutual information and random forest feature importance assessment to filter a subset of features directly related to the equipment's physical degradation mechanism, eliminating redundant and irrelevant features to construct the health assessment feature vector. The feature selection parameters can be configured through the system backend to ensure the relevance and effectiveness of the feature vectors.

[0047] The deep learning anomaly detection unit incorporates a pre-trained encoder-decoder anomaly detection model. The model supports both autoencoder (AE) and variational autoencoder (VAE) architectures, allowing users to select the appropriate model based on device type and data characteristics. The encoder consists of 3-5 fully connected or convolutional layers, with configurable neurons per layer. It maps the input health evaluation feature vector to a low-dimensional latent space through progressive dimensionality reduction. The decoder employs a network structure symmetrical to the encoder, reconstructing the low-dimensional latent space vector into an output vector consistent with the input feature dimension. The model training module supports offline training and online updates. Offline training is based on historical normal operation data, while online updates are implemented using incremental learning algorithms to ensure the model adapts to changes in device status. This unit calculates the mean squared error (MSE) or mean absolute error (MAE) between the input features and the model-reconstructed features to obtain an anomaly score representing a deviation from normal operation. This anomaly score is then sent to the dynamic failure rate prediction unit and the comprehensive evaluation mapping unit.

[0048] The dynamic failure rate prediction unit incorporates a Weibull proportional hazards model. This unit receives data from the feature vector construction unit regarding the equipment's operational years, the average lifespan of similar equipment, and anomaly scores from the deep learning anomaly detection unit. Based on historical failure data, it fits the shape and scale parameters of the Weibull distribution using maximum likelihood estimation, and substitutes these parameters into the cumulative distribution function of the Weibull proportional hazards model to calculate the probability of conditional failure of the equipment within a specific future time period. This unit supports real-time uploading of failure data and dynamic adjustment of model parameters to ensure the accuracy of the conditional failure probability calculation. After calculation, the conditional failure probability is sent to the comprehensive evaluation mapping unit.

[0049] The comprehensive evaluation mapping unit incorporates a fuzzy comprehensive evaluation system, which includes a factor set definition module, a weight set calculation module, a comment set setting module, a membership degree calculation module, and a normalization module. The factor set definition module uses anomaly scores and conditional failure probabilities as evaluation factors; the weight set calculation module uses the analytic hierarchy process (AHP) to calculate the weight coefficients of the two factors; the comment set setting module divides health into five levels: excellent, good, average, poor, and severe; the membership degree calculation module uses a triangular fuzzy distribution or a trapezoidal fuzzy distribution to calculate the membership degree of each evaluation factor to different comment levels; and the normalization module uses the min-max normalization method to map the comprehensive membership degree to a preset health score range, ultimately outputting a quantified health evaluation result, which is then sent to the operation and maintenance strategy matching module.

[0050] The equipment importance assessment module includes a hierarchical structure construction unit, an indicator quantification unit, a weight calculation unit, and an importance scoring unit. These units work together to achieve accurate assessment of equipment importance. The hierarchical structure construction unit receives power supply reliability operation data from the data acquisition and fusion module. Combining the equipment's location and load importance in the network topology, it constructs a hierarchical model with "power supply reliability impact" as the top-level objective, load loss magnitude, impact on the number of users during power outages, network topology reconstruction capability, and average time required to restore power as the criterion layers, and all equipment to be assessed as the solution layer. The model structure can be visualized through the system interface and supports manual adjustment and confirmation.

[0051] The index quantification unit employs a power flow calculation module, a fault simulation module, and a topology analysis module to quantify the criteria-level indicators. The power flow calculation module incorporates two algorithms: the Newton-Raphson method and the PQ decomposition method. It can select the appropriate algorithm based on the distribution network scale, calculating the power flow distribution during normal equipment operation based on the current network topology and load data. The fault simulation module simulates power flow changes after a single component failure by setting equipment outage states, calculating the magnitude of load loss. The topology analysis module analyzes the transfer paths, transfer capacity, and number of switching operations after equipment failure, quantifying the network topology reconfiguration capability. Simultaneously, by combining historical emergency repair data and user information, it quantifies the impact of power outages on the number of users and the average time required to restore power, obtaining the first and second sets of quantified indicators.

[0052] The weight calculation unit uses the analytic hierarchy process (AHP) to calculate the weights of the criteria-level indicators. It incorporates a 1-9 scale to construct the judgment matrix and allows users to input pairwise comparison results via the system interface, automatically generating the judgment matrix. The consistency verification module calculates the consistency index (CI) and the random consistency ratio (CR). When CR ≥ 0.1, it prompts the user to adjust the elements of the judgment matrix until the consistency requirements are met. The weight solution module calculates the weight coefficients of each criterion-level indicator using the eigenvalue method. These weight coefficients are visualized and can be exported. The importance scoring unit standardizes the first and second sets of quantitative indicators, uses the min-max normalization method to eliminate the influence of dimensions, and performs a linear weighted sum based on the weight coefficients to obtain a comprehensive importance score for each device. After classifying the importance levels, it generates importance evaluation results and sends them to the operation and maintenance strategy matching module.

[0053] The operation and maintenance strategy matching module includes a matrix configuration unit, a matching execution unit, a resource verification unit, and a strategy adjustment unit. The module has a built-in operation and maintenance strategy matrix database that stores preset two-dimensional decision tables. The matrix configuration unit allows users to configure health level ranges, importance level ranges, and the corresponding operation and maintenance strategy types for each cell through the system interface. Configuration results are synchronized to the operation and maintenance strategy matrix database in real time. It supports adding, modifying, and deleting strategy types to meet the operation and maintenance needs of different power distribution networks.

[0054] The matching execution unit receives the health evaluation results from the equipment health evaluation module and the importance evaluation results from the equipment importance assessment module. Based on the classification standards for health and importance level ranges, it determines the current equipment's level range, locates the corresponding cell in the maintenance strategy matrix database, and obtains the predefined first maintenance strategy type. The resource verification unit interfaces with the human resource management system, material management system, and time window management system to obtain real-time information on available human resources (number of maintenance personnel, skill level, work status), material resources (spare parts inventory, tool availability), and time window resources (off-peak periods, permitted power outage periods). It verifies whether these resources meet the execution requirements of the first maintenance strategy type and generates a resource verification report.

[0055] The strategy adjustment unit optimizes the strategy based on the resource verification report results. If the resources meet the requirements, the first operation and maintenance strategy type is confirmed as the final strategy. If the resources do not meet the requirements, the dynamic adjustment mechanism is activated. The second operation and maintenance strategy type corresponding to the adjacent cell is searched in the operation and maintenance strategy matrix. An alternative strategy is selected based on the principle of "minimizing impact and maximizing resource adaptation". The final strategy matching result is generated and sent to the operation and maintenance strategy generation module.

[0056] The operation and maintenance strategy generation module includes an instruction generation unit, a knowledge base management unit, a geographic information integration unit, and a resource scheduling unit. The module has a built-in maintenance knowledge base and a strategy instruction template library, supporting rapid instruction generation and standardized output. The knowledge base management unit stores standard maintenance procedures, spare parts lists, technical specifications, safety requirements, and other information corresponding to different equipment types and operation and maintenance strategy types. It supports offline import, online editing, and automatic updates of the knowledge base, ensuring the accuracy and timeliness of the knowledge.

[0057] The instruction generation unit receives the strategy matching results sent by the operation and maintenance strategy matching module. Based on the final strategy type, it calls the corresponding strategy instruction template and, combined with information such as equipment identification, health evaluation results, and importance evaluation results, generates differentiated operation and maintenance strategy instructions. When the final strategy is a planned maintenance strategy, it calls the standard maintenance procedures and spare parts list from the maintenance knowledge base, calculates the estimated working hours based on historical maintenance data, and generates a planned maintenance instruction containing specific work order content, execution time window, and resource allocation scheme. When the final strategy is a condition monitoring enhancement strategy, it generates a monitoring parameter adjustment instruction containing sampling frequency adjustment values, diagnostic algorithm model identifiers, and early warning threshold parameters. When the final strategy is an immediate fault repair strategy, it integrates the functions of the geographic information system and resource scheduling system to generate an emergency repair work order instruction containing fault information, personnel information, material information, optimal path, and operation steps.

[0058] The geographic information integration unit incorporates a power distribution network geographic information system, accurately displaying information such as the physical location of equipment, surrounding road distribution, emergency repair site locations, and power grid topology. It supports map zooming, panning, and querying, providing geographic data support for route planning. The resource scheduling unit uses genetic algorithms or simulated annealing algorithms, combined with geographic information data and real-time resource status data, to calculate the optimal dispatch routes for maintenance personnel and spare parts delivery routes, ensuring that maintenance resources can quickly reach the site. All generated instructions are encapsulated as structured task objects, stored and transmitted in JSON format, and include execution priority and time constraints. Instructions can be sent to maintenance personnel through system interface display, SMS notifications, and mobile app push notifications, facilitating rapid reception and execution.

[0059] The system also includes a closed-loop feedback optimization module and a global collaborative adjustment module to achieve continuous optimization of operation and maintenance methods and collaborative management and control of network-wide operation and maintenance. The closed-loop feedback optimization module includes a data acquisition unit, an effect evaluation unit, a sample library management unit, and a model optimization unit. After the execution of operation and maintenance strategy instructions, the data acquisition unit continuously collects new online monitoring data and power supply reliability operation data of the equipment after a complete operation and maintenance cycle, as feedback data. The effect evaluation unit recalculates the equipment health evaluation results based on the feedback data, compares them with the results before execution to obtain the change in health status, and evaluates the effectiveness of the strategy. The sample library management unit stores information such as the strategy effectiveness evaluation results, equipment identification, strategy type, and change in health status in the strategy effect sample library. The model optimization unit periodically uses sample library data to jointly optimize and iterate the status evaluation model of the equipment health evaluation module and the operation and maintenance strategy matrix of the operation and maintenance strategy matching module, dynamically correcting model parameters and matrix configurations to improve the system's evaluation accuracy and strategy matching applicability.

[0060] The global collaborative adjustment module includes a batch processing unit, a conflict detection unit, a global optimization unit, and a plan generation unit. The batch processing unit periodically executes differentiated operation and maintenance methods on all equipment in the distribution network in batches, generating a preliminary set of differentiated operation and maintenance strategy instructions. The conflict detection unit performs time, space, and resource conflict detection on the instruction set and identifies conflicting instruction combinations. The global optimization unit establishes a global optimization model with the goal of maximizing the overall power supply reliability and minimizing the total operation and maintenance cost, and uses a multi-objective evolutionary algorithm to solve the model to obtain a Pareto optimal solution set. The plan generation unit selects the optimal solution based on decision preferences, coordinates the operation and maintenance strategy types and execution time windows of some equipment, and generates a global collaborative operation and maintenance plan that eliminates conflicts and achieves Pareto optimality, guiding the orderly development of operation and maintenance work across the entire network.

[0061] The system's hardware deployment supports three modes: local deployment, cloud deployment, and hybrid deployment. Local deployment is suitable for scenarios with high data security requirements and limited network conditions. Cloud deployment is suitable for scenarios with large-scale distribution networks and dispersed operation and maintenance nodes. Hybrid deployment combines the advantages of both, enabling local storage of core data and cloud expansion of non-core functions. The system's software interface includes a data visualization panel, an operation and maintenance strategy management interface, an equipment evaluation result display interface, a command issuance interface, and a system configuration interface. It supports users in performing operations such as data query, strategy configuration, command management, and system maintenance. The interface design is simple and intuitive, and the operation is convenient.

[0062] The system also includes auxiliary functions such as a permission management module, a log management module, and an alarm notification module. The permission management module uses role-based access control policies to assign different operating permissions to different users (such as system administrators, operation and maintenance managers, and maintenance personnel) to ensure system security. The log management module records system operation logs, operation logs, and fault logs, supporting log querying, exporting, and analysis to facilitate system fault diagnosis and operation and maintenance auditing. The alarm notification module automatically triggers alarm notifications in cases of severely abnormal equipment health, critical equipment with poor health in importance assessments, or abnormal execution of operation and maintenance strategies. These notifications are sent to relevant personnel via SMS, email, and app push notifications to ensure timely handling and the safe and stable operation of the power distribution network. As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the above-disclosed embodiments are merely illustrative and not exhaustive. All modifications within the scope of this invention or equivalent to this invention are included in this invention.

Claims

1. A differentiated operation and maintenance method based on real-time weak points of equipment, characterized in that, Includes the following steps: Real-time acquisition of online monitoring data and power supply reliability operation data of equipment in the power distribution network to obtain a multi-source dataset of equipment status; Based on the multi-source dataset of device status, a status evaluation model is used to calculate the current health status score of the device and obtain the health evaluation result. Based on the power supply reliability operation data, combined with the device's location and load importance in the network topology, the analytic hierarchy process (AHP) is used to assess the impact of device failures on power supply reliability, resulting in an importance evaluation result. The health evaluation result and the importance evaluation result are then input into a preset operation and maintenance strategy matrix for matching. The operation and maintenance strategy matrix defines the operation and maintenance strategy types corresponding to different combinations of health and importance levels. Based on the matching result, differentiated operation and maintenance strategy instructions are generated for the device. These instructions include specific operation and maintenance measures, execution time windows, and resource allocation schemes.

2. The differentiated operation and maintenance method based on real-time weak links of equipment according to claim 1, characterized in that, The steps of acquiring real-time online monitoring data and power supply reliability operation data of equipment in the distribution network to obtain a multi-source dataset of equipment status specifically include: synchronously accessing the data streams of the equipment online monitoring system, the distribution automation system, and the power supply service command system through data interfaces to obtain raw multi-source data including equipment electrical quantity data, equipment operating condition data, and power supply interruption event records; performing time-scale alignment and data cleaning on the raw multi-source data, removing outliers and filling missing values ​​to obtain normalized multi-source time series data; for the normalized multi-source time series data, using a sliding time window mechanism to extract statistical features within each sliding time window, the statistical features including mean, variance, kurtosis, and trend slope to obtain a primary feature set; and associating and fusing the primary feature set with static attribute data from the equipment ledger, the static attribute data including equipment model, commissioning year, and family defect information to form the multi-source dataset of equipment status that integrates dynamic operating status and static background attributes.

3. The differentiated operation and maintenance method based on real-time weak links of equipment according to claim 1, characterized in that, The step of calculating the current health status score of the equipment using a state evaluation model to obtain the health evaluation result specifically includes: selecting a subset of features directly related to the physical degradation mechanism of the equipment from the multi-source dataset of equipment status to form a health evaluation feature vector; inputting the health evaluation feature vector into a pre-trained deep learning anomaly detection model, where the deep learning anomaly detection model maps the input features to a low-dimensional latent space through an encoder and reconstructs them through a decoder, calculating the reconstruction error between the input features and the reconstructed features to obtain an anomaly score representing the deviation of the equipment status from the normal operating mode; inputting the anomaly score into a dynamic failure rate prediction module based on the Weibull proportional hazards model, and calculating the probability of conditional failure of the equipment in a specific future time period by combining the equipment's years of operation with the average lifespan of similar equipment; fusing the anomaly score and the probability of conditional failure, and normalizing and mapping them to a preset health score range using a fuzzy comprehensive evaluation method, and finally outputting the quantified health evaluation result.

4. The differentiated operation and maintenance method based on real-time weak links of equipment according to claim 1, characterized in that, The steps for evaluating the impact of equipment failures on power supply reliability using the analytic hierarchy process (AHP) to obtain importance evaluation results specifically include: constructing a hierarchical model with "impact on power supply reliability" as the top-level objective. The criteria layer of the hierarchical model includes the magnitude of load loss caused by equipment failure, the impact on the number of users during power outages, network topology reconfiguration capabilities, and the average time required to restore power; based on the location of the equipment in the network topology, quantifying the load loss caused by equipment failure and the number of users during power outages through power flow calculations and single-component failure simulations to obtain a first set of quantitative indicators; analyzing the network topology to calculate the equipment's ability to be supplied with load through other paths after a failure and the number of switches required to operate, to obtain a second set of quantitative indicators; comparing each indicator in the criteria layer of the hierarchical model pairwise to construct a judgment matrix, and calculating the weight coefficients of each indicator after passing a consistency check; standardizing the first and second sets of quantitative indicators, and performing a linear weighted sum based on the weight coefficients to calculate the comprehensive importance score for each device, thus obtaining the importance evaluation result.

5. The differentiated operation and maintenance method based on real-time weak links of equipment according to claim 1, characterized in that, The step of inputting the health evaluation results and importance evaluation results into a preset operation and maintenance strategy matrix for matching specifically includes: the operation and maintenance strategy matrix is ​​a two-dimensional decision table, whose horizontal axis is divided into multiple health level intervals according to the numerical range of the health evaluation results, and whose vertical axis is divided into multiple importance level intervals according to the numerical range of the importance evaluation results; according to the health level interval to which the current device's health evaluation result belongs, and the importance level interval to which the importance evaluation result belongs, the corresponding cell is located in the two-dimensional decision table, and the first operation and maintenance strategy type is predefined in this cell; after matching the first operation and maintenance strategy type, a resource constraint check process is triggered to check whether the currently available human resources, material resources, and time window resources meet the execution requirements of the first operation and maintenance strategy type; if the resources meet the requirements, the first operation and maintenance strategy type is directly confirmed as the final strategy; if the resources do not meet the requirements, a dynamic strategy adjustment mechanism is initiated to find the second operation and maintenance strategy type corresponding to the cell adjacent to the current cell in the operation and maintenance strategy matrix that meets the resource requirements as an alternative strategy.

6. A differentiated operation and maintenance method based on real-time weak links of equipment according to claim 5, characterized in that, The steps for generating differentiated operation and maintenance strategy instructions based on the matching results specifically include: when the final strategy is a planned maintenance strategy, the strategy generation engine calls the maintenance knowledge base, matches the corresponding standard maintenance procedures, required spare parts list, and estimated working hours from the standard maintenance operation library according to the specific values ​​of the equipment type and health evaluation results, and generates a planned maintenance instruction containing specific work order content; when the final strategy is a condition monitoring enhancement strategy, the strategy generation engine adjusts the parameters of the online monitoring device related to the equipment, including increasing the monitoring data sampling frequency, enabling the diagnostic algorithm model, and setting stricter early warning thresholds, and generates a monitoring parameter adjustment instruction; when the final strategy is an immediate fault repair strategy, the strategy generation engine combines the geographic information system and the resource scheduling system to calculate the optimal dispatch path for maintenance personnel and the spare parts delivery path, and automatically generates a repair work order instruction containing personnel, materials, paths, and emergency operation steps; all generated instructions are encapsulated as structured task objects, with accompanying execution priority and time constraint information.

7. A differentiated operation and maintenance method based on real-time weak links of equipment according to claim 1, characterized in that, It also includes closed-loop feedback and optimization steps after the execution of the operation and maintenance strategy: after executing the differentiated operation and maintenance strategy instructions, continuously collect new online monitoring data and power supply reliability operation data of the equipment after a complete operation and maintenance cycle to obtain feedback data; Based on the feedback data, the health evaluation result of the device is recalculated and compared with the health evaluation result before the strategy was executed to obtain the change in health status. Analyze the correlation between the changes in the health status and the type of operation and maintenance strategy implemented, and evaluate the effectiveness of the strategy; The strategy effectiveness assessment results, device identification, executed strategy type, and corresponding health status change are used as a training sample and stored in the strategy effect sample library. The updated strategy effect sample library is used periodically to jointly optimize and iterate the status evaluation model and the operation and maintenance strategy matrix to dynamically correct the accuracy of the health evaluation and the applicability of the operation and maintenance strategy matching.

8. A differentiated operation and maintenance method based on real-time weak links of equipment according to claim 1, characterized in that, It also includes a global operation and maintenance strategy adjustment step that considers network-level collaboration: periodically execute the differentiated operation and maintenance method based on real-time weak links of equipment as described in claim 1 on all equipment in the distribution network in batches to generate a preliminary differentiated operation and maintenance strategy instruction set for all equipment; Conflict detection is performed on all preliminary differentiated operation and maintenance strategy instructions to identify instruction combinations that conflict in time, space, or resources. A global optimization model is established with the goal of maximizing the overall power supply reliability and minimizing the total operation and maintenance cost. The set of preliminary differentiated operation and maintenance strategy instructions is input into the global optimization model as decision variables. A multi-objective evolutionary algorithm is used to solve the global optimization model. Under the premise of satisfying global resource constraints, the operation and maintenance strategy types and execution time windows of some devices are coordinated and adjusted to output a global collaborative operation and maintenance plan that eliminates conflicts and achieves Pareto optimality.

9. A differentiated operation and maintenance system based on real-time weak points of equipment, based on the differentiated operation and maintenance method based on real-time weak points of equipment according to any one of claims 1 to 8, characterized in that, include: The system includes a data acquisition and fusion module, an equipment health assessment module, an equipment importance assessment module, an operation and maintenance strategy matching module, and an operation and maintenance strategy generation module. The data acquisition and fusion module is configured to collect online monitoring data and power supply reliability operation data of equipment in the distribution network in real time, process and fuse the data to obtain a multi-source dataset of equipment status, and send the multi-source dataset of equipment status to the equipment health evaluation module and the equipment importance assessment module. The device health assessment module is configured to calculate the current health status score of the device based on the received multi-source dataset of device status using a status assessment model, obtain the health assessment result, and send the health assessment result to the operation and maintenance strategy matching module. The device importance assessment module is configured to assess the impact of device failures on power supply reliability based on the received power supply reliability operation data, combined with the device's location and load importance in the network topology, using the analytic hierarchy process (AHP) to obtain an importance assessment result, and then send the importance assessment result to the operation and maintenance strategy matching module. The operation and maintenance strategy matching module is configured to input the received health evaluation result and importance evaluation result into a preset operation and maintenance strategy matrix for matching, determine the corresponding operation and maintenance strategy type, and send the matching result to the operation and maintenance strategy generation module; the operation and maintenance strategy generation module is configured to generate differentiated operation and maintenance strategy instructions for the device based on the received matching result, the differentiated operation and maintenance strategy instructions include specific operation and maintenance measures, execution time windows and resource allocation schemes.

10. A differentiated operation and maintenance system based on real-time weak links of equipment according to claim 9, characterized in that, The equipment health evaluation module includes: a feature vector construction unit, a deep learning anomaly detection unit, a dynamic failure rate prediction unit, and a comprehensive evaluation mapping unit. The feature vector construction unit is configured to select a subset of features directly related to the physical degradation mechanism of the equipment from the multi-source dataset of equipment states, construct a health evaluation feature vector, and send the health evaluation feature vector to the deep learning anomaly detection unit. The deep learning anomaly detection unit is configured to input the received health evaluation feature vector into a pre-trained encoder-decoder structure anomaly detection model, and obtain a characterization of the equipment state bias by calculating the reconstruction error between the input features and the model's reconstructed features. The system receives an anomaly score indicating deviation from normal operating mode and sends this score to the dynamic failure rate prediction unit and the comprehensive evaluation mapping unit. The dynamic failure rate prediction unit is configured to calculate the probability of conditional failure of the equipment within a specific future time period using a Weibull proportional hazards model, based on the received anomaly score and data on the equipment's operational years and the average lifespan of similar equipment. This probability is then sent to the comprehensive evaluation mapping unit. The comprehensive evaluation mapping unit is configured to fuse the received anomaly score and the probability of conditional failure, normalize and map them to a preset health score range using a fuzzy comprehensive evaluation method, and finally output a quantified health evaluation result.