A power distribution network maintenance plan automatic arrangement method and system

Abstract

The application discloses a kind of power distribution network maintenance plan automatic arrangement method and system, method includes: the comprehensive data vector of equipment is obtained by pre-processing to multi-source data;Health degradation index of equipment is calculated based on comprehensive data vector, and equipment failure probability is calculated based on health degradation index and comprehensive data vector;The maintenance priority index of each equipment is calculated based on failure probability, and candidate maintenance object is screened according to priority index and generates candidate maintenance task;Multi-dimensional space-time constraint model is constructed;The comprehensive distance of any two candidate maintenance tasks is calculated based on comprehensive distance function, and candidate maintenance task is clustered to form task group according to comprehensive distance;In the range of multi-dimensional space-time constraint model, each task group carries out multi-objective optimization, generates initial maintenance plan;Grid safety check is carried out to initial maintenance plan.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and in particular to a method and system for automatically scheduling maintenance plans for distribution networks. Background Technology

[0002] Distribution network maintenance is a crucial and fundamental step in ensuring the reliable operation of power equipment and supporting the safe and stable operation of the system. A well-planned maintenance schedule can not only improve power supply reliability and ensure stable system operation, but also further reduce the area and duration of power outages, thereby enhancing the overall economic efficiency of maintenance work.

[0003] Currently, there are three main methods for scheduling maintenance plans in power distribution networks: periodic maintenance, reactive maintenance, and condition-based maintenance. Periodic maintenance is performed on a fixed schedule, which is simple to implement but can easily lead to over-maintenance or under-maintenance. Reactive maintenance is a passive response, resulting in significant power outage losses. Condition-based maintenance, based on equipment condition assessment, has certain advantages. In terms of maintenance decision-making, condition-based maintenance determines priorities through failure rate models, health indices, or economic indicators, and introduces opportunistic maintenance strategies, cost-effectiveness evaluations, and optimization methods such as genetic algorithms and particle swarm optimization to generate plans, which has made some progress.

[0004] However, existing condition-based maintenance scheduling methods for distribution networks still have significant shortcomings. First, constraint modeling is incomplete, focusing too much on the equipment's own condition or a single economic objective, failing to incorporate spatiotemporal factors such as maintenance team location, road access time, work windows, power outage permits, weather conditions, material arrival, and the timing of multiple tasks into a unified model. This results in theoretically feasible plans but poor engineering execution. Second, a lack of coordination mechanisms means that planned maintenance, defect handling, and emergency repairs are independent, easily leading to task conflicts, redundant resource allocation, and insufficient power supply to critical users. Third, the interrelationships of space, time, and resources are not fully utilized, resulting in limited automation and executability.

[0005] Therefore, there is an urgent need to propose an automatic scheduling method for distribution network maintenance plans that comprehensively considers multidimensional spatiotemporal constraints and has collaborative optimization and dynamic adjustment capabilities. Summary of the Invention

[0006] To address the shortcomings of existing technologies, such as incomplete constraint modeling, lack of collaborative mechanisms for multiple types of maintenance tasks, and insufficient mining and utilization of spatiotemporal resource relationships, which lead to poor executability of the generated maintenance plans and susceptibility to resource conflicts, this invention provides an automatic arrangement method and system for distribution network maintenance plans. The specific technical solution is as follows: An automatic scheduling method for distribution network maintenance plans includes the following steps: S1. Preprocessing multi-source data to obtain a comprehensive data vector of equipment; S2. Calculating the health degradation index of equipment based on the comprehensive data vector; constructing a dynamic fault model of equipment based on the health degradation index and the comprehensive data vector; calculating the equipment failure probability within the planning period based on the dynamic fault model; calculating the maintenance priority index of equipment based on the failure probability; filtering candidate maintenance objects according to the priority index; and generating candidate maintenance tasks for the candidate maintenance objects; S3. A multi-dimensional spatiotemporal constraint model that includes at least time, space, resources, and power grid operation; S4. Based on a comprehensive distance function... Calculate the comprehensive distance between any two candidate maintenance tasks, and cluster and merge the candidate maintenance tasks into task groups based on the comprehensive distance; S5. Construct a multi-objective optimization function for each task group, and iteratively solve it within the scope of the multi-dimensional spatiotemporal constraint model to generate an initial maintenance plan; S6. Perform grid safety verification on the initial maintenance plan. If the verification fails, perform iterative correction or infeasibility processing until any of the following termination states are reached: all tasks pass the verification, some tasks are postponed and the remaining tasks pass the verification, or an infeasibility result list is output; correspondingly, the formal maintenance plan, the postponed task list, or the infeasibility result list are output.

[0007] Preferably, step S2 includes the following sub-steps: calculating the health degradation index of the equipment using a weighted summation method based on the equipment comprehensive data vector; constructing a dynamic failure probability model based on the health degradation index and the comprehensive data vector, the dynamic failure probability model taking into account dynamic failure rate, basic failure rate, aging impact factor, load impact factor, and maintenance impact factor; calculating the failure probability of the equipment based on the dynamic failure probability model; calculating the equipment maintenance priority index based on failure probability, equipment importance, defect level, user impact, and power outage impact; comparing the equipment maintenance priority index with a preset maintenance threshold, and if the priority index is greater than the preset maintenance threshold, including the equipment in the candidate maintenance object set; generating a corresponding maintenance task for each candidate maintenance object in the candidate maintenance object set.

[0008] Preferably, step S5 includes the following sub-steps: taking task groups as optimization units, encoding the execution order, shift allocation, and planned start time of each task group into initial population individuals; constructing a multi-objective optimization function with the optimization objectives of minimizing equipment operation risk, minimizing power outage losses, minimizing maintenance costs, minimizing total maintenance time, minimizing resource utilization imbalance, and minimizing plan disturbance; performing non-dominated sorting and iteration on the multi-objective optimization function of each task group under the constraints of the multi-dimensional spatiotemporal constraint model to obtain a non-dominated solution set, wherein the target value of the plan disturbance is set to zero when generating the initial maintenance plan; and selecting the optimal solution of each non-dominated solution set as the initial maintenance plan using a weighted comprehensive evaluation.

[0009] Preferably, step S4 includes the following sub-steps: S41. Treat each candidate maintenance task as an initial task group; S42. Treat each candidate maintenance task as an initial task group, calculate the comprehensive distance between all cross-group task pairs between the two groups using a comprehensive distance function, and determine the maximum value as the comprehensive distance between the two groups; wherein, the comprehensive distance function takes into account geographical distance, electrical topology distance, time window difference, and resource requirement difference; S43. In each iteration, select the two task groups with the smallest comprehensive distance between groups that is not greater than a preset clustering threshold and merge them to form a new task group; S44. Repeat steps S42 to S43 until the comprehensive distance between any two task groups is greater than the preset clustering threshold, and stop merging to obtain the final task group.

[0010] Preferably, step S6 includes the following sub-steps: S61. Based on the initial maintenance plan, the planning period is divided into several discrete time periods according to the task start time, task end time, and switch operation time. The network topology is updated and power flow calculation is performed for each time period to obtain the power flow calculation results for each time period; S62. The power flow calculation results are verified using a multi-dimensional spatiotemporal constraint model. The verification includes node voltage verification, line current carrying capacity verification, transfer capacity verification, power supply verification for important users, operation safety and operation sequence verification, personnel qualification and resource verification, and weather and external conditions verification. If all time periods pass the verification, proceed to step S64; S63. If any verification fails, a verification list containing the default type, default value, affected tasks, and suggested repair strategies is generated, and a progressive correction process including local correction, global re-optimization, and infeasibility handling is executed in sequence. Depending on the different termination states of the correction process, the process enters the final plan output, delayed task output, or infeasibility result output branch; S64. The termination status is determined by any of the following: all tasks pass verification, some tasks are postponed but the remaining tasks pass verification, or an infeasible results list is output. The output is the formal maintenance plan, the postponed task list, or the infeasible results list.

[0011] Preferably, the method further includes step S7. When executing the final maintenance plan, monitor changes in external conditions and perform local rolling optimization on the affected tasks and their related tasks.

[0012] Preferably, step S7 includes the following sub-steps: S71. Continuously monitor changes in external conditions, and trigger rolling optimization when external conditions change; S72. When rolling optimization is triggered, identify tasks directly affected by the changes in external conditions as a set of directly affected tasks, include tasks that share resources, are in the same power outage area, or have a time-series dependency with the directly affected tasks into a set of associated tasks, and take the union of the set of directly affected tasks and the set of associated tasks as the rolling optimization object; S73. After determining the rolling optimization object, freeze the start time, shift, and power outage range of the remaining tasks; S74. Obtain the latest equipment status and load forecast at the trigger time. Data from measurements, weather forecasts, road traffic, and on-site operations are used to cover the original plan inputs and update the multidimensional spatiotemporal constraint model accordingly. The arrangement of the rolling optimization objects in the original plan is encoded as the initial population seed, and the weight of the plan disturbance degree objective in the multi-objective optimization is increased. The rolling optimization objects are re-optimized based on the original plan to generate candidate update schemes. S75. The candidate update schemes are subjected to safety verification. If the verification is passed, the scheme is accepted. If it fails, the scope of related task extraction is expanded or low-priority tasks are removed until an update scheme that passes the verification is formed. The list of differences between the update scheme and the original plan is output and sent to the field for execution.

[0013] Preferably, step S1 includes the following sub-steps: collecting multi-source data; performing format conversion, anomaly removal, missing data completion, time synchronization, spatial coordinate unification and normalization on the multi-source data to obtain a comprehensive data vector for each device, wherein the comprehensive data vector includes basic device attributes, real-time load information, historical fault information, defect status information, historical maintenance information, environmental impact information, geospatial location and information affecting user load.

[0014] Preferably, the constraints of the multidimensional spatiotemporal constraint model include: Time constraints: Each maintenance task must be completed within the permitted work time window; ,in, For the task The earliest time that can be allowed to start, For the task The latest time that must be completed, The start time of the plan, Duration of the task; Space constraints: Maintenance team m from task Location to Mission The travel time to the location is The consecutive execution time of the two tasks should meet the following requirements. Traffic time is determined by the road network impedance function. Calculation, where Free passage time for roads. For traffic flow, For road traffic capacity, and For road impedance adjustment parameters; Resource constraint: At any time t, the amount of maintenance resources used must not exceed the total amount of resources, i.e. ,in, Indicates task The demand for the k-th type of resource This represents the total amount of the k-th type of resource. This is a task execution status indicator. Power grid operation constraints: Line current meets The node voltage satisfies ,in, For maintenance of lines or equipment under the topology The operating current at time t, This is the maximum allowable current for the line. Let be the voltage magnitude of node n at time t. and These are the lower and upper limits of the permissible voltage, respectively.

[0015] This invention also provides an automatic scheduling system for distribution network maintenance plans, applied to the above-mentioned automatic scheduling method for distribution network maintenance plans, comprising: The multi-source data access module is used to preprocess multi-source data to obtain the device's comprehensive data vector; The maintenance task generation module calculates the health degradation index of the equipment based on the comprehensive data vector; constructs a dynamic fault model of the equipment based on the health degradation index and the comprehensive data vector; calculates the equipment failure probability within the planned cycle based on the dynamic fault model; calculates the maintenance priority index of the equipment based on the failure probability; filters candidate maintenance objects according to the priority index; and generates candidate maintenance tasks for the candidate maintenance objects. The multidimensional spatiotemporal constraint modeling module is used to construct multidimensional spatiotemporal constraint models that include constraints related to time, space, resources, power grid operation, security, and business rules. The collaborative optimization orchestration module is used to calculate the comprehensive distance between any two candidate maintenance tasks based on the comprehensive distance function, and to cluster and merge the candidate maintenance tasks into task groups based on the comprehensive distance. The initial maintenance plan generation module is used to construct multi-objective optimization functions for each task group, iteratively solve them within the scope of the multi-dimensional spatiotemporal constraint model, and generate an initial maintenance plan. The safety verification and scheme modification module is used to perform power grid safety verification on the initial maintenance plan. If the verification fails, iterative modification or infeasibility processing is performed until any of the following termination states are reached: all tasks pass the verification, some tasks are postponed and the remaining tasks pass the verification, or an infeasibility result list is output. The corresponding output is the formal maintenance plan, the list of postponed tasks, or the list of infeasibility results. The local rolling optimization module is used to monitor changes in external conditions and perform local rolling optimization on affected tasks and their related tasks when executing the final maintenance plan.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention effectively solves multiple technical problems in existing power distribution network maintenance plan scheduling, such as reliance on manual experience, incomplete consideration of constraints, lack of collaborative optimization among maintenance tasks, and failure to integrate emergency repair optimization with daily plan scheduling, through a complete process from multi-source data access, equipment status assessment, candidate task generation, multi-dimensional spatiotemporal constraint modeling to collaborative optimization solution, and safety verification closed-loop correction. This method achieves automatic generation and dynamic adjustment of maintenance plans, significantly reducing the subjectivity and lag of manual scheduling; through multi-dimensional spatiotemporal constraint model joint constraints, the maintenance plan simultaneously meets on-site operating conditions and power grid safety operation requirements; through task clustering and merging and simultaneous shutdown and repair, duplicate power outages and duplicate work assignments are reduced, improving resource utilization efficiency; and through safety verification and closed-loop correction mechanisms, the plan is ensured to pass all safety verifications before issuance, reducing the risks of temporary changes and execution. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0018] Figure 1 This is a flowchart of the automatic scheduling method for power distribution network maintenance plans according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0023] Example 1 Please refer to Figure 1 , Figure 1 This is a flowchart of a method for automatically scheduling distribution network maintenance plans. This embodiment provides a method for automatically scheduling distribution network maintenance plans, which includes the following steps: S1. Preprocess the distribution network data to obtain the comprehensive data vector of the equipment.

[0024] Specifically, step S1 includes the following sub-steps: S11. Collect data from multiple sources.

[0025] In this example, the multi-source data includes equipment ledger systems, power distribution automation systems, production management systems, defect management systems, power outage planning systems, geographic information systems, meteorological systems, traffic information systems, and mobile work terminals. The multi-source data includes at least: equipment type, equipment number, years of operation, historical maintenance records, defect level, fault records, real-time operating status, load curves, switch status, feeder topology, user importance level, maintenance team location, personnel qualifications, vehicle status, tool configuration, spare parts inventory, road travel time, weather forecasts, and work safety restrictions.

[0026] S12. Perform format conversion, anomaly removal, missing data completion, time synchronization, spatial coordinate unification and normalization on the multi-source data to obtain the comprehensive data vector of each device. The comprehensive data vector includes the device's basic attributes, real-time load information, historical fault information, defect status information, historical maintenance information, environmental impact information, geospatial location and information affecting user load.

[0027] To ensure that data from different systems can participate in subsequent calculations in a unified manner, this step performs format conversion, anomaly removal, missing data completion, time synchronization, and spatial coordinate unification on multi-source heterogeneous data.

[0028] Specifically, the format conversion includes: establishing a mapping relationship between source fields and standard fields based on the device's unique code and data dictionary; uniformly converting device types, defect levels, switch statuses, and user levels from different systems into preset enumeration codes; uniformly converting quantities such as current, voltage, load, distance, and time into standard units; and uniformly converting date and time to the same time zone and the "year-month-day hour:minute:second" format. Fields for which mapping cannot be established are written into an abnormal data list and sent for manual confirmation.

[0029] Anomaly removal includes rule-based validation and statistical validation. Rule-based validation identifies data that exceeds the equipment's rated range, violates logical relationships, or is inconsistent with the feeder topology; statistical validation uses the interquartile range (ICM) or three-standard-deviation method to identify abrupt changes. Anomalies that can be verified by adjacent time points or other data sources are corrected, while anomalies that cannot be verified are marked as invalid and do not participate directly in subsequent calculations.

[0030] Missing data is handled according to data type: static data such as equipment rated parameters and commissioning date are prioritized for completion from equipment ledgers or the most recent valid record; continuous time series such as load and environmental quantities are completed using linear interpolation or spline interpolation when the proportion of consecutive missing data does not exceed 10%; discrete state quantities are completed using the most recent valid value; when key fields are missing or the proportion of consecutive missing data exceeds 10%, the corresponding equipment is marked as having incomplete data and is transferred to manual review.

[0031] Time synchronization is based on the calculation step size of the maintenance plan. To unify the time base, The time interval can be 15 min, 30 min, or 1 hour. Data from each system is resampled to a unified time using the most recent timestamp matching or linear interpolation method; when the deviation between the source data timestamp and the target time exceeds... At that time, instead of forcing a match, it is marked as not timely enough.

[0032] In this embodiment, 30 minutes is used as the calculation step size for the maintenance plan. The maximum permissible deviation between real-time operating quantities such as current, voltage, and switch status and the target time is taken as 5 minutes; the maximum permissible deviation of load forecast, meteorological, and traffic data is taken as 15 minutes; and the update time deviation of static data such as equipment ledgers is taken as 24 hours. Data exceeding the corresponding limits is marked as insufficient timeliness and is not directly overwritten by the most recent valid value.

[0033] When spatial coordinates are unified, the coordinates of each system are transformed to the same geographic coordinate system or projected coordinate system, and the equipment location is determined through a three-level association of "unique equipment code - feeder topology node - spatial coordinates". For data with only addresses or tower numbers, geographic codes and topological adjacency relationships are used to supplement the coordinates; when multiple candidate locations conflict, the record with the same unique equipment code and electrical topology is given priority, and the remaining conflicts are transferred to manual verification.

[0034] Furthermore, in order to eliminate the influence of data with different dimensions on the calculation results, various indicators are normalized.

[0035] For positive indicators, their normalized expression is: For negative indicators, their normalized expression is: in, For the i-th device, the k-th original index, The maximum value of the indicator. The minimum value of the indicator. The value is the normalized index value.

[0036] After preprocessing, the integrated data vector of the device is obtained. The expression for the integrated data vector is: in, Let i be the comprehensive data vector of the i-th device. For the i-th device's basic attributes, This provides the real-time load information for the i-th device. This contains historical fault information for the i-th device. This is the defect status information for the i-th device. This contains the historical maintenance information for the i-th device. This provides environmental impact information for the i-th device. Indicates the geospatial location of the i-th device. This represents the user and load information of the i-th device.

[0037] The equipment's basic attributes include equipment type, unique equipment code, voltage level, rated capacity, manufacturer, model, commissioning date, design life, and feeder to which it belongs; real-time load information includes current, voltage, active power, reactive power, load rate, peak value in the past 24 hours, and average value in the past 7 days; historical fault information includes the number of faults, fault type, fault duration, outage load, and average fault interval in the past 1, 3, and 5 years; defect status information includes defect code, defect level, discovery time, duration of non-elimination, and deterioration trend; historical maintenance information includes the time of the most recent maintenance, maintenance type, work items, test results, quality acceptance conclusion, and post-maintenance fault records; environmental impact information includes temperature, humidity, pollution level, rainfall, salt spray, thunderstorms, icing, and wind speed; geospatial location includes latitude and longitude, pole or station number, transformer area, and road access distance; and affected user and load information includes the number of affected users, important user level, important load capacity, and expected power outage.

[0038] Step S2. Calculate the health degradation index of the equipment based on the comprehensive data vector; construct a dynamic fault model of the equipment based on the health degradation index; calculate the equipment failure probability within the planned cycle based on the dynamic fault model of the equipment; calculate the maintenance priority index of each piece of equipment based on the equipment failure probability; filter candidate maintenance objects according to the priority index; and generate candidate maintenance tasks for the candidate maintenance objects.

[0039] Specifically, step S2 includes the following steps: S21. Based on the comprehensive data vector of the equipment, the health degradation index of the equipment is calculated using the weighted summation method.

[0040] Specifically, for the i-th device, the formula for its health degradation index is: in, The equipment health degradation index, Let be the normalized value of the k-th state index. The corresponding indicator weights, and satisfying In this example, the equipment's operating status, aging degree, defect level, fault history, load level, and environmental factors are used as status indicators to calculate the equipment's health degradation index. A higher health degradation index value indicates a higher overall degree of degradation. Specifically, during the calculation, positive indicators (those with higher values ​​indicating higher degradation) are normalized using a positive indicator normalization formula; negative indicators (those with higher values ​​indicating better equipment condition) are normalized using a negative indicator normalization formula.

[0041] Furthermore, the operational status indicators include voltage over-limit rate, current over-limit rate, number of abnormal temperature rises, partial discharge or insulation monitoring anomalies, and number of switch failures or malfunctions; the aging degree indicators include the ratio of years in operation to design life, insulation aging trend, and remaining life of key components; the defect indicators include defect level, number of days without elimination, and number of times the same defect recurs; the fault history indicators include the fault frequency in the past 3 years, the recurrence rate of faults in the past year, and the average fault interval; the load indicators include the average load rate in the past 24 hours, the peak load rate, and the cumulative duration of load rates exceeding 70%; and the environmental indicators include temperature, humidity, pollution, salt spray, thunderstorms, and icing risk. All indicators are positively processed and uniformly mapped to [0,1].

[0042] It should be noted that the index weights adopt the entropy weight method. The entropy weight method automatically assigns weights based on the dispersion of different devices across various state indicators. This reduces the subjectivity of manual weighting based on experience and avoids the problems of complex judgment matrix construction and reliance on expert judgment for consistency verification found in the analytic hierarchy process (AHP). Furthermore, compared to methods that rely on historical data training with a large number of fault samples, the entropy weight method has lower requirements for sample labels and is more suitable for engineering scenarios where fault samples in distribution network equipment are relatively sparse and equipment types vary significantly. The more significant the difference in an index among different devices, the stronger its ability to distinguish equipment states, and the larger its corresponding weight. Conversely, if an index shows little variation among devices, its distinguishing effect is weak, and its corresponding weight is smaller.

[0043] Specifically, let the number of devices participating in the evaluation be... The number of status indicators is , No. The device in the The normalized value of the item indicator First calculate the... The first item under the indicator The proportion of each device: in, It is a very small positive number, used to avoid When =0, a logarithmic calculation anomaly occurs, and its range of values ​​can be: ~ , This is the normalized value of the k-th state index.

[0044] Throughout this text, extremely small positive numbers used to avoid denominators of zero or logarithmic anomalies are uniformly denoted as ε, with a range of values ​​of [insert range here]. ~ In this embodiment, .

[0045] Then calculate the first... Information entropy of the indicator: in, Indicates the first Information entropy of the indicator For the first The first item under the indicator The proportion of each device. When the difference in a certain indicator is small among the devices, its The distribution is relatively uniform, and the information entropy is relatively high. A larger entropy indicates that the indicator has a weaker ability to distinguish device status; when an indicator differs significantly between different devices, its entropy is smaller, indicating that the indicator contains more effective distinguishing information.

[0046] Further calculation of the first Coefficient of variation of the indicators: For the first The coefficient of difference for each indicator.

[0047] Finally, we get the number. Weight of each indicator: To reach the first The weight of each indicator.

[0048] The overall degree of equipment degradation is indicated by calculating the health degradation index. A higher health degradation index indicates a higher degree of equipment degradation.

[0049] S22. Construct a dynamic failure probability model based on the health degradation index and comprehensive data vector. The dynamic failure probability model takes into account the dynamic failure rate, basic failure rate, aging impact factor, load impact factor and maintenance impact factor.

[0050] Specifically, the expression for the dynamic failure probability model is: In the formula, Let be the dynamic failure rate of device i at time t. Based on the failure rate, As an influencing factor of aging, As the load impact factor, As an environmental impact factor, To inspect for degradation or repair of influencing factors.

[0051] Among them, the basic failure rate This is obtained from statistics on the historical number of failures and cumulative operating time of equipment of the same type and voltage level. Specifically, it uses... Perform a smoothing estimate. Among these, The number of failures during the statistical period. For cumulative runtime, , Smoothing parameters to avoid distortion in small samples.

[0052] The smoothing parameter 'a' represents the prior equivalent number of failures, ranging from 0.1 to 1.0; the smoothing parameter 'b' represents the prior equivalent cumulative runtime, ranging from 10 to 50 equipment-years. In this embodiment, a = 0.5 and b = 20 equipment-years are used to avoid estimating the basic failure rate as 0 when the sample size of similar equipment is small or when there are no failures during the statistical period.

[0053] Aging Influencing Factors Based on years of operation, design life, and health degradation index Confirmed. Specifically, in this example, the Weibull risk ratio method is used, with the health degradation index as a covariate to adjust the risk ratio of equipment as it ages. The Weibull distribution has long been used for modeling equipment lifespan, failure time, and aging failure rate; its risk ratio is typically expressed as: in, For shape parameters, is the scale parameter. When The risk rate increases with equipment service time, making it suitable for describing the failure risk after equipment enters the aging and wear-out period. NIST lists the Weibull model as a widely used life distribution model in reliability engineering. In another embodiment, a piecewise function method can be used to calculate the aging impact factor. This method directly sets a fixed risk multiple range based on the equipment's life stage and fine-tunes it using a health degradation index. The higher the equipment health degradation index or the closer the service life is to the design life, the larger the factor value, accelerating the increase in the failure rate.

[0054] Load Influence Factor Calculated based on the actual load rate. Specifically, the actual load rate of the equipment is calculated using the following formula: ,in, The actual load of the equipment at time t. The rated load capacity of the equipment. The load impact factor is calculated by comparing the actual load rate of the equipment with the economic operating range, where the load rate does not exceed the overload threshold. If the load factor is set to 1, and the load exceeds the preset overload threshold, the load impact factor is calculated based on the actual load rate and the overload threshold. Load Influence Factor It increases exponentially, based on the square of the load factor. The overload threshold is set to 70% of the rated load.

[0055] In this embodiment, a load rate of 0.30–0.70 is defined as the economic operating range. The load impact factor is set to 1.00 when the load rate is not higher than 0.70; 1.10 when the load rate is greater than 0.70 but not higher than 0.85; 1.30 when the load rate is greater than 0.85 but not higher than 1.00; 1.80 when the load rate is greater than 1.00 but not higher than 1.10; and 2.50 when the load rate exceeds 1.10, with a cutoff value not exceeding 3.00. For equipment with a smaller rated capacity or poor heat dissipation, the 0.70 threshold can be lowered to 0.60–0.65 based on the manufacturer's permissible load curve.

[0056] Environmental impact factors The environmental impact factor is calculated by normalizing and weighting environmental indicators such as temperature, humidity, pollution level, salt spray, thunderstorm, and icing. It should be noted that environmental indicators such as temperature, humidity, pollution level, salt spray, thunderstorm, and icing are derived from multi-source data. After normalizing each indicator, the entropy weight method is used to calculate the weight of each indicator, and then the results are summed to obtain the environmental impact factor.

[0057] When all environmental conditions are within the allowable range for the equipment, the environmental impact factor is 1.00; when there is a single general adverse condition, it is 1.10 to 1.30; when there are two or more adverse conditions at the same time, or when there is a warning of heavy pollution, high temperature, or icing, it is 1.50 to 2.00; when the conditions for prohibiting operation or seriously threatening the operation of the equipment are met, the impact factor is no longer used as the sole factor, but directly forms meteorological or safety hard constraints.

[0058] Maintenance Influence Factors It is determined by the time since the last maintenance, the type of maintenance, the results of the maintenance quality acceptance, and the fault records after maintenance.

[0059] The maintenance impact factor is determined by multiplying the four types of coefficients and truncating to [0.5, 3.0]: when the time since the last maintenance does not exceed 50% of the specified cycle, is 50%–100% of the cycle, 100%–150% of the cycle, and exceeds 150%, the time coefficient is 0.80, 1.00, 1.20, and 1.50, respectively; when the maintenance type is major repair, minor repair, special handling, inspection and testing, or temporary emergency repair, the type coefficient is 0.75, 0.90, 1.00, and 1.10, respectively; when the quality acceptance conclusion is excellent, qualified, requires rectification, or unqualified, the quality coefficient is 0.85, 1.00, 1.20, and 1.50, respectively; when there are no faults within 180 days after maintenance, one similar fault occurs, or more than two similar faults occur, the fault record coefficient is 1.00, 1.20, and 1.50, respectively.

[0060] It should be noted that each influencing factor is usually limited to the range of 0.5 to 3 to avoid a single factor causing an abnormal amplification of the failure rate.

[0061] S23. Calculate the failure probability of the equipment based on the dynamic failure probability model.

[0062] The probability of equipment failure within the planned period can be expressed as: in, This marks the end of the maintenance plan evaluation cycle. This is the start time of the maintenance plan assessment cycle, and > The evaluation period can be set to 1 day, 1 week, 1 month or 1 year, depending on the type of plan.

[0063] S24. Calculate the equipment maintenance priority index based on failure probability, equipment importance, defect level, user impact, and power outage impact.

[0064] Furthermore, equipment maintenance priority index The calculation is based on equipment failure probability, equipment importance, defect severity, power outage impact range, and impact on key users. The expression is as follows: in, For equipment Probability of failure during the maintenance plan assessment period; For equipment importance indicators; This is an indicator of defect severity. This refers to the extent of the power outage's impact. For user sensitivity or key user impact metrics; The weight coefficients of the corresponding indicators, and satisfying the following conditions: Among them, the defect severity index The importance index of equipment is determined based on its current defect level. Values ​​for no defect, general defect, severe defect, and critical defect are 0, 0.33, 0.67, and 1, respectively. The value is determined based on the equipment type in the equipment register and its functional location in the feeder topology. The equipment register data is associated with the feeder topology data based on the unique equipment code. The value is 0.20 for low-voltage terminal equipment or ordinary terminal branch equipment, 0.40 for ordinary distribution transformers or ordinary branch line equipment, 0.70 for main feeder equipment or sectionalizing switches, 0.85 for tie switches, critical ring main units, or equipment performing load transfer functions, and 1.00 for substation busbars, main transformers, or feeder outlet switches. When the same equipment meets two or more level conditions simultaneously, the larger value is taken. Power outage impact range index. The calculation is based on the number of users still affected by power outages after individual equipment is taken out of service, the unsupplied load capacity, and the number of lost transformer areas. Before calculating the equipment maintenance priority index, a pre-assessment scenario for individual equipment shutdown is established for each piece of equipment. The assessment time is the moment when the predicted downstream load of the equipment is the highest within the planned period. The power transfer scheme with the largest restored load capacity is selected from the pre-established feasible power transfer scheme library, and the line current carrying capacity and node voltage are checked for the selected power transfer scheme. If no power transfer scheme passes the check, the power outage impact range is calculated as if there is no power transfer condition. All equipment participating in the status assessment and with complete data constitutes the set of equipment to be evaluated. ,equipment The number of users who still experience power loss after being individually shut down and the aforementioned power transfer scheme implemented is: The load capacity that still cannot be restored to power is The number of transformer substations still without power is Then, respectively, according to: as well as Perform normalization processing, and follow the following: Calculate the power outage impact range indicators, among which, For the corresponding weight coefficients, satisfying As an optional configuration, it is advisable to User sensitivity or key user impact metrics Based on the user levels and their unsupplied load capacity after the equipment is individually taken out of operation and the above pre-assessed power transfer plan is implemented, let the set of users still affected by the power outage be . ,but: in, For equipment After exiting the program, the user Unsupplied load capacity For users Importance coefficient, To avoid extremely small positive numbers with a denominator of zero, the importance coefficients for Level 1 Important Users, Level 2 Important Users, Temporary Important Users, General Users, and Interruptible Load Users are set to 1.00, 0.80, 0.60, 0.30, and 0.10, respectively. When there are no unattended users after the equipment is taken out of service, let... When the remaining power outage includes uninterruptible loads such as hospital emergency systems, emergency command centers, rail transit control systems, or critical communication hubs, and power cannot be maintained through dual power sources, backup power sources, generators, or other power supply measures, then... This direct assignment rule takes precedence over the aforementioned weighted calculation rule. The weighting coefficients are determined using a combination of initial weighting using entropy weighting and business adjustments. First, all devices participating in the status evaluation and with complete data within the planning period are used as the calculation sample. Then, based on each device's... Calculating the initial entropy weight based on the dispersion of the five indicators Based on the principles of prioritizing equipment safety, serious defects, and important users, a business correction coefficient greater than 0 is set for each indicator. and in accordance with: The initial entropy weights are corrected and normalized to obtain the final weight coefficients. When the number of devices to be evaluated is less than two, or when the value of a certain indicator is exactly the same across all devices, that indicator cannot form an effective entropy weight based on its dispersion. In this case, a preset weight is used as a backup configuration, which can be selected as an optional configuration. Therefore, the equipment importance indicators are directly derived from the equipment ledger and feeder topology, while the power outage impact range indicators and user sensitivity indicators are derived from the pre-assessment results of individual equipment shutdowns. These indicators are consistent with the data and judgment criteria used in subsequent power transfer capacity, line current carrying capacity, node voltage, and power supply verification for important users. This forms a closed-loop calculation process from multi-source data preprocessing, single equipment shutdown pre-assessment, calculation of various indicators, weight determination, maintenance priority calculation, to selection of candidate maintenance targets.

[0065] S25. Compare the equipment maintenance priority index with the preset maintenance threshold. If the priority index is greater than the preset maintenance threshold, the equipment is included in the candidate maintenance object set. Otherwise, the equipment is not included in the current planning cycle.

[0066] After completing the calculation of the equipment maintenance priority index, the equipment maintenance priority index will be... Compared with a preset maintenance threshold, if the priority index is greater than the preset maintenance threshold, the equipment is included in the candidate maintenance object set. Furthermore, the preset maintenance threshold ranges from [0.50, 0.80], and its specific value is based on historical maintenance work orders. The distribution and planned cycle can be determined using maintenance capacity and allowable outage scale. Preferably, the preset maintenance threshold is 0.65.

[0067] S26. Generate a corresponding maintenance task for each candidate maintenance object in the candidate maintenance object set.

[0068] Furthermore, the maintenance task is defined as follows: in, This represents the i-th maintenance task. Indicates the object to be inspected. Indicates the type of maintenance work. Indicate the work location. Indicates the expected duration of the operation. To allow for job time windows, Represents the resource demand vector. This is a task priority index. This indicates the area affected by the power outage.

[0069] S3. Construct a multi-dimensional spatiotemporal constraint model that includes time, space, resources, and power grid operation.

[0070] Regarding time constraints, each maintenance task must be completed within the allowed working time window. (Task description follows.) The start time is The duration of the task is Allow job window to be Then we have: in, For the task The earliest time that can be allowed to start, For the task The latest time that must be completed, The start time of the plan, The duration of the task.

[0071] If task Must precede the task If executed, the task priority relationship is constrained as follows: in, For the task The start time, For the task The start time is , For the task The duration of the task.

[0072] The task priority is determined according to the following rules: Within the same equipment or the same power outage operation chain, the task order is forcibly ordered according to the standard operating procedure of "power outage isolation → voltage testing and grounding → maintenance work → quality acceptance → power restoration". When there are material, vehicle, or professional process dependencies between tasks, the preceding task takes precedence over the subsequent task. For example, if task B can only begin after the materials required for task A have arrived, vehicles have been allocated, or the preceding process has been completed, then task A takes precedence over task B. For tasks executed by the same work group with overlapping time windows, priority is determined by the task priority index. Sorting from highest to lowest; for tasks that depend on the operation of power transfer or tie switches in the electrical topology, they are arranged in the order of completing power transfer first, followed by power outage maintenance.

[0073] In terms of spatial constraints, the road network is abstracted as a graph. The free passage time for each road edge r is Traffic flow is The road capacity is The road impedance can then be expressed in the following form: in, and For road impedance adjustment parameters, The value range is [0.05, 0.30]. The value range is [2, 6], preferably, =0.15, =4.

[0074] Where μ is the traffic congestion sensitivity coefficient and ν is the congestion growth index. μ ranges from 0.05 to 0.30, and ν ranges from 2 to 6; in this embodiment, the commonly used BPR road impedance parameters μ=0.15 and ν=4 are adopted. Road traffic flow is updated every 15 minutes; when real-time traffic data is unavailable, the historical average traffic flow for the same period of the same week is used.

[0075] The shortest travel time between the two work points is calculated based on road impedance: in, For the task To the mission The set of feasible paths For one of the paths, Let be the impedance of road edge r at time t.

[0076] If maintenance team m completes its task Then proceed to the mission Then the start time must meet the following conditions: in, For the task The start time, For the task The start time, Let t be the shortest travel time between the two work locations at time t.

[0077] In terms of resource constraints, each task needs to meet the requirements for personnel, vehicles, tools, and spare parts.

[0078] set up Indicates task Whether it is to be carried out by team m, , This refers to the number of available maintenance teams; each task can only be performed by one primary responsible team, i.e. .

[0079] For any time t, the maintenance resource usage must not exceed the total resource limit: in, Indicates task The demand for the k-th type of resource This represents the total amount of the k-th type of resource. For task execution status indication, when hour, Select 1 if the value is 1, otherwise select 0.

[0080] Regarding grid operation constraints, maintenance plans must not create unacceptable risks of power outages, overloads, or power loss for critical users. For each maintenance period, the post-maintenance network topology is determined based on the equipment to be shut down and power transfer operations, and power flow calculations are performed in conjunction with load forecasting or real-time measurements.

[0081] Define the boundary for quantifying power outage losses: Let the unsupplied load at node n due to the maintenance plan at time t be . The importance weight of the nodes is The weighted load loss can then be expressed as: in, The weighted load loss of node n at time t. For node n, the unsupplied load at time t, As the importance weight of the nodes, The total number of load nodes or user aggregation nodes participating in the power outage loss statistics.

[0082] The importance weight of a node is determined based on user level and load characteristics and normalized to [0,1]: 1.00 for Level 1 important users, 0.80 for Level 2 important users, 0.60 for temporary important users, 0.30 for general users, and 0.10 for interruptible loads. When a node contains multiple types of users, the weighted sum is calculated according to the load capacity ratio of each type of user; when uninterruptible loads such as hospitals, emergency command, and rail transit control are involved, the weight is directly set to 1.00. This weight is also used in the calculation of the power outage loss target in step S5 to ensure consistency between the safety verification and the optimization target.

[0083] For the operating current of a line or equipment, the following should be met: in, For maintenance of lines or equipment under the topology The operating current at time t, This is the maximum allowable current for the line. Preferably, to maintain an operating margin, the maximum allowable current for the line is... It can be set to 0.80 to 1.00 times the rated allowable current.

[0084] For node voltage constraints, the following should be satisfied: in, Let be the voltage magnitude of node n at time t. and These are the lower and upper limits of the permissible voltage, respectively.

[0085] Furthermore, the multidimensional spatiotemporal constraint model also includes safety and business rule constraints, including operational safety distance constraints, operation ticket sequence constraints, important user supply constraints, personnel qualification hard matching constraints, and meteorological and external environment constraints. The above constraints are logical decision-type constraints.

[0086] S4. Calculate the comprehensive distance between any two candidate maintenance tasks based on the comprehensive distance function, and cluster and merge the candidate maintenance tasks to form task groups based on the comprehensive distance.

[0087] Specifically, step S4 includes the following sub-steps: S41. Treat each candidate maintenance task as an initial task group.

[0088] S42. For any two task groups, calculate the combined distance of all cross-group task pairs between the two groups, and determine the maximum value among them as the combined inter-group distance between the two groups.

[0089] Specifically, a comprehensive distance function is used to calculate the comprehensive distance between each pair of candidate maintenance tasks. This function takes into account geographical distance, electrical topology distance, time window differences, and resource requirement differences. The comprehensive distance between any two candidate maintenance tasks is calculated based on this function to measure the degree of synergy between them.

[0090] The expression for the composite distance function is: in, Indicates geographical distance. Indicates electrical topology distance, Indicates the difference in time windows. Indicates differences in resource demand. , , , These are the weighting coefficients.

[0091] Furthermore, Indicates task To the mission The difference in geographical distance between them is calculated using the following formula: in, For the task With the task The actual road distance or straight-line distance between work sites shall be based on the shortest distance calculated from the road network. The preset maximum acceptable distance for collaborative maintenance can be determined based on the service radius of the power supply station or work team, or by statistical analysis of historical maintenance work orders in the same area. For example, it can be the 80%-90% percentile of the daily service radius of the same work team, the typical working radius of the same feeder, or the distance of historical combined maintenance tasks. The smaller the value, the closer the two tasks are geographically, and the more suitable they are for joint construction.

[0092] In this embodiment, the maximum acceptable road distance for collaborative maintenance of urban tasks is 10 km, and for suburban or rural power distribution networks it is 20 km. If no region is distinguished, the default distance is 15 km. If the estimated travel time on the road exceeds 45 minutes, even if the distance meets the requirements, the tasks will not be prioritized for merging into the same task group.

[0093] Indicates task With the task The electrical topological distance between them. Specifically, the distribution network is abstracted as a graph structure. ,in A distribution network node set, which includes switches, transformers, branch boxes, ring main units, or user access points. This is the distribution network edge set, representing line or electrical connection relationships. The shortest electrical path length between the corresponding device nodes of the two tasks is... Then we have: in, The maximum electrical topology distance can be preset according to the division rules of the same feeder, same branch, same transformer area, or same power outage range.

[0094] In another embodiment, Alternatively, values ​​can be assigned according to discrete rules: if two tasks are located within the same power outage area or the same branch line, then... If they are located on the same feeder but different branches, then take... If it is located on an adjacent feeder that can be supplied via a tie switch, then take... If there is no obvious electrical connection between the two, then take . The smaller the value, the stronger the electrical connection between the two tasks, making them more suitable for simultaneous shutdown and repair.

[0095] Indicates task With the task The difference in task time windows. (Set task) The allowed job time window is ,Task The allowed job time window is The overlap length of the time windows for both is: The time window difference can then be expressed as: When the time windows of two tasks completely overlap, the time window difference This indicates the best time compatibility; when two tasks do not have overlapping job windows. This indicates that the windows do not overlap, and This indicates that it is not suitable to merge them into the same job window.

[0096] Indicates task With the task The differences in resource requirements. Assume the task... With the task For the first The demand for each type of resource is as follows and If the number of resource types is K, then the resource demand difference can be expressed as: in, For the number of resource types, This is a very small positive number, used to avoid the denominator being zero. When the personnel qualifications, vehicles, tools, and material types required for two tasks are very similar... The smaller the value, the better the resource sharing. If the total resources required after merging two tasks exceed the currently available resources, or if there are hard constraints such as unqualified personnel, conflicting specialized tools, or insufficient critical spare parts, then a direct decision will be made. This indicates that the conditions for collaborative orchestration are not met.

[0097] Weighting coefficient These represent the relative importance of geographical distance, electrical topology distance, time window difference, and resource requirement difference in task relevance determination, respectively, satisfying the following: This invention uses the entropy weight method to determine The calculation method is the same as the weight calculation method in the health deterioration index in step S2, and will not be repeated here.

[0098] S43. In each iteration, select the two task groups with the smallest inter-group comprehensive distance that is not greater than the preset clustering threshold and merge them to form a new task group.

[0099] S43. Repeat steps S42 to S43 until the inter-group comprehensive distance between the two task groups is greater than the preset clustering threshold, and obtain the final task group.

[0100] The final output task groups are such that the combined distance between any two candidate maintenance tasks within each final task group is no greater than the preset clustering threshold.

[0101] Furthermore, a preset clustering threshold is set. This can be determined based on historical maintenance work orders and manually combined maintenance records. The specific method is as follows: statistically analyze task pairs that have historically been manually assigned to simultaneous shutdown and maintenance, adjacent joint maintenance, or resource sharing, and calculate their comprehensive distance function. The distribution is analyzed, and the 80%–90% quantile is used as the initial threshold. If historical samples are lacking, the initial threshold can be set according to the operation and inspection rules, such as taking a preset clustering threshold. And in subsequent operations, it will be rolled over and adjusted based on the actual merge success rate, resource conflict rate, and number of planned adjustments. When the conditions are met... At that time, the task was considered With the task Those with strong correlation can be included in the same candidate maintenance group; when At that time, it was considered that the two tasks were not sufficiently related and were not prioritized for merging and scheduling.

[0102] Preset clustering threshold The selectable range is 0.40 to 0.60, and this embodiment uses 0.50. After going live, the value will be reviewed quarterly based on the task merging success rate and resource conflict rate: when the resource conflict rate after merging is higher than 10%, the value will be reduced by 0.05; when the merging success rate is higher than 85% and the resource conflict rate is lower than 5%, the value will be increased by 0.05, but the adjusted value will still be limited to 0.40 to 0.60.

[0103] S5. Construct multi-objective optimization functions for each task group, iteratively solve them within the scope of the multi-dimensional spatiotemporal constraint model, and generate an initial maintenance plan.

[0104] Specifically, step S5 includes the following sub-steps: S51. Using task groups as the optimization unit, the execution order, execution team allocation, and planned start time of each task group are encoded as initial population individuals.

[0105] For the task set of the task group The r-th candidate maintenance plan is coded as follows: in, Indicates task The candidate maintenance group to which it belongs. This indicates the order in which tasks are executed within a group or work team. This indicates the maintenance team number that performed the task. This indicates the planned start time of the task. The task's order within the group or shift execution sequence is determined by the equipment maintenance priority index. Determined, based on equipment maintenance priority index Tasks are sorted, with those having higher priority indices being assigned first. This coding method can simultaneously describe the grouping of maintenance tasks, their execution order, shift allocation, and time schedule.

[0106] Depend on Individual combinations of candidate maintenance plans form the initial population. : in, For population size, N represents the number of candidate maintenance tasks to be scheduled.

[0107] The population size is determined by the number of candidate tasks N, taking max(40, 4N), and limited to between 40 and 200; in this embodiment, it is set to 100 when N=25. If the number of candidate tasks exceeds 50, the population size is set to 200, and partition optimization or rolling window is preferred to reduce the size of a single solution.

[0108] S52. With the optimization objectives of minimizing equipment operation risk, power outage loss, maintenance cost, total maintenance time, resource utilization imbalance, and plan disturbance, the maintenance plan is non-dominated sorted and iterated under the constraints of a multi-dimensional spatiotemporal constraint model to obtain a non-dominated solution set; wherein, when generating the initial maintenance plan, the target value of the plan disturbance is set to zero; in the rolling optimization in step S7, the plan disturbance is used to quantify the degree of deviation of the updated scheme from the original plan.

[0109] In this step, each individual in the initial population... Decoding is performed to generate a detailed maintenance plan, which includes the start and end times, executing team, resource allocation, outage scope, and power transfer scheme for each task. It should be noted that the power transfer scheme comes from a pre-generated library of feasible schemes that meet the requirements of energized path connectivity, tie line capacity, and protection coordination. During decoding, candidate schemes are selected based on the outage scope of the task. The selected power transfer scheme serves as input for power flow, line current carrying capacity, and power supply verification for important users. Furthermore, after the formal plan passes verification, it is output to the outage operation ticket or dispatch review stage.

[0110] Then, a multidimensional spatiotemporal constraint model was used to verify the constraints of the scheme.

[0111] Furthermore, the constraints include: Time window constraints: ; Spatial constraints: ; Resource capacity constraints: ; Line current carrying constraints: ; Node voltage constraints: ; in, Allow the job window for the task. For the duration of the task, For the work group From the task To the mission Traffic time For the task For the first The demand for such resources For the first Total amount of resources of this type For line current, This represents the node voltage.

[0112] If the solution does not meet any of the above constraints, constraint repair will be initiated according to the type of breach.

[0113] When a time window constraint is violated, the task is shifted forward or backward within the allowed window, and its successor task is moved synchronously; if the window length is insufficient, the task group is split or low-priority tasks are postponed. When spatial constraints are violated, the task order of the same work group is rearranged first. If this is still not satisfied, a qualified work group that is closer in distance is assigned. When resource capacity constraints are violated, try the following in sequence: stagger concurrently executed tasks, reallocate idle resources, replace work groups, and split task groups. When line current carrying capacity or node voltage constraints are violated, try adjusting the transfer path, reducing the single transfer load, splitting the outage area, changing the maintenance period, or canceling the task merging in sequence.

[0114] After each repair operation is performed, it is reassessed whether the affected period meets the above constraints of time window, space, resource capacity, line current carrying capacity, and node voltage. If all constraints are met, the repair is considered successful. If the repair is successful, the candidate maintenance plan is retained and further optimization is carried out. If all repair strategies fail, the individual is marked as infeasible and its selection probability is reduced through constraint penalty terms.

[0115] The penalty for breach of contract can be represented as: in, Indicates the first Normalized degree of default for each constraint This represents the total number of constraints included in the current candidate plan. Furthermore, constraints such as line current carrying capacity constraints and resource capacity constraints apply, ensuring that the actual value does not exceed the upper limit. ,in, This represents the portion of the actual value that exceeds the upper limit. This indicates that the larger of the absolute value of the upper limit and the smallest positive number ε should be used; the actual value must not be lower than the lower limit. , This represents the portion of the lower limit value that exceeds the actual value. This indicates that the absolute value of the lower limit is greater than or equal to ε; for "yes or no" logical constraints, when satisfied... = 0, when not satisfied =1.

[0116] Based on the penalty term for constraint violation, the objective function for a solution with constraint violation is modified as follows: in, For the j-th objective function after adding the penalty, For the j-th optimization objective, the original calculated value is... To constrain the penalty coefficient, ∈[ Preferably, its initial value is taken as... During the iteration process, according to Adaptive increase, The initial constraint penalty coefficient, This represents the maximum number of iterations. Let j be the j-th objective after adding the penalty for breach of contract.

[0117] The constraint penalty coefficient κ ranges from 10^2 to 10^6. In this embodiment, κ is initially set to 10^3 and increases linearly with the number of iterations. When hard constraints such as personnel qualifications, power failure of important users, or operation sequence are violated, they are directly treated as infeasible solutions, and soft constraint repair is not only carried out by the penalty term.

[0118] Furthermore, for each maintenance plan Calculate the six-dimensional target vector .

[0119] in, Indicates equipment operational risk. Indicates losses due to power outages. This indicates the cost of maintenance. This indicates the total maintenance time. Indicates the degree of uneven resource utilization. This indicates the degree of disturbance to the original plan during the rolling optimization process.

[0120] Furthermore, the equipment operation risk objective can be expressed as: in, Let be the priority index of the i-th maintenance task. The number of candidate tasks to participate in the optimization. This indicates the waiting time from the planned start time of the task to the actual start time of maintenance. The longer the waiting time for high-risk equipment, the greater the risk penalty.

[0121] The power outage loss target can be expressed as: in, Assign importance weights to load nodes or user aggregation nodes. The total number of load nodes participating in the power outage loss calculation. For the set of discrete verification times, Let t be the unsupplied load. This represents the time interval between adjacent verification times.

[0122] The maintenance cost target can be expressed as: in, for, For labor costs, For vehicle costs, For material costs, Costs related to power outages.

[0123] The time window objective can be represented as: in, Indicates the total duration of the plan. For the task The start time, For the duration of the task, This is the start time of the planned period.

[0124] To calculate the actual workload of a work group, the working time of work group m can be defined separately: in, Let m be the set of tasks to be performed by work group m. This represents the set of adjacent task pairs in the execution sequence for this work group. This refers to the travel time between adjacent work sites. Used for workload and resource balance evaluation.

[0125] The work duration of a work team is used to calculate the workload and resource utilization imbalance of the work team, and participates in the non-dominated ranking of the objective function. The task complexity coefficient for ordinary operations is 1.00, for high-altitude or near-electrical operations it is 1.20, and for operations involving multi-disciplinary collaboration, complex power transfer, or replacement of large equipment it is 1.50. The effective working time of a single shift is preferably no more than 8 hours. If it exceeds 10 hours, overtime penalties will be imposed, and if it exceeds 12 hours, it will be considered a breach of the hard constraints on resources and safety.

[0126] The goal of balanced resource utilization is expressed as: in, It indicates the degree of uneven resource utilization and is used to measure whether the use of resources such as maintenance teams, vehicles, and tools is balanced among different maintenance teams. Let m be the workload of the m-th maintenance team. This represents the average workload for all work groups.

[0127] Furthermore, the workload of the work team It can be calculated based on a combination of task duration, travel time, and task difficulty: in, Indicates task Whether by work group Execute; if executed, the value is 1; otherwise, it is 0. The working hours of team m; This is the task complexity coefficient, which can be determined based on the type of work, hazard level, personnel qualification requirements, or the difficulty of historical work orders.

[0128] The larger the value, the greater the difference in workload between work groups, indicating an imbalance in resource utilization or overload in some work groups.

[0129] Furthermore, This indicates the degree of disturbance to the original plan during the rolling optimization process, used to avoid frequent adjustments to the maintenance plan that could lead to chaos in on-site execution. In the initial plan generation phase, since there is no original plan to refer to, the target for the degree of plan disturbance is... =0; This target is enabled during subsequent rolling optimizations to quantify the offset of the update plan from the original plan's start time and the number of shift changes.

[0130] in, For maintenance tasks At the originally planned start time, Indicates task Start time in the update plan; This indicates the adjusted work group number. This indicates the work group number before the adjustment. For indicator functions; Change the penalty coefficient for the work group.

[0131] The penalty coefficient ρ for team change is used to convert a team change into an equivalent time disturbance. The value range is 1 to 4 hours. In this embodiment, it is 2 hours. When it involves a work order that has been processed, a site briefing that has been completed, or a task that requires special qualifications, it is 4 hours. For tasks that have not yet been issued and do not require special qualifications, it can be 1 hour.

[0132] To facilitate local search, solution comparison, and final selection, a weighted comprehensive score is defined: in, Let j be the normalized value of the j-th objective. For the corresponding weights, and This score does not replace the Pareto non-dominated ranking; it is only used to assist in the comparison of candidate solutions at the same level and for the final selection of the best solution.

[0133] Furthermore, The weights for each objective correspond to equipment operation risk, power outage loss, maintenance cost, total maintenance time, resource utilization imbalance, and plan disturbance degree, respectively, satisfying the following: In this example, the method of "entropy weight method + business priority adjustment" is used to determine the priority. Specifically, the system first calculates the normalized values ​​of each candidate maintenance plan on six objectives based on the candidate maintenance plan set, forming an objective evaluation matrix: in, For the first The candidate solution is in the... Normalized values ​​on each objective j=1,2,3...,6.

[0134] Calculate the first Under the first goal The proportion of each option: in, It is a very small integer, and its value range can be 1. ~ Preferred = .

[0135] Then calculate the first... Information entropy of a target: in, For the first The information entropy of each target.

[0136] Calculate the difference coefficient based on information entropy: in, For the first The difference coefficient of each objective.

[0137] Calculate the first based on the difference coefficient. Initial weights for each objective: in, For the first The initial weights of each objective.

[0138] To reflect the business principles of "safety first, reliable power supply first, and taking into account economy and operational stability" in distribution network maintenance, a business correction coefficient is further introduced. The final weights are obtained as follows: Among them, the business correction coefficient It can be determined according to operation and maintenance management requirements, power supply reliability assessment requirements, important user protection requirements, and maintenance plan management rules.

[0139] As an optional implementation configuration, the business correction coefficients corresponding to equipment operation risk, power outage loss, maintenance cost, total maintenance time, resource utilization imbalance, and plan disturbance degree can be taken as 1.20, 1.20, 0.90, 0.90, 0.80, and 1.30, respectively, and then multiplied with the entropy weight result and normalized; during regular offline orchestration, the plan disturbance target is turned off, and its weight is reset to 0.

[0140] Furthermore, all maintenance tasks in the population are non-dominated and ordered. This is achieved by comparing two maintenance plans. and For the two maintenance plans and ,like Non-inferior in all objectives And superior to at least one objective Then it is believed Dominate .

[0141] To facilitate comparisons between different objectives, after calculating the optimization objective, the population of each generation is normalized. in, For the original calculated value of the j-th target, Let be the minimum value of the j-th target. For the maximum value of the j-th target, For a very small positive number, when = At that time, Set all values ​​to 0.

[0142] Based on dominance relationships, the population is divided into multiple non-dominated layers: Among them, the first layer This represents the optimal non-dominated solution set in the current population. The lower the level, the better the overall performance of the solution.

[0143] The specific steps are as follows: Traverse the entire population, compare each candidate plan pairwise, and when a candidate plan individual... All objective functions are no worse than candidate program individuals And outperforms candidate program individuals in at least one objective. Then it is believed Dominate Thus, for each candidate program individual Calculate its dominance (by) (Dominant) set of individuals and domination of the individual (dominance) ) quantity ;Will Candidate plan individuals with a value of 0 are added to the first non-dominated layer; subsequent non-dominated layers are generated sequentially until all candidate plan individuals are stratified.

[0144] To prevent the algorithm from converging to a local optimum too early, the crowding distance is calculated for candidate plans within the same non-dominated layer.

[0145] Specifically, the crowding distance is calculated separately for each objective. First, the crowding distance for a single objective function is calculated: for the same undominated layer... There are 10 candidate plan individuals, sorted in ascending order according to their j-th objective; the crowding distance of boundary individuals is set to infinity, and the distance of the remaining candidate plan individuals r to the j-th objective is... ,in, Let the objective function value be the value of the individual ranked after individual r for the j-th objective. Let r be the objective function value of the individual preceding individual r for the j-th objective. The maximum value of all individuals in the current non-dominated layer on the j-th objective. This represents the minimum value for all individuals in the current non-dominated layer on the j-th objective. After calculating the congestion distance for each of the five objective functions (equipment operation risk, power outage loss, maintenance cost, total maintenance time, and resource utilization imbalance), the final congestion distance is obtained by summing all objectives. The greater the congestion distance, the sparser the distribution of the candidate maintenance plan in the area, and the higher its retention value.

[0146] Furthermore, a tournament selection method is used to determine the parent individuals. Specifically, candidate maintenance plans in the initial population are compared pairwise. When comparing two candidate maintenance plans, the one with the lower non-dominance level is preferentially selected as the parent individual; if the two non-dominance levels are the same, the one with the greater crowding distance is selected as the parent individual. The number of generated parent individuals is equal to the number of offspring individuals; that is, if the initial population contains If there are 10 candidate plans, then the parent and offspring individuals will each have 100% of the 100% ... One candidate program.

[0147] Perform crossover operations on the selected parent individuals to generate new maintenance plans.

[0148] The expression for the interleaving operation is: Cross-operations can be implemented through task sequence cross-operations, maintenance group cross-operations, and shift assignment cross-operations. In this example, a combined task sequence and shift time cross-operation approach is used, with the parent... and Generate offspring .

[0149] Specifically, from the father generation Inherited high-priority tasks are sorted from the parent generation. The system inherits part of the work group allocation and time window arrangement, thereby generating new offspring individuals.

[0150] For example, the father generation The order of tasks is father generation for If selected from inherit If the relative order is determined, the offspring will be retained first. Press again Fill in the blanks in the order that do not appear. , ,form Team assignment and time window are taken from Subsequently, constraints were used to resolve team conflicts and time window violations.

[0151] After the parent individuals undergo crossover, the newly generated offspring individuals undergo mutation.

[0152] The expression for the mutation operation is: Mutation operations with preset probabilities It acts on offspring individuals with a pre-set low probability. The process of randomly perturbing its encoding to increase population diversity.

[0153] Specifically, the system iterates through each newly generated offspring individual. Generate a random number between 0 and 1. If the random number is less than or equal to the preset probability... If the random number is greater than the preset probability, then mutation is performed on the offspring individual. If so, then the offspring will remain unchanged. Desirable ~0.20, where N is the total number of tasks.

[0154] When mutation is required, one mutation operator is randomly selected from the list to operate on the individual. Then, based on the selected mutation operator, the corresponding component in the offspring's genetic code is directly modified. Furthermore, mutation operators include task advance, task combination or splitting, shift / replacement, start time shift, and adjacent task exchange. Specifically, task advance means that, under the premise of satisfying pre-constraints, the start time of a high-risk or high-priority task is adjusted forward (moved by 1-4 steps). Task combination / merging or splitting indicates merging tasks from two adjacent areas into the same maintenance group; or splitting a task group with severe resource conflicts into independent tasks; Team replacement indicates randomly changing the execution team number of a task or task group (reassigning from available teams); Start time shift indicates shifting the start time forward or backward within the allowed work window, with the step size taken from the base time step. 1 to 4 times; adjacent task exchange refers to exchanging the execution order of two adjacent tasks in the same shift's execution sequence.

[0155] After completing the crossover and mutation operations, the constraint repair mechanism is executed again to ensure that the new scheme meets time constraints, resource constraints, power grid operation constraints, and safety constraints. The final generated individual... They officially become members of the new generation of offspring, participating in subsequent merging of parent generations and elite selection.

[0156] To improve the algorithm's convergence speed, after each generation of evolution, 10%–30% of the superior individuals in the current population are subjected to local search, with 20% being the optimal choice. Within the same non-dominated layer, priority is given to individuals with larger crowding distances and weighted comprehensive scoring functions. Smaller individuals. Further, local search rules include: 1. Prioritize tasks with high equipment risk and long waiting times. Specifically, calculate the risk waiting contribution of each task, select the task with the largest contribution, and prioritize them without disrupting their prerequisite relationships and time windows. Move forward by step size.

[0157] The formula for calculating the risk waiting contribution is as follows: , Let be the maintenance priority index for the i-th task. This is the scheduled start time for the task within the current maintenance plan. This is the start time of the planned period.

[0158] 2. Group tasks with the same or similar power outage areas into the same maintenance window. Specifically, filter tasks based on the same power outage area or overall distance. Less than the preset clustering threshold For pairs of tasks, merging is performed when their shared time windows overlap, resources are not exceeded, and the power outage loss does not increase after merging. "Resources not exceeding limits" means that the total number of personnel, vehicles, tools, spare parts, etc., required for the combined execution of the two tasks within the same time period cannot exceed the total available resources of the system. "Power outage loss not increasing" means that the calculated combined power outage loss after merging the two tasks into the same power outage window must not exceed the sum of the power outage losses if the two tasks were executed separately.

[0159] 3. Transfer tasks with excessive workload to less busy work groups. Specifically, calculate the workload of each work group. With average load The difference will be used to assign transferable tasks from the highest-loaded shift to qualified shifts with lower loads.

[0160] 4. Reorder consecutive tasks with significant travel time to shorten the total travel time between different work locations for work teams, thereby reducing overall maintenance time and improving efficiency. Specifically, the system iterates through each pair of consecutive tasks in the current work schedule of the same work team and calculates the travel time between them. The system filters out task connections that exceed a preset threshold (e.g., 30 minutes). For these time-consuming connections, the system rearranges them within a certain range before and after them (e.g., the current task and its 2-3 subsequent tasks) using either adjacent swapping or a 2-opt method. The rearranged order must not violate the established sequential dependencies between tasks and must satisfy time window constraints; otherwise, the solution is discarded. The total traffic time before and after the rearrangement is compared. If the total traffic time after the rearrangement is less than the total traffic time before the rearrangement, the order adjustment is accepted, and the planned start times of all tasks in the shift are updated synchronously. If the time spent after the rearrangement does not decrease (or even increases), the adjustment is abandoned, and the original order is restored.

[0161] 5. Adjust the power transfer path or split the maintenance group for schemes that do not meet the power grid safety verification. Specifically, based on the power grid safety verification results, identify the specific causes and equipment locations leading to safety issues. Without changing the outage equipment and maintenance time, switch the current outage area to an alternative power transfer path from the pre-generated "feasible power transfer scheme library" based on the specific locations of current overruns, voltage overruns, or insufficient power transfer capacity. If "adjusting the power transfer" still cannot resolve the overruns, it indicates that the current task group is "too many or too large," and the task group needs to be split into smaller units. Specifically, split a task group containing multiple devices and a large outage area into two or more independent sub-task groups with smaller outage areas. Execute the split sub-task groups at different times to reduce the impact load of a single maintenance on the power grid, allowing the power transfer path to carry the remaining load. After each "adjusting the power transfer" or "splitting the maintenance group" operation, a complete power flow calculation and multi-dimensional constraint verification must be performed on the scheme immediately to determine whether all overruns have been eliminated and no new problems have arisen. Adjustments are accepted only if the overall objective function improves and all hard constraints are met; otherwise, the original solution is restored.

[0162] After generating a new offspring population, merge the parent and offspring populations: in, For the t-th generation parent population, Both are offspring populations generated by selection, crossover, mutation, and constraint repair, and contain One candidate plan; merging populations Contains 2 One candidate program.

[0163] Then merge the total group Recalculate the non-dominated sorting and crowding distance, and select the top... One excellent candidate program for the next generation population: The algorithm terminates when any of the following conditions are met: the maximum number of iterations reaches 200 generations; the number of solutions in the Pareto non-dominated solution set and the optimal value of each objective remain unchanged for 20 consecutive generations; the relative improvement of the optimal weighted comprehensive score is less than 0.1%; or the computation time for a single arrangement reaches 300 seconds. When the number of candidate tasks exceeds 100, the maximum number of iterations can be increased to 500 generations and the upper limit of computation time can be increased to 900 seconds.

[0164] After the algorithm terminates, a set of Pareto non-dominated solutions is obtained.

[0165] S53. The optimal solution of each non-dominated solution set is selected by weighted comprehensive evaluation, and the optimal solutions are combined to form the initial maintenance plan.

[0166] To facilitate practical implementation, this invention further employs a weighted comprehensive evaluation method to select the final solution from the Pareto non-dominated solution set. A weighted comprehensive score is calculated for each candidate solution. The lower the score, the better the solution.

[0167] Select that satisfies all security constraints and The minimum acceptable solution should be used as the initial maintenance plan: S6. Perform power grid safety verification on the initial maintenance plan. If the verification fails, perform iterative correction or infeasibility handling until any of the following termination states are reached: all tasks pass the verification, some tasks are postponed and the remaining tasks pass the verification, or an infeasibility result list is output; correspondingly, the formal maintenance plan, the postponed task list, or the infeasibility result list is output.

[0168] Specifically, step S6 includes the following sub-steps: S61. Based on the initial maintenance plan, the planning period is divided into several discrete time periods according to the task start time, task end time and switch operation time. The network topology is updated and power flow calculation is performed for each time period to obtain the power flow calculation results for each time period.

[0169] In this step, the entire planned cycle of the initial maintenance plan is divided into several discrete time periods that are connected sequentially, according to the time when the power grid topology changes. The cutting points are determined based on the task start time, task end time, and switch operation time. The switch operation time indicates the time when the tie switch needs to be opened or closed when implementing the power transfer scheme. For example, if a planned cycle is one day (0:00~24:00), in which a task starts at 8:00, a closing operation occurs at 9:30, and a task ends at 12:00, the cycle is divided into 4 discrete time periods: [0:00, 8:00], [8:00, 9:30], [9:30, 12:00], [12:00, 24:00].

[0170] The network topology is updated for each discrete time period, specifically including: Identify out-of-service equipment: Based on the maintenance tasks being carried out during this period, identify which equipment is out of service.

[0171] Perform switching operations: Determine the open / closed status of the tie switch and sectionalizing switch during this period according to the bound power transfer plan.

[0172] Based on the above, the post-maintenance network topology corresponding to this time period is formed, and the load forecast data for this time period is overlaid to form a complete power flow calculation input model. It should be noted that forming the post-maintenance network topology corresponding to this time period is a basic and mature technical method in the field of power system safety verification, and will not be elaborated on again.

[0173] For each discrete time period of the reconstructed topology, a power flow calculation is performed independently to obtain the numerical solutions for the voltage magnitude of each node and the current of each branch within that time period. Then, it is determined whether the power flow calculation has converged. If the calculation has converged, complete voltage and current data are obtained, and the process proceeds to the subsequent verification stage. If the calculation has not converged (e.g., due to islanding or circulating current causing iterative divergence), the scheme for that time period is directly determined to be infeasible. In this case, there is no need to proceed to the subsequent verification stage; instead, a verification list is immediately generated and a correction process is triggered.

[0174] S62. Verify the power flow calculation results using a multidimensional spatiotemporal constraint model. Verification includes node voltage verification, line current carrying capacity verification, transfer capacity verification, power supply verification for important users, operation safety and operation sequence verification, personnel qualification and resource verification, and weather and external conditions verification. If all time periods pass the verification, proceed to step S64.

[0175] Specifically, the node voltage verification involves checking each node voltage amplitude obtained from the power flow calculation to see if it meets the node voltage constraints in the multidimensional spatiotemporal constraint model, and recording the maximum exceedance. If the voltage is below the lower limit or above the upper limit, the verification fails.

[0176] The line current carrying capacity check is as follows: it is determined whether the current of each line and transformer branch obtained from the power flow calculation meets the line current carrying capacity constraints in the multidimensional spatiotemporal constraint model. If the constraints are not met, the test is deemed unsuccessful.

[0177] The transfer capacity verification is as follows: determine whether the transfer path meets the requirements of electrical connection, operation conditions of the tie switch, and the remaining capacity of the feeder, transformer and tie line is greater than or equal to the load to be transferred, and at the same time verify the protection coordination and ring network operation rules; if any one of them is not met, it will fail.

[0178] The power supply verification for critical users involves verifying particularly important and first-level load users by spatially / topologically associating the outage area with the list of critical users. If a particularly important user experiences a power outage and no temporary backup measures (such as a generator truck) are configured, the verification is deemed unsuccessful.

[0179] The work safety and operation sequence verification involves: verifying the logic and safety distances of the planned operation ticket; based on the voltage level and work type, checking the power outage isolation point, grounding wire connection position, and the impact of parallel installations on the same pole. Ensure the operation sequence complies with the "Electric Power Safety Work Regulations" (e.g., strictly prohibiting the operation of disconnecting switches under load). Any operation logic error will result in a failure to pass the verification.

[0180] Personnel qualifications and resource verification: A thorough check is conducted on each team member's high-voltage electrician certificate, the supervisor's qualifications, vehicle condition, and the availability of special tools and spare parts. If the configuration falls below the requirements of the standard work package, the application is deemed unsuccessful.

[0181] Weather and external conditions verification: Associate the mission location / time period with weather warnings (strong winds, thunderstorms, high temperatures, icing) and road control information. If weather conditions exceed the operational threshold (e.g., wind speed > level 6) or roads are closed, the mission is deemed unfeasible for that time period.

[0182] A verification result is considered passed only if the power flow calculation results for all discrete time periods do not trigger any failure conditions in all seven types of verifications mentioned above. If any verification for any time period fails to meet the standard, a verification list is immediately generated. The verification list includes the specific location of the violation, the type of violation, the value of the violation, a list of affected tasks, and suggested repair directions.

[0183] S63. If any check fails, a check list is generated containing the default type, default value, affected tasks, and suggested repair strategies. A progressive correction process is then executed, which includes local correction, global re-optimization, and infeasibility handling. Depending on the termination status of the correction process, the process proceeds to the final plan output, delayed task output, or infeasibility result output branch.

[0184] Specifically, step S63 includes the following sub-steps: 1. Preset the maximum number of corrections. and number of consecutive times without effective improvement Initialize the current number of corrections n=0, and the number of consecutive times without effective improvement m=0; as an optional configuration, the maximum number of corrections... The number of times can be taken is 20 to 50, and in this embodiment it is taken is 30, which is the number of times there is no effective improvement. It can be done 2 to 5 times; in this embodiment, it is done 3 times.

[0185] 2. Determine whether the breach meets the local correction condition: When the breach only involves some tasks or some discrete time periods, and there is no need to change the candidate maintenance task set, task group division, and the shift allocation and execution sequence of other unaffected tasks, the local correction method is adopted; otherwise, the global re-optimization is triggered.

[0186] During partial correction: the planned start time, execution team, task sequence, outage range, and resource allocation of tasks not affected by the default are frozen, and adjustments are only made to directly affected tasks and their related tasks; the partial correction includes shifting the task start time within the allowed work time window, adjusting the task execution sequence within the same task group, reallocating currently available maintenance teams or resources, changing the transfer scheme from the feasible transfer scheme library, staggering tasks with resource conflicts, and adjusting the task duration or the interval between adjacent tasks without changing the task group division; after each partial correction, only the time period involved in the correction operation and its adjacent topology switching time period are updated, the network topology of the corresponding time period is re-determined, and power flow calculation and verification are performed on the affected time period; the partial correction does not re-execute the task group global multi-objective optimization in step S5, nor does it change the original arrangement of unaffected tasks.

[0187] Circumstances that trigger global re-optimization: The local correction count reaches the preset maximum number of local corrections but still fails to pass the verification; all local correction strategies for the current default type have been executed but the default has not been eliminated; local correction generates new hard constraint defaults; the default involves multiple task groups and cannot be eliminated while freezing other task arrangements; eliminating the default requires splitting or merging task groups, changing the candidate maintenance task set, postponing or removing maintenance tasks, reallocating shifts and resources across task groups, or significantly changing the task execution order and planned start time; After triggering global re-optimization, return to step S5, use the current maintenance plan, verification list, and identified default information as input for re-optimization, use the previous maintenance plan as the initial population seed, and increase the penalty weight of the corresponding default constraint according to the verification list, re-execute the multi-objective optimization of step S5, generate a new initial maintenance plan, and then re-execute the complete verification of step S6 for all discrete time periods of the entire planning cycle.

[0188] 3. Prioritize hard constraint conflicts, followed by grid operation overruns, and then resource and time conflicts. Select the appropriate correction strategy from the following options: adjust power transfer paths, change task execution time periods, reallocate maintenance teams and resources, adjust task execution order, split task groups, and postpone low-priority tasks. Only re-execute optimization and safety verification for the time periods and tasks affected by the correction operation.

[0189] 4. After completing the nth revision, compare the current verification result with the previous verification result; when the number of defaults decreases, or when the number of defaults remains unchanged, at least one default value of the same type decreases and no new hard constraint defaults occur, the current revision is deemed an effective improvement, and the number of consecutive no effective improvements is reset to zero; when the number of defaults does not decrease, the default values ​​of each type do not decrease, or a new hard constraint default occurs after the revision, the current revision is deemed not to have formed an effective improvement, and the number of consecutive no effective improvements is incremented by 1.

[0190] 5. When the revised maintenance plan passes all safety checks, stop the revision and proceed to step S64; when the number of revisions reaches... The number of consecutive times without effective improvement reached If all the preset correction strategies for the current default type have been executed but there is still a hard constraint default, stop repeatedly correcting the current task set and determine it as infeasible under the current constraint conditions, then proceed to the next step.

[0191] 6. When the current task set is determined to be infeasible, perform the following three-level recursive processing: Level 1: Identify the directly affected tasks and their associated tasks in the checklist that cause constraint conflicts, and prioritize splitting the task groups containing the aforementioned tasks; Level 2: If a feasible solution cannot be formed after the task group is split, the affected tasks will be adjusted to subsequent planning cycles or removed from the current optimization task set in order of equipment maintenance priority index from low to high. At least one task will be adjusted each time, and multi-objective optimization and safety verification will be re-executed for the remaining tasks. Level 3: Mandatory Task Retention and Manual Intervention – For critical defects, tasks involving personal safety, and tasks that cannot be postponed according to laws, regulations, or operating procedures, the postponement or removal process will not be implemented. Instead, these tasks will be retained as mandatory tasks and handled by suspending other ordinary maintenance tasks that conflict with them, expanding the scope of available resources, or transferring them to manual scheduling.

[0192] 7. If, after task group splitting and low-priority task postponement, all remaining tasks pass the safety check, the tasks that pass the check will be used as the final maintenance plan for the current planning cycle, and a list of postponed tasks will be output. If there are still unresolved hard constraint conflicts among the remaining tasks, the solutions that fail the safety check will not be used as the final maintenance plan, but a list of infeasible results will be output. The list of infeasible results shall include at least the tasks that could not be scheduled, the corresponding equipment, the type of breach, the breach value, the conflict period, related tasks, the corrective strategies that have been implemented, the reasons for infeasibility, and the suggested handling methods, and the relevant tasks will be transferred to the manual review or emergency dispatch process.

[0193] S64. Use any one of the following states as the termination state: "All tasks passed verification", "Remaining tasks passed verification after some tasks were postponed", or "Output list of infeasible results". Output the formal maintenance plan, the list of postponed tasks, or the list of infeasible results.

[0194] S7. During the execution of the maintenance plan, the plan shall be partially updated according to real-time status changes.

[0195] Partial updates follow the principles of "completed tasks remain unchanged, currently executing tasks remain unchanged in principle, unaffected tasks remain unchanged as much as possible, and only affected tasks and their related tasks are reordered," and optimize the degree of disturbance to the original plan during the rolling optimization process. Limit the scope of adjustment.

[0196] Specifically, step S7 includes the following sub-steps: S71. Continuously monitor changes in external conditions and trigger scrolling optimization when external conditions change.

[0197] Specifically, changes in external conditions include: increased equipment defect level, health degradation index or maintenance priority index exceeding preset threshold, excessive increase in road travel time, excessive load forecast deviation, changes in user power outage window or sudden resource unavailability. When any of the above occurs, rolling optimization is triggered.

[0198] Furthermore, to avoid excessively frequent dynamic rolling updates, quantified thresholds are set for various triggering conditions. Let the dynamic rolling update trigger flag be... When any triggering condition is met, let: Initiate rolling updates; otherwise: The original maintenance plan remains unchanged.

[0199] Specifically, the equipment defect level adopts discrete variables. This indicates that, according to the defect classification rules, no defects, general defects, serious defects, and critical defects are assigned values ​​of 0, 1, 2, and 3 respectively: The defect level is considered to have increased to meet the rolling update condition when the following conditions are met simultaneously: and: in, For the new defect level of the equipment, For the old defect level of the equipment, if If the value is 3, a rolling update will be triggered immediately.

[0200] For the health degradation index and maintenance priority index, the following trigger conditions can be set: or: in, For the new health deterioration index, The old health deterioration index, This is the priority index for new maintenance. This is the old maintenance priority index. It is a very small positive number, used to avoid the denominator being zero.

[0201] Regarding the road traffic obstruction, the original plan was to have the shifts from the task... To the mission The passage time is The real-time updated passage time is When the following conditions are met: or: It is believed that changes in road conditions have affected the execution of the maintenance plan, and rolling optimization should be triggered. Among these, For the team from the task To the mission The new maintenance schedule's opening hours. For the team from the task To the mission The original scheduled opening hours for maintenance. It is a very small positive number.

[0202] For load forecasting deviation, let the node or feeder at time... The predicted load is The actual load or the latest forecast load is When the following conditions are met: Rolling updates are triggered periodically. For general power distribution lines, the following values ​​are used: For critical users, heavy-load feeders, or scenarios with limited power supply capacity, the following should be considered: Regarding the change in the user's allowed power outage window, let the original allowed power outage window be: The updated power outage window is as follows: A rolling update is triggered when any of the following conditions are met: or: in, To allow for the start time of work in the updated maintenance plan. The updated maintenance plan allows for the completion time of the work. This is the permitted start time for the original maintenance plan. This is the permitted end time for the original maintenance plan. Estimate the operation time for the task. For safety margin, a value of 30 min is preferred.

[0203] S72. When rolling optimization is triggered, identify the tasks directly affected by the change in external conditions as the set of directly affected tasks, include the tasks that share resources, are in the same power outage area, or have a time-series dependency relationship with the directly affected tasks into the set of associated tasks, and take the union of the set of directly affected tasks and the set of associated tasks as the rolling optimization object.

[0204] When a rolling update is triggered, optimization is performed only on the affected task set and its associated task set. Let the affected task set be... Its associated task set is Then the object of scroll optimization is: Directly affected tasks are those that cannot be executed as originally planned due to changes in external conditions. Examples include tasks where road blockages prevent the work team from reaching the affected area, maintenance tasks for equipment with increased defect levels, and tasks with changing power outage windows. Related tasks are those that share resources, are in the same power outage area, or have a temporal dependency with the affected tasks. Resource sharing means tasks using the same work team, vehicle, or tools; being in the same power outage area means tasks located within the same power outage zone, and adjustments to the outage zone will affect them; temporal dependency means tasks with preceding or succeeding constraints.

[0205] S73. After determining the rolling optimization object, freeze the start time, shift, and power outage range of the remaining tasks.

[0206] The start time, shift, and power outage range of frozen tasks will not be modified in this rolling optimization. That is, the variables of frozen tasks will be removed from the optimization variables, and in each generation of evolution, the values ​​of these variables will be forced to remain at the original values ​​as planned, without any changes. Only the variables of the rolling optimization object will be subjected to genetic operations such as selection, crossover, and mutation.

[0207] Freeze completed tasks, tasks in progress, and tasks nearing completion and already approved. Release the start time, shift allocation, grouping, and transfer scheme variables of the rolling optimization object, and keep the original planned arrangements of the remaining tasks not included in the rolling optimization object unchanged.

[0208] S74. Obtain the latest equipment status, load forecast, weather forecast, road traffic and on-site operation feedback data at the trigger time, cover the original plan input and update the multidimensional spatiotemporal constraint model accordingly; encode the arrangement of the rolling optimization object in the original plan as the initial population seed, increase the weight of the plan disturbance degree objective in the multi-objective optimization, and re-perform multi-objective optimization on the rolling optimization object based on the original plan to generate candidate update schemes.

[0209] After the constraints are updated, a feasibility pre-verification should be performed on all tasks that have not yet started within the current planning period: After updating the multidimensional spatiotemporal constraint model using the latest equipment status, load forecast, weather forecast, road access, and on-site feedback data, a feasibility pre-verification should be performed on all tasks that have not yet started within the current planning period, not just the initially determined rolling optimization objects. If a frozen task encounters situations such as unmet weather conditions, inaccessible roads, resource or shift conflicts, line overload, node voltage exceeding limits, insufficient transfer capacity, or insufficient power supply to important users under the updated constraint model, then the frozen task should be re-identified as a directly affected task, its start time, shift, and outage range should be unfrozen, and it should be added to the rolling optimization objects; at the same time, tasks that share resources, are in the same outage area, have time-series dependencies, or are coupled with the task should also be added to the rolling optimization objects.

[0210] When rolling optimization is initiated, the start time, shift, grouping, and transfer scheme of tasks within the original maintenance plan's rolling optimization object are treated as a complete candidate and directly added to the initial population of the new population. The latest external data acquired when rolling optimization is triggered is updated in the constraint model, and a new set of multi-dimensional spatiotemporal constraint models suitable for the current operating conditions is generated. Specifically, external data includes equipment status, load forecasts, weather forecasts, road traffic information, and on-site feedback.

[0211] Then, increase the perturbation level in the objective function. The weights are determined, and within the updated multidimensional spatiotemporal constraint model, the original plan is used as the baseline for rolling optimization of the target object to perform multi-objective optimization again, generating candidate update schemes.

[0212] S75. Perform a security check on the candidate update scheme. If the check passes, the scheme is accepted. If it fails, the scope of related task extraction is expanded or low-priority tasks are removed until an update scheme that passes the check is formed. Output a list of differences between the update scheme and the original plan and send it to the site for execution.

[0213] When the verification fails due to insufficient definition of the original set of related tasks, the scope of related task extraction is expanded: newly emerging conflicting tasks are included in the related set, forming an expanded rolling optimization object. After expanding the scope, the process returns to step S72 to perform multi-objective optimization again, generating new candidate solutions before re-entering step S75 for verification. If the verification still fails after expanding the related scope, the tasks with the lowest maintenance priority index are identified, removed from the current planning cycle, and postponed to subsequent cycles. This releases the time window, team resources, and power outage space occupied by these low-priority tasks, allowing the remaining high-priority tasks to meet all constraints. After removing the low-priority tasks, steps S74 to S75 are re-executed until a solution that passes the verification is generated.

[0214] Once the candidate update plan passes all safety checks, a structured difference list is automatically generated based on the differences between the "updated plan" and the "original plan". The difference list includes: new tasks, canceled tasks, delayed tasks, changes in work groups, changes in time, and changes in supply transfer.

[0215] In summary, by employing a three-tiered progressive calculation system combining health index, dynamic failure rate model, and maintenance priority index, and integrating the entropy weight method for automatic identification of equipment state differences, the system achieves the screening of candidate maintenance targets and the automatic generation of task attributes, thus resolving the subjectivity and lag issues inherent in manual experience-based screening. Furthermore, by unifying time windows, road network accessibility, resource capacity, power grid operation, and safety rules into mathematical constraints, the maintenance plan simultaneously meets both on-site operational conditions and power grid safety operation requirements from the outset. Finally, by using a comprehensive distance function, the system achieves automatic task clustering based on geographical proximity, electrical topology correlation, time window compatibility, and resource demand similarity. By combining merged scheduling with improved NSGA-II algorithm's six-objective collaborative optimization, constraint repair, and local search strategies, simultaneous shutdown and repair, adjacent joint repair, and resource sharing were achieved, effectively reducing the number of repeated power outages and work organization costs. A closed-loop iterative mechanism of safety verification and scheme modification ensures that the final output plan can only be issued for execution after passing all hard constraint verifications, reducing the risk of temporary changes and execution due to unreasonable plans. Through local rolling optimization triggered by quantified thresholds and a plan disturbance penalty strategy, targeted adjustments are made only to affected tasks and their related tasks, balancing the overall stability of the original plan with the ability to respond quickly to sudden disturbances on site.

[0216] Example 2 This embodiment provides an automatic scheduling system for distribution network maintenance plans, applied to the automatic scheduling method for distribution network maintenance plans in Embodiment 1, which includes: The multi-source data access module is used to preprocess multi-source data to obtain the device's comprehensive data vector; The maintenance task generation module calculates the health degradation index of the equipment based on the comprehensive data vector; constructs a dynamic fault model of the equipment based on the health degradation index and the comprehensive data vector; calculates the equipment failure probability within the planned cycle based on the dynamic fault model; calculates the maintenance priority index of the equipment based on the failure probability; filters candidate maintenance objects according to the priority index; and generates candidate maintenance tasks for the candidate maintenance objects. The multidimensional spatiotemporal constraint modeling module is used to construct multidimensional spatiotemporal constraint models that include constraints related to time, space, resources, power grid operation, security, and business rules. The collaborative optimization orchestration module is used to calculate the comprehensive distance between any two candidate maintenance tasks based on the comprehensive distance function, and to cluster and merge the candidate maintenance tasks into task groups based on the comprehensive distance. The initial maintenance plan generation module is used to construct multi-objective optimization functions for each task group, iteratively solve them within the scope of the multi-dimensional spatiotemporal constraint model, and generate an initial maintenance plan. The safety verification and scheme modification module is used to perform power grid safety verification on the initial maintenance plan. If the verification fails, iterative modification or infeasibility processing is performed until any of the following termination states are reached: all tasks pass the verification, some tasks are postponed and the remaining tasks pass the verification, or an infeasibility result list is output. The corresponding output is the formal maintenance plan, the list of postponed tasks, or the list of infeasibility results. The local rolling optimization module is used to monitor changes in external conditions and perform local rolling optimization on affected tasks and their related tasks when executing the final maintenance plan.

[0217] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for automatically scheduling maintenance plans for a power distribution network, characterized in that, Includes the following steps S1. Preprocess the multi-source data to obtain the comprehensive data vector of the device; S2. Calculate the health degradation index of the equipment based on comprehensive data vectors; A dynamic equipment failure model is constructed based on the health degradation index and comprehensive data vector, and the probability of equipment failure within the planning cycle is calculated based on the dynamic equipment failure model. Based on the probability of equipment failure, calculate the maintenance priority index of the equipment, screen candidate maintenance objects according to the priority index, and generate candidate maintenance tasks for the candidate maintenance objects. S3. Construct a multi-dimensional spatiotemporal constraint model that includes at least time, space, resources, and power grid operation; S4. Calculate the comprehensive distance between any two candidate maintenance tasks based on the comprehensive distance function, and cluster and merge the candidate maintenance tasks to form task groups based on the comprehensive distance; S5. Construct multi-objective optimization functions for each task group, iteratively solve them within the scope of the multi-dimensional spatiotemporal constraint model, and generate an initial maintenance plan; S6. Perform power grid safety verification on the initial maintenance plan. If the verification fails, perform iterative correction or infeasibility handling until any of the following termination states are reached: all tasks pass the verification, some tasks are postponed and the remaining tasks pass the verification, or an infeasibility result list is output; correspondingly, the formal maintenance plan, the postponed task list, or the infeasibility result list is output.

2. The automatic scheduling method for power distribution network maintenance plans according to claim 1, characterized in that, Step S2 includes the following sub-steps: Based on the comprehensive data vector of the equipment, the health degradation index of the equipment is calculated using the weighted summation method. A dynamic failure probability model is constructed based on the health degradation index and comprehensive data vector. The dynamic failure probability model takes into account the dynamic failure rate, basic failure rate, aging impact factor, load impact factor and maintenance impact factor. Calculate the failure probability of the equipment based on the dynamic failure probability model; The equipment maintenance priority index is calculated based on failure probability, equipment importance, defect level, user impact, and power outage impact. The equipment maintenance priority index is compared with the preset maintenance threshold. Once the priority index is determined to be greater than the preset maintenance threshold, the equipment is included in the candidate maintenance object set. Generate a corresponding maintenance task for each candidate maintenance object in the candidate maintenance object set.

3. The automatic scheduling method for power distribution network maintenance plans according to claim 1, characterized in that, Step S5 includes the following sub-steps: Using task groups as the optimization unit, the execution order, shift allocation, and planned start time of each task group are encoded as individuals in the initial population. A multi-objective optimization function is constructed with the optimization objectives of minimizing equipment operation risk, power outage losses, maintenance costs, total maintenance time, resource utilization imbalance, and plan disturbance. Under the constraints of the multi-dimensional spatiotemporal constraint model, the multi-objective optimization function of each task group is non-dominated and iterated to obtain a non-dominated solution set. The plan disturbance objective value is set to zero when generating the initial maintenance plan. The optimal solution for each non-dominated solution set is selected as the initial maintenance plan using a weighted comprehensive evaluation.

4. The automatic scheduling method for power distribution network maintenance plans according to claim 1, characterized in that, Step S4 includes the following sub-steps: S41. Treat each candidate maintenance task as an initial task group; S42. Treat each candidate maintenance task as an initial task group, and use the comprehensive distance function to calculate the comprehensive distance between all cross-group task pairs between the two groups. The maximum value is determined as the comprehensive distance between the two groups. The comprehensive distance function takes into account geographical distance, electrical topology distance, time window difference, and resource requirement difference. S43. In each iteration, select the two task groups with the smallest inter-group comprehensive distance that is not greater than the preset clustering threshold and merge them to form a new task group; S44. Repeat steps S42 to S43 until the inter-group comprehensive distance between any two task groups is greater than the preset clustering threshold, then stop merging to obtain the final task groups.

5. The automatic scheduling method for distribution network maintenance plans according to claim 1, characterized in that, Step S6 includes the following sub-steps: S61. Based on the initial maintenance plan, the planning period is divided into several discrete time periods according to the task start time, task end time and switch operation time. The network topology is updated and power flow calculation is performed for each time period to obtain the power flow calculation results for each time period. S62. The power flow calculation results are verified using a multidimensional spatiotemporal constraint model. The verification includes node voltage verification, line current carrying capacity verification, power transfer capacity verification, power supply to important users verification, operation safety and operation sequence verification, personnel qualification and resource verification, and weather and external conditions verification. If all time periods pass the verification, proceed to step S64. S63. If any check fails, a check list containing the default type, default value, affected tasks and suggested repair strategies is generated, and a progressive correction process including local correction, global re-optimization and infeasibility handling is executed in sequence. Depending on the different termination states of the correction process, it enters the final plan output, delayed task output or infeasibility result output branch respectively. S64. The termination status is any of the following: all tasks pass verification, some tasks are postponed and the remaining tasks pass verification, or an infeasible result list is output. The output is the formal maintenance plan, the postponed task list, or the infeasible result list.

6. The automatic scheduling method for power distribution network maintenance plans according to claim 1, characterized in that, It also includes step S7. When executing the final maintenance plan, monitor changes in external conditions and perform local rolling optimization on the affected tasks and their related tasks.

7. The automatic scheduling method for power distribution network maintenance plans according to claim 6, characterized in that, Step S7 includes the following sub-steps: S71. Continuously monitor changes in external conditions and trigger scrolling optimization when external conditions change; S72. When rolling optimization is triggered, identify tasks that are directly affected by changes in external conditions as the set of directly affected tasks, include tasks that share resources, are in the same power outage area, or have a time-series dependency relationship with the directly affected tasks into the set of associated tasks, and take the union of the set of directly affected tasks and the set of associated tasks as the rolling optimization object. S73. After determining the rolling optimization target, freeze the start time, shift, and power outage range of the remaining tasks; S74 acquires the latest equipment status, load forecast, weather forecast, road traffic and on-site operation feedback data at the trigger time, overwrites the original plan input and updates the multidimensional spatiotemporal constraint model accordingly; encodes the arrangement of the rolling optimization object in the original plan as the initial population seed, increases the weight of the plan disturbance degree objective in the multi-objective optimization, and re-optimizes the rolling optimization object based on the original plan to generate candidate update schemes; S75. Perform a security check on the candidate update scheme. If the check passes, the scheme is accepted. If it fails, the scope of related task extraction is expanded or low-priority tasks are removed until an update scheme that passes the check is formed. Output a list of differences between the update scheme and the original plan and send it to the site for execution.

8. The automatic scheduling method for power distribution network maintenance plans according to claim 1, characterized in that, Step S1 includes the following sub-steps: Collect data from multiple sources; The data from multiple sources is processed through format conversion, anomaly removal, missing data completion, time synchronization, spatial coordinate unification, and normalization to obtain a comprehensive data vector for each device. The comprehensive data vector includes basic device attributes, real-time load information, historical fault information, defect status information, historical maintenance information, environmental impact information, geospatial location, and information affecting user load.

9. The automatic scheduling method for distribution network maintenance plans according to claim 1, characterized in that, The constraints of the multidimensional spatiotemporal constraint model include: Time constraints: Each maintenance task must be completed within the permitted work time window; ,in, For the task The earliest time that can be allowed to start, For the task The latest time that must be completed, The start time of the plan, Duration of the task; Space constraints: Maintenance team m from task Location to Mission The travel time to the location is The consecutive execution time of the two tasks should meet the following requirements. Traffic time is determined by the road network impedance function. Calculation, where Free passage time for roads. For traffic flow, For road traffic capacity, and For road impedance adjustment parameters; Resource constraint: At any time t, the amount of maintenance resources used must not exceed the total amount of resources, i.e. ,in, Indicates task The demand for the k-th type of resource This represents the total amount of the k-th type of resource. This is a task execution status indicator. Power grid operation constraints: Line current meets The node voltage satisfies ,in, For maintenance of lines or equipment under the topology The operating current at time t, This is the maximum allowable current for the line. Let be the voltage magnitude of node n at time t. and These are the lower and upper limits of the permissible voltage, respectively.

10. An automatic scheduling system for power distribution network maintenance plans, characterized in that, The automatic scheduling method for distribution network maintenance plans according to any one of claims 1 to 9 includes: The multi-source data access module is used to preprocess multi-source data to obtain the device's comprehensive data vector; The maintenance task generation module calculates the health degradation index of the equipment based on the comprehensive data vector; constructs a dynamic fault model of the equipment based on the health degradation index and the comprehensive data vector; calculates the equipment failure probability within the planned cycle based on the dynamic fault model; calculates the maintenance priority index of the equipment based on the failure probability; filters candidate maintenance objects according to the priority index; and generates candidate maintenance tasks for the candidate maintenance objects. The multidimensional spatiotemporal constraint modeling module is used to construct multidimensional spatiotemporal constraint models that include constraints related to time, space, resources, power grid operation, security, and business rules. The collaborative optimization orchestration module is used to calculate the comprehensive distance between any two candidate maintenance tasks based on the comprehensive distance function, and to cluster and merge the candidate maintenance tasks into task groups based on the comprehensive distance. The initial maintenance plan generation module is used to construct multi-objective optimization functions for each task group, iteratively solve them within the scope of the multi-dimensional spatiotemporal constraint model, and generate an initial maintenance plan. The safety verification and scheme modification module is used to perform power grid safety verification on the initial maintenance plan. If the verification fails, iterative modification or infeasibility processing is performed until any of the following termination states are reached: all tasks pass the verification, some tasks are postponed and the remaining tasks pass the verification, or an infeasibility result list is output. The corresponding output is the formal maintenance plan, the list of postponed tasks, or the list of infeasibility results. The local rolling optimization module is used to monitor changes in external conditions and perform local rolling optimization on affected tasks and their related tasks when executing the final maintenance plan.

Images