A method for optimizing the status perception and digital intelligent maintenance process of primary equipment in substations

By constructing a multi-source state perception and maintenance process digital twin for primary substation equipment, the system dynamically matches equipment degradation trends with grid constraints, optimizes the maintenance process, solves the rigidity problem of the maintenance process, and realizes efficient and flexible maintenance plan generation and execution, thereby improving maintenance quality and efficiency.

CN122472741BActive Publication Date: 2026-11-10YANBIAN ELECTRICAL BUREAU
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202610604647.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-05-06
Publication Date
2026-11-10
Estimated Expiration
2046-05-06

AI Technical Summary

Technical Problem

Existing technologies cannot dynamically adapt to equipment degradation trends and power grid supply constraints during the maintenance of primary substation equipment. This results in rigid maintenance procedures, making it impossible to complete maintenance within the limited power outage window, or causing maintenance delays that affect power supply. Furthermore, there are issues such as time conflicts between multiple shifts and equipment status deviations during the maintenance process.

Method used

A multi-source state perception and maintenance process digital twin of substation primary equipment is constructed. Through multi-dimensional modeling, data fusion and joint optimization calculation, differentiated maintenance window optimization schemes are generated. The entire process is virtually simulated and pre-verified before and after maintenance, and the maintenance process is adjusted and optimized in real time to dynamically match equipment status and grid constraints.

Benefits of technology

It enables efficient completion of maintenance within a limited power outage window, reduces maintenance rework rate and the risk of power outage delay, improves maintenance quality and efficiency, reduces the recurrence of similar faults, and forms a full-process digital and intelligent closed-loop management and control system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122472741B_ABST
    Figure CN122472741B_ABST
Patent Text Reader

Abstract

This invention relates to the field of substation primary equipment condition monitoring technology, and discloses a method for substation primary equipment condition perception and intelligent digital maintenance process optimization. By constructing a three-dimensional digital twin that includes equipment condition evolution, maintenance operation procedures, and grid operation constraints, and integrating grid supply guarantee constraints to perform joint optimization calculations, it achieves dynamic matching of deterioration trends and power outage windows. Based on window duration constraints, it adaptively trims procedures and merges parallel operations, compressing the standard 5-hour maintenance into a 3-hour window. At the same time, it performs a full-process virtual simulation before maintenance to avoid procedure deviations, timing conflicts, resource mismatches, and safety blind spots in advance. During maintenance, it dynamically corrects procedures and resource allocation in real time based on the actual root causes of faults on site, forming an intelligent digital closed loop of perception, diagnosis, decision-making, execution, and evaluation. This effectively solves the problems of difficult maintenance during power supply guarantee windows and difficult on-site deviation adjustment, significantly improving maintenance efficiency and quality.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of substation primary equipment condition monitoring technology, and in particular to a method for optimizing the condition perception and digital intelligent maintenance process of substation primary equipment. Background Technology

[0002] With the continuous expansion of the power grid and the increasing demands for power supply reliability, the operation and maintenance management of primary substation equipment is gradually shifting from the traditional periodic maintenance model to a condition-based maintenance model. In the traditional power system operation and maintenance system, the condition assessment of primary substation equipment mainly relies on periodic preventive tests and offline monitoring data. Maintenance plans are often based on fixed time periods or historical fault statistics of equipment. In recent years, Internet of Things (IoT) technology, smart sensor technology, and digital twin technology have been widely used in power systems. By deploying various types of online monitoring devices on primary substation equipment, real-time data acquisition of equipment operating status has been achieved. At the same time, the widespread adoption of data communication technology enables the efficient transmission of massive amounts of status data to centralized monitoring platforms or cloud data centers. Based on the equipment condition assessment results, combined with multi-dimensional factors such as grid risk, maintenance resources, and maintenance costs, differentiated maintenance strategies and process optimization have been developed, which has become an important development direction for improving the quality and efficiency of substation equipment operation and maintenance.

[0003] For example, Chinese Patent CN120317861B discloses a cloud-based intelligent management and control system and method for the entire process of substation maintenance. This invention solves the problem of positioning drift and failure caused by strong electromagnetic fields in substations in existing technologies by integrating a multi-source positioning compensation module and combining UWB positioning, visual SLAM, and electromagnetic interference sensors to dynamically calculate compensation coefficients. It achieves millimeter-level precision real-time navigation in high-voltage equipment areas, significantly improving the accuracy of maintenance path planning and operational safety. Furthermore, this invention uses a dynamic access control module to dynamically adjust equipment operation permission levels based on biometric credibility parameters and real-time electromagnetic environment credibility scores, solving the problem of traditional static role allocation in complex electromagnetic environments.

[0004] The risk of excessive or unreasonable restrictions is mitigated, enabling flexibility and precision in permission allocation and effectively reducing the probability of misoperation in energized areas.

[0005] The above-mentioned and existing related technologies often have the following drawbacks:

[0006] 1. In real-world scenarios where new power systems have a high proportion of renewable energy integration, core primary equipment such as main transformers, high-voltage circuit breakers, and GIS in 220kV and above hub substations are continuously impacted by peak summer demand, sharp load increases during the evening rush hour, and drastic fluctuations in wind power output at night during critical periods such as power supply for important events. This leads to frequent short-term nonlinear and abrupt deterioration in the insulation and mechanical characteristics of the equipment. During the power supply guarantee period, the power grid can only provide a very narrow rigid outage window. At the same time, there are situations where multiple devices simultaneously experience deterioration warnings and require coordinated maintenance. This results in existing technologies only being able to output static and rigid standardized maintenance procedures. It is impossible to predict the optimal maintenance window period based on the dynamic evolution trend of equipment deterioration and match the power grid supply guarantee constraints, nor can it adaptively trim unnecessary process nodes and merge parallel procedures based on the limited outage duration. This leads to either the inability to arrange maintenance during the power supply guarantee period, resulting in equipment operating with defects and eventually tripping due to faults, causing large-scale power outages, or the maintenance outage duration far exceeding the grid's allowable range, affecting renewable energy consumption and power supply to important users.

[0007] 2. Due to the common occurrence of actual internal fault points and root causes of deterioration in substation primary equipment maintenance sites, which significantly deviate from the conclusions output by previous online monitoring and status diagnosis, frequent unexpected situations arise during maintenance, such as overlapping work schedules of multiple shifts, mismatch between spare parts and on-site needs, and blind spots in safety measures. This makes it impossible for the existing rigid standardized maintenance process to complete the virtual execution and pre-verification of the entire process and all elements before maintenance, thus failing to avoid problems such as mismatch between maintenance procedures and the actual state of the equipment and work schedule conflicts. Furthermore, it is impossible to dynamically adjust and optimize the maintenance process and procedures based on the actual root causes of faults and unexpected situations found on-site during maintenance. This results in rework, excessive power outage time, substandard maintenance quality, and repeated occurrence of the same type of fault. Summary of the Invention

[0008] The technical problem to be solved by this invention is that the existing technology has the disadvantages of tight power supply window, many field deviations, and rigid process that is difficult to adapt. To this end, we propose a method for optimizing the status perception and digital intelligent maintenance process of substation primary equipment.

[0009] To achieve the above objectives, this application adopts the following technical solution: a method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment, comprising the following steps:

[0010] Step 1: Construct a digital twin of the multi-source state perception and maintenance process of the primary equipment of the substation, and determine the multi-dimensional modeling direction of the digital twin. The multi-dimensional modeling direction includes the equipment state evolution modeling dimension, the maintenance operation procedure modeling dimension, and the power grid operation constraint modeling dimension. Set an initial mirror node for each modeling dimension.

[0011] Step 2: Based on the initial mirror node, perform node expansion and data fusion according to the modeling direction to obtain equipment state evolution data, maintenance operation process data, and power grid operation constraint data. Generate mirror sub-nodes for each modeling dimension, and construct state awareness rules and process optimization rules for each sub-node.

[0012] Step 3: Establish a dynamic matching relationship between equipment status evolution trends and maintenance windows, perform joint optimization calculations that integrate power grid supply constraints, and generate differentiated maintenance window optimization schemes;

[0013] Step 4: Based on the duration constraint of the window period optimization scheme, perform adaptive process trimming and process merging on the mirrored sub-nodes of the maintenance operation process to generate a process-level maintenance process optimization scheme;

[0014] Step 5: Before the maintenance is carried out, a full-process virtual simulation and pre-verification is performed based on the digital twin to identify and avoid problems such as process adaptability deviation, work sequence conflict, resource mismatch, and safety blind spot.

[0015] Step 6: During the maintenance process, based on the real-time synchronization of the digital twin, the maintenance process and procedures are dynamically adjusted and optimized according to the root causes of the on-site faults, and real-time optimization scheduling instructions are generated.

[0016] Step 7: Integrate post-maintenance evaluation feedback data and iteratively optimize digital twin parameters.

[0017] Preferably, in step 1, the initial mirror node of the equipment state evolution modeling dimension carries the online monitoring data of the core primary equipment and its degradation rate assessment model;

[0018] The initial mirror node of the maintenance operation process modeling dimension carries the standard process library, process time and process dependency relationship;

[0019] The initial mirror node of the power grid operation constraint modeling dimension carries the power grid operation mode, power outage window period, and multi-equipment collaborative maintenance constraints during the power supply guarantee period.

[0020] Preferably, in step 3, the joint optimization calculation adopts a multi-objective optimization algorithm, specifically: the degradation trend prediction results output by the mirror sub-node of the equipment state evolution modeling dimension are used as the equipment risk input, and the power outage window constraints output by the mirror sub-node of the power grid operation constraint modeling dimension are used as the boundary of the feasible solution space. The optimization objectives are to minimize the equipment fault risk integral, maximize the window period utilization rate, and minimize the impact on power grid operation, and to solve for the optimal maintenance window period and multi-equipment collaborative maintenance scheme.

[0021] Preferably, in step 3, the joint optimization calculation adopts a maintenance decision optimization module based on reinforcement learning. Specifically, the degradation trend prediction results output by the mirror sub-node of the equipment state evolution modeling dimension, the power outage window constraints output by the mirror sub-node of the power grid operation constraint modeling dimension, the maintenance resource status data obtained in step 2, and the historical decision feedback are used as the state space. The maintenance window period selection, process pruning scheme, and resource allocation scheme are used as the action space. The reward function is to minimize the equipment failure risk and the impact on power grid operation and maximize the window period utilization rate. The maintenance decision scheme is generated by combining offline training and online decision-making.

[0022] Preferably, in step 4, the adaptive process trimming and process merging specifically involve:

[0023] Based on the window duration constraint in the maintenance window period optimization scheme generated in step 3, identify and cut unnecessary processes in the standard process library that do not match the deterioration trend prediction results.

[0024] Based on the process dependencies carried by the initial mirror nodes of the maintenance operation process modeling dimension in step 1, identify processes that can be executed in parallel and establish parallel timing sequences.

[0025] Based on the maintenance operation process data in step 2, the spare parts inventory data and personnel team configuration data are obtained to generate a trimmed process sequence, ensuring that the total process time does not exceed the window time constraint.

[0026] Preferably, in step 5, the virtual execution and pre-verification of the entire process specifically includes:

[0027] Input the process information from the process-level maintenance process optimization scheme generated in step 4 into the virtual simulation engine;

[0028] Based on the 3D model of primary equipment in substations and personnel operation flow simulation technology, the maintenance process is fully simulated during virtual execution and pre-verification.

[0029] During the simulation, the system automatically identifies compatibility discrepancies, timing conflicts, resource mismatches, and security blind spots, and generates adjustment plans.

[0030] Preferably, in step 5, the virtual execution and pre-verification adopt a distributed parallel simulation architecture, specifically as follows:

[0031] The maintenance process sequence in the process-level maintenance process optimization scheme generated in step 4 is divided into independent simulation task blocks according to equipment units and work areas.

[0032] Distribute simulation task blocks to multiple computing nodes for parallel simulation;

[0033] Nodes synchronize and share resource status through message queues;

[0034] After the simulation of each node is completed, the results are merged to generate a full-scene virtual simulation report.

[0035] Preferably, in step 6, the dynamic adjustment and optimization of the maintenance process and procedures specifically includes:

[0036] During the maintenance process, the system receives real-time data on the root causes of faults and equipment status deviations from on-site personnel via mobile terminals and synchronizes this data to the digital twin.

[0037] The digital twin is based on the real root cause of the fault and refers to the process dependency relationship carried by the initial mirror node of the maintenance operation process modeling dimension in step 1 to generate the adjusted process sequence.

[0038] Based on the window duration in the maintenance window optimization plan generated in step 3 and the time already completed for the current process, the remaining window time is calculated, the sequence of subsequent processes is automatically rearranged, and real-time optimization scheduling instructions are generated and pushed to the field operation terminal.

[0039] Preferably, in step 6, the cloud-edge collaborative decision-making architecture is dynamically adjusted to replace the centralized cloud decision-making architecture, specifically as follows:

[0040] Deploying edge computing nodes at the maintenance site will bring the real-time synchronization and dynamic correction capabilities of the digital twin to the work site.

[0041] Edge nodes collect operational data in real time and make dynamic adjustment decisions locally.

[0042] The cloud-based digital twin periodically receives synchronized data from edge nodes, performs global optimization analysis, and sends the optimized model parameters back to the edge nodes.

[0043] Preferably, in step 7, the iterative optimization specifically involves:

[0044] Collect data on the actual time consumption, actual resource usage, on-site fault and diagnostic deviation records, and comparison data between virtual simulation and actual execution deviations for each process during the maintenance implementation;

[0045] Based on the collected feedback data, update the fault mode library weight parameters of the mirror sub-nodes in the equipment state evolution modeling dimension, update the process time library and process dependency parameters of the mirror sub-nodes in the maintenance operation process modeling dimension, and update the window period selection strategy weight parameters of the mirror sub-nodes in the power grid operation constraint modeling dimension.

[0046] The technical effects and advantages of this invention are as follows:

[0047] This invention overcomes the limitations of single-dimensional decision-making in existing technologies by constructing a three-dimensional digital twin that includes equipment state evolution, maintenance operation procedures, and grid operation constraints. It integrates constraints such as equipment degradation trends, maintenance procedure attributes, grid operation modes, outage windows, and load forecasting into a unified modeling framework, providing a complete data foundation for multi-constraint maintenance decisions. Based on this, by establishing a dynamic matching relationship between equipment state evolution trends and maintenance windows, it performs joint optimization calculations that integrate grid supply guarantee constraints, constructing a comprehensive optimization index R for the maintenance process. This quantitatively evaluates and collaboratively optimizes multiple factors such as equipment degradation rate, health index decay, window utilization rate, impact of renewable energy fluctuations, and maintenance resource availability, achieving a dynamic balance between equipment risk and grid constraints. This avoids the dilemma of being unable to perform maintenance during the power guarantee period, leading to fault tripping or excessive outages affecting power supply. Based on window duration constraints, it performs adaptive trimming and process merging of maintenance operation procedures. By identifying unnecessary procedures for trimming, establishing parallel timing sequences for parallelizable procedures, and generating process sequences through spare parts and personnel configuration, it realizes the transformation of standard 5-hour maintenance procedures into... The 3-hour window of process optimization significantly improved maintenance efficiency within the limited window. Simultaneously, before maintenance implementation, a full-process virtual simulation and pre-verification were performed using 3D models and personnel movement simulation technology. This proactively identified and mitigated potential problems such as process compatibility deviations, multi-shift cross-operation timing conflicts, spare parts mismatches, and safety blind spots, significantly reducing maintenance rework rates and the risk of exceeding power outage duration limits. During maintenance, real-time feedback data from the field was received via mobile terminals and synchronized to the digital twin. When the actual root cause of the fault deviated from the initial diagnosis, subsequent process sequences, resource allocation, and operation timings were dynamically adjusted, generating real-time optimized scheduling instructions that were pushed to the field operation terminals. This enabled real-time dynamic correction during maintenance, effectively improving maintenance quality and reducing the recurrence rate of similar faults. Finally, by integrating post-maintenance evaluation feedback data, the parameters of each mirror sub-node of the digital twin were iteratively optimized, updating fault mode library weights, process time libraries, parallel relationship parameters, and window selection strategies. This enabled the system to continuously evolve, forming a fully intelligent closed-loop management system. Attached Figure Description

[0048] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0049] Figure 1 This is a flowchart illustrating the operation of a method for optimizing the status perception and digital maintenance process of primary substation equipment according to the present invention.

[0050] Figure 2 This is a schematic diagram of the cloud-edge collaborative decision-making architecture deployment of the present invention;

[0051] Figure 3 This is a comparison diagram of the dynamic correction timing of the maintenance process of the present invention;

[0052] Figure 4 This is a multi-dimensional correlation scatter matrix diagram of a method for optimizing the status perception and digital maintenance process of primary substation equipment according to the present invention.

[0053] Figure 5 This is a diagram showing the dynamic matching relationship between the degradation rate and the selected maintenance window period in this invention.

[0054] Figure 6 The present invention illustrates the distribution, trend, and correlation coefficients between pairs of variables. Detailed Implementation

[0055] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0056] Example 1: Refer to Figure 1-6 As shown, this invention provides a technical solution: a method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment, comprising the following steps:

[0057] Step S1: Construct a digital twin of the multi-source state perception and maintenance process of the substation primary equipment, determine the multi-dimensional modeling direction of the digital twin, set an initial mirror node for each modeling dimension, and the initial mirror node carries the basic state attribute information of the modeling dimension. The multi-dimensional modeling direction includes the equipment state evolution modeling dimension, the maintenance operation procedure modeling dimension, and the power grid operation constraint modeling dimension.

[0058] In this embodiment, the maintenance scenario of a 220kV hub substation during the peak summer power supply period is used as an example for illustration. First, it is necessary to construct a digital twin of the multi-source state perception and maintenance process of the substation primary equipment. This construction process includes determining the multi-dimensional modeling direction. In this scenario, the multi-dimensional modeling direction is specifically set as the equipment state evolution modeling dimension, the maintenance operation procedure modeling dimension, and the power grid operation constraint modeling dimension. For each modeling dimension, a corresponding initial mirror node must be set. These initial mirror nodes carry the basic state attribute information of each modeling dimension.

[0059] Based on the data types required for each modeling dimension, the equipment state evolution data required for the equipment state evolution modeling dimension includes:

[0060] Online oil chromatography monitoring data of core primary equipment such as main transformers, high-voltage circuit breakers, and GIS; concentrations of characteristic gases such as acetylene, hydrogen, and ethylene; trends in partial discharge; mechanical characteristic parameters; current waveforms, operating times, stroke curves, and temperature field distribution data of opening and closing coils; as well as historical change curves and degradation rate assessment models of the above parameters.

[0061] The maintenance operation process data required for modeling the maintenance operation process includes: a standard process library for various maintenance operations, such as the process sequence for transformer cover lifting maintenance: power outage operation → safety measure arrangement → oil draining → cover lifting → fault location → defect handling → reassembly → oil filling → testing → power supply; standard working hours for each process; a list of required tools and spare parts; personnel qualification requirements; dependencies between processes; safety measure templates, etc.

[0062] The grid operation constraint data required for grid operation constraint modeling includes: grid operation mode during the power supply guarantee period, such as main wiring diagram, load transfer path, power flow limit of key sections, power supply requirements of important users, new energy output forecast curve, available power outage window periods, and power outage window overlap constraints when multiple devices are under coordinated maintenance.

[0063] Step S2: Based on the multi-dimensional modeling direction, perform node expansion and data fusion of the digital twin. According to the data types required by each modeling dimension, obtain equipment status evolution data, maintenance operation process data, and power grid operation constraint data from the corresponding data sources, generate mirror sub-nodes for each modeling dimension, and build dimension-specific status perception rules and process optimization rules for each sub-node.

[0064] In the above-mentioned 220kV hub substation power supply maintenance scenario, after the initialization of the digital twin is completed, it is necessary to expand the nodes and fuse the data based on the determined multi-dimensional modeling direction. Specifically, according to the data types required by each modeling dimension, the corresponding data is obtained from the corresponding data source to support the generation of mirror sub-nodes of each modeling dimension.

[0065] The equipment status evolution data includes: oil chromatography trend data, partial discharge continuous monitoring waveforms, real-time mechanical characteristic parameters, and infrared temperature measurement data obtained from the online monitoring system of power equipment. Among them, oil chromatography data is used to analyze the composition and concentration changes of dissolved gases in the insulating oil inside the transformer, partial discharge data is used to identify the type and development trend of insulation defects, and mechanical characteristic data is used to evaluate the operating mechanism status of switchgear such as circuit breakers.

[0066] The maintenance operation process data includes: historical maintenance records of substations obtained from the production management system (PMS), standardized operation instructions, process time database, personnel team information, spare parts inventory data, etc. Among them, the standardized operation instructions define the standard process sequence and operation specifications for various maintenance operations, the process time database records the standard time consumption of each process, the personnel team information includes the professional qualifications of maintenance personnel and their current task status, and the spare parts inventory data reflects the model, quantity and storage location of various spare parts.

[0067] The power grid operation constraint data includes: the power grid operation mode during the power supply guarantee period, load forecast curves, renewable energy output forecasts, important user lists, and pre-arranged power outage window plans obtained from the dispatch management system (OMS / EMS). Among them, the power grid operation mode determines the load transfer path and system stability risk when equipment is out of power. The load forecast curve is used to judge the power supply margin of the power grid in each period. The renewable energy output forecast is used to identify special periods such as large-scale photovoltaic power generation and wind power fluctuations. The power outage window plan clarifies the allowed equipment outage time period and the maximum outage duration.

[0068] Then, based on the extracted data, corresponding mirrored sub-nodes are generated for the equipment state evolution modeling dimension, maintenance operation procedure modeling dimension, and power grid operation constraint modeling dimension, respectively. For each sub-node, dimension-specific state perception rules and process optimization rules are constructed. These rules will be used for subsequent dynamic evaluation of equipment state and adaptive optimization of maintenance process.

[0069] Step S3: Establish a dynamic matching relationship between equipment state evolution trend and maintenance window period. Based on the degradation trend prediction results output by the mirror sub-node of the equipment state evolution modeling dimension and the power outage window constraint conditions output by the mirror sub-node of the power grid operation constraint modeling dimension, perform joint optimization calculations that integrate power grid supply guarantee constraints to generate differentiated maintenance window period optimization schemes.

[0070] In the aforementioned power supply maintenance scenarios, one of the core issues that digital twins need to address is that during the peak summer power supply period, the acetylene concentration in the main transformer oil chromatogram shows a non-linear increasing trend. The load surge during the midday solar power generation period exacerbates local overheating, while the power grid can only provide a rigid 3-hour power outage window from 2:00 AM to 5:00 AM. Existing technologies can only output static strategies that recommend immediate power outage maintenance, which cannot solve the problem of matching this window period with the equipment degradation trend.

[0071] In this embodiment, the main transformer oil chromatography data is analyzed in real time through the mirrored sub-nodes of the equipment state evolution modeling dimension. Combined with the hybrid neural network model, the evolution trajectory of acetylene concentration in the next 72 hours is predicted, and the equipment health index decay curve is calculated. When the prediction result shows that the acetylene concentration will exceed the warning threshold after 48 hours, the system inputs the deterioration trend data into the joint optimization module.

[0072] Meanwhile, the mirror sub-nodes of the power grid operation constraint modeling dimension extract all available power outage window information during the power supply guarantee period, including 2:00-5:00 AM, 1:30-4:30 AM the next day, and 2:30-5:30 AM the third day, as well as the collaborative maintenance constraints when multiple devices, such as main transformers and high-voltage circuit breakers, simultaneously experience deterioration warnings. These constraints must be completed within the same window or staggered between windows.

[0073] Based on the above inputs, the joint optimization module performs multi-objective optimization calculations, with optimization objectives including: minimizing equipment failure risk, maximizing window period utilization, and minimizing the impact on power grid operation.

[0074] In the aforementioned power supply maintenance scenario, the joint optimization module performs multi-objective optimization based on the comprehensive optimization index R of the maintenance process. The calculation process of R is as follows:

[0075] First, construct a sub-formula for the urgency of equipment risks: ;

[0076] in, The importance weights for equipment are assigned, for example, a weight of 0.9 for main transformers, 0.7 for high-voltage circuit breakers, and 0.6 for GIS bays. For example, the daily increase rate of acetylene concentration in the main transformer oil chromatography is 2.5 ppm / day. This represents the remaining time for the health index threshold. For example, if the current health index is 85 points and the threshold is set to 70 points, the remaining time calculated based on the current rate of deterioration is 6 days. This represents the current cumulative degradation; for example, the total hydrocarbon content in the oil chromatogram has reached 120 ppm. The maximum allowable cumulative degradation is, for example, a total hydrocarbon warning value of 150 ppm as specified in the regulations.

[0077] Secondly, construct the formula for power grid supply adaptability: ;

[0078] in, This is a window utilization correction factor, with values ​​determined according to the power supply guarantee level: 0.8 for Level 1 power supply guarantee and 0.6 for Level 2 power supply guarantee. This refers to the available power outage window duration, for example, a 3-hour outage period approved by the power grid dispatch center from 2:00 AM to 5:00 AM. The expected baseline duration for the window is set to 4 hours. This refers to the current fluctuations in renewable energy output, such as a sudden drop of 50MW in photovoltaic output at midday. The maximum fluctuation range that the system can withstand is calculated to be 80MW based on the power grid in this region.

[0079] Next, construct the maintenance resource availability sub-formula: ;

[0080] in, The grid disturbance risk coefficient is set at 1.2 for areas with high renewable energy penetration and 1.0 for conventional areas. This refers to the number of currently available maintenance teams. For example, there are two maintenance teams available: Substation Maintenance Team 1 and Team 2. This represents the minimum number of work teams required for a given process. For example, overhauling a suspended structure requires two work teams. This refers to the current system load factor, for example, a load factor of 65% during the midday peak photovoltaic power generation period. The upper limit of system load rate is usually set at 85%.

[0081] Then, construct the process compressibility formula: ;

[0082] in, The compression elasticity coefficient for the process is 0.8. Parallelism between processes is defined as the ratio of the number of processes that can be parallelized to the total number of processes. For example, if there are 10 processes in total... This allows for parallel processing of 4 items with a parallelism of 0.4, while the baseline parallelism is set to 0.5. This is the parallelism saturation coefficient, with a value of 5, which controls the steepness of the logistic function.

[0083] Construct the spare parts support formula: ;

[0084] in, The spare parts availability factor is set to 0.6. The quantity of spare parts to be matched is specified, for example, 2 sets of sleeve seals and 1 torque wrench have been matched, totaling 3 items. This represents the total required number of spare parts, for example, 4 items are needed. For spare parts allocation response time, for example, it takes 1.5 hours to transport spare parts from the warehouse to the site. The baseline response time is set to 2 hours.

[0085] Construct a sub-formula for diagnostic reliability: ;

[0086] in, The diagnostic risk factor is set to 0.7. The online monitoring system gives a confidence level of 0.85 for on-site fault diagnosis. The baseline confidence level is set to 0.8. The sensitivity coefficient for diagnostic bias is set to 3.

[0087] Constructing the formula for job conflict: ;

[0088] in, This is the conflict coefficient for overlapping operations, with a value of 0.5. This refers to the overlapping area of ​​multiple work groups' work areas. For example, the overlapping area between the work areas of a primary work group and a test work group is 15 square meters. This refers to the total working area, for example, the total area of ​​the maintenance area is 100 square meters. This refers to the process timing deviation, for example, the actual start time being 0.2 hours later than planned. To allow the maximum timing deviation, it is set to 0.5 hours.

[0089] Finally, the main formula for the comprehensive optimization index of the integrated maintenance process is as follows: ;

[0090] The optimized output is to select the window period from 2:30 to 5:30 a.m. on the third day, and to merge and arrange the main transformer maintenance process and the circuit breaker maintenance process in parallel within this window. The circuit breaker maintenance is completed during the same period when the main transformer is out of power, thus solving the problem of coordinated maintenance of the two devices within a single window.

[0091] This optimization scheme uses a digital twin to jointly predict the dynamic evolution trend of equipment degradation and the power grid supply constraints, solving the problem that existing technologies cannot match the optimal maintenance timing within a rigid window period.

[0092] Step S4: Based on the window duration constraint in the maintenance window period optimization scheme, perform adaptive process trimming and process merging on the mirrored sub-nodes of the maintenance operation process modeling dimension to generate a process-level maintenance process optimization scheme for limited power outage duration.

[0093] In the aforementioned power supply maintenance scenario, the optimized maintenance window is from 2:30 AM to 5:30 AM, with a total duration of only 3 hours. However, the standard maintenance procedure library for this main transformer requires 5 hours for a complete overhead cover maintenance, including 0.5 hours for power outage operation, 0.5 hours for safety measure setup, 1.0 hour for oil drainage, 0.5 hours for overhead cover installation, 1 hour for fault location and handling, 0.5 hours for reassembly, 0.5 hours for oil injection, and 0.5 hours for testing and power restoration. Existing technology cannot complete all procedures within 3 hours.

[0094] In this embodiment, the mirrored sub-nodes of the maintenance operation process modeling dimension perform adaptive process pruning and process merging based on the optimized window period constraints:

[0095] First, based on the deterioration root cause localization results output by the mirror sub-nodes of the equipment state evolution modeling dimension, oil chromatography analysis points to high-energy discharge inside the medium-pressure side bushing, the fault point is clearly located, and the system automatically identifies unnecessary procedures. For example, since the fault point has been accurately located inside the bushing, there is no need to conduct a comprehensive inspection of the tap changer, iron core, and windings. Therefore, the comprehensive inspection procedure after lifting the cover can be reduced to a directional inspection of the bushing and leads after lifting the cover.

[0096] Secondly, based on the dependency graph between processes, processes that can be executed in parallel are identified. In standard processes, there is some parallel space between oil draining and safety measure arrangement, between hood and spare parts preparation can be completely parallel, and between reassembly and pre-oil injection test preparation can be parallel.

[0097] Finally, based on spare parts inventory data and personnel team configuration, the system automatically generates a trimmed process sequence, compressing the original 5-hour process into 2 hours and 50 minutes, with a 10-minute buffer reserved to meet the 3-hour window constraint.

[0098] The process sequence after cutting includes: power outage operation (0.5h) → safety measures arrangement and oil drainage in parallel (1h) → hoisting cover and spare parts preparation in parallel (1h) → targeted fault handling (0.5h) → reassembly and oil filling in parallel (0.5h) → testing and power supply (0.5h).

[0099] This process optimization scheme at the process level achieves adaptive process tailoring and process merging based on a limited power outage window, solving the problem that existing technologies output rigid standardized processes and cannot complete maintenance within a limited window period.

[0100] Step S5: Before the maintenance is carried out, the entire process and all elements are virtually executed and pre-verified based on the digital twin. The process-level maintenance process optimization plan is fully simulated in the virtual space to identify and avoid the compatibility deviation between the maintenance process and the actual state of the equipment, the timing conflict of multiple shifts working together, the mismatch between spare parts and on-site needs, and the blind spots in the layout of safety measures.

[0101] In the above-mentioned power supply maintenance scenario, after generating the process optimization plan at the work level, this embodiment starts the virtual execution and pre-verification stage before the maintenance is carried out.

[0102] Specifically, the trimmed process sequence, including the start time, duration, required personnel, required tools and spare parts, and safety requirements for each process, is input into the virtual simulation engine of the digital twin. This engine, based on a high-precision 3D model, a BIM model of the substation's primary equipment, a point cloud model of the work site, and personnel movement simulation technology, completely simulates the entire maintenance process in virtual space.

[0103] During the simulation, the system automatically identified the following potential problems:

[0104] The virtual simulation revealed that, due to the high position of the bushing on the medium-voltage side of the main transformer, the directional fault handling process requires the assistance of an aerial work platform. However, the original process sequence did not include the scheduling time for the aerial work platform. The system automatically inserted the aerial work platform positioning process between the hoisting and directional fault handling processes and added 10 minutes of scheduling time.

[0105] The virtual simulation showed that when the two work teams were conducting parallel pre-refilling and pre-oiling tests, their work areas overlapped, potentially causing insufficient safety distance. The system automatically adjusted the parallel timing, delaying the start time of the test team's work by 20 minutes to ensure that the work areas did not overlap.

[0106] The virtual simulation, which used data from the spare parts management system, revealed a slight difference between the model of the sleeve seal ring in the spare parts warehouse and the one on the field equipment. The system automatically triggered the spare parts verification process, requiring on-site personnel to recheck the seal ring size before lifting the cover and to replenish the spare seal ring.

[0107] The virtual simulation, through 3D spatial analysis, revealed that a grounding wire installation point was obstructed by scaffolding during the safety measure deployment process, preventing it from being installed in the standard position. The system automatically generated an adjustment plan, re-marking the grounding wire installation point at an adjacent location and updating the safety measure deployment diagram.

[0108] Through the above-mentioned virtual execution and pre-verification, this embodiment avoids problems such as the compatibility deviation between the maintenance procedures and the actual state of the equipment, and the conflict of operation sequence before the maintenance is carried out, thus solving the defect of the existing technology that cannot detect and avoid on-site emergencies before maintenance.

[0109] Step S6: During the maintenance process, based on the real-time collected equipment status feedback data and on-site operation execution data, the digital twin is synchronized and dynamically corrected in real time. According to the real root causes of faults and emergencies found on-site, the maintenance process and procedures are dynamically adjusted and optimized, and real-time optimized scheduling instructions are generated during the maintenance process.

[0110] In the aforementioned power supply maintenance scenario, the maintenance officially commenced at 2:30 AM. Throughout the process, the digital twin maintained real-time synchronization with the on-site operational data.

[0111] After the cover removal process was completed, the on-site maintenance personnel found that the actual fault point differed from the previous diagnosis: oil chromatography analysis pointed to internal discharge in the bushing, but after the cover was actually removed, creepage marks were found on the surface of the bushing insulator, while the internal conductor was intact. At the same time, it was found that the sealing gasket at the connection between the bushing and the transformer body showed signs of aging and cracking.

[0112] The discrepancy between the actual root cause of the fault and the initial diagnostic conclusion was detected in real time via a digital twin input from a mobile terminal, and the system immediately initiated a dynamic correction process:

[0113] The mirror sub-nodes of the equipment state evolution modeling dimension receive fault images and descriptions from the field and update the fault type from high-energy discharge inside the bushing to creepage on the bushing surface and aging of the sealing gasket.

[0114] The mirrored sub-nodes of the maintenance operation process modeling dimension automatically adjust subsequent processes based on the corrected fault type. The originally planned targeted fault handling process has been changed to insulation surface treatment and gasket replacement. Due to the addition of the gasket replacement process, the system recalculates the remaining time and assesses whether the window period has been exceeded.

[0115] The system detects that a special tool and sealant are needed to replace the sealing gasket. It automatically checks the current status of the tool in the spare parts library and generates a coordination instruction to notify another team to use the tool within 20 minutes and hand it over to the current work point.

[0116] Based on the remaining window time and the duration of the new process, the system reschedules the subsequent process sequence, further compresses the oil injection and test preparation in parallel, and activates a buffer early warning mechanism to prompt on-site personnel to speed up the work pace.

[0117] Through the above-mentioned real-time synchronization and dynamic correction, this embodiment realizes the dynamic adjustment and optimization of maintenance process and procedures based on the actual root causes of faults and emergencies found on site during the maintenance process, which solves the problem that the existing technology cannot adjust during the process, resulting in maintenance rework and excessive power outage time.

[0118] Step S7: Integrate the evaluation feedback data after the maintenance is carried out, iteratively optimize the parameters of each mirror sub-node of the digital twin, update the rule parameters of the equipment state evolution model, maintenance operation procedure model, and power grid operation constraint model, and generate the iteratively optimized digital twin.

[0119] After the maintenance is completed, this embodiment collects data on the entire maintenance process, including: actual time spent on each process, actual resource usage, on-site fault and diagnostic deviation records, comparison of virtual simulation and actual execution deviations, and actual usage time during the window period.

[0120] Based on this feedback data, the system optimizes the parameters of the mirrored child nodes for each modeling dimension:

[0121] The fault characteristics of this bushing surface creepage were added to the fault mode library, and the weight parameters of the fault diagnosis model were updated to improve the accuracy of subsequent identification of similar faults.

[0122] The process sequence and time data after this trimming will be incorporated into the standard time library. The parallel relationship parameters of directional fault handling and spare parts preparation after the lifting of the cover in the process dependency graph will be updated to optimize the process merging logic for similar scenarios in the future.

[0123] The data from the early morning window and load curve used in this study were incorporated into the optimization model to update the weight parameters of the window selection strategy and improve the accuracy of window matching during future power supply guarantee periods.

[0124] Through the aforementioned iterative optimizations, the digital twin achieves continuous self-evolution, constantly improving the accuracy and efficiency of subsequent maintenance process optimization.

[0125] Example 2: Step S1: Based on Example 1, a maintenance decision optimization module based on reinforcement learning is introduced. The equipment degradation trend prediction results, power grid operation constraints, maintenance resource status, and historical decision feedback are used as the state space of the reinforcement learning agent. The maintenance window selection, process pruning scheme, and resource allocation scheme are used as the action space. The reward function is to minimize the equipment failure risk and the impact on power grid operation and maximize the window utilization rate. The globally optimal maintenance decision scheme is generated by combining offline training and online decision-making.

[0126] In this optimized embodiment, the scenario of coordinated maintenance of five 220kV hub substations during the peak summer power supply period is used as an example for illustration.

[0127] The joint optimization module in the basic embodiment adopts a deterministic optimization algorithm. In complex scenarios with multiple devices working together for maintenance and multiple constraints coupled together, it suffers from problems such as solution space explosion and difficulty in convergence. This optimization embodiment introduces a reinforcement learning agent to solve this technical problem.

[0128] The degradation trend prediction curves of 12 core primary equipment in 5 substations, the health index change rate in the next 72 hours, the power grid operation mode of each substation, the main wiring topology, the load transfer capacity, and all available outage windows during the power supply guarantee period are encoded into a high-dimensional state vector. Each window includes the start time, maximum duration, maintenance resource status, location and availability of 3 maintenance teams, and inventory and dispatch time of each type of spare parts in the spare parts warehouse.

[0129] The maintenance actions performed on each piece of equipment within each power outage window are defined as actions, including: whether maintenance is scheduled, which simplification plan is adopted (standard plan / simplification plan A / simplification plan B), resource allocation plan, which work team is used, and which spare parts are allocated. The action space is approximately 10^ 4 Magnitude.

[0130] The reward function has two dimensions:

[0131] Positive Reward: Successfully completing maintenance and eliminating potential equipment malfunctions within the window period;

[0132] Negative incentives include: equipment failure and tripping during the window period, window period exceeding the time limit, impact on power supply to important users, and waste of maintenance resources.

[0133] Based on three years of historical maintenance data and power grid operation data, a simulation environment was built for offline training. After 20,000 rounds of training, the agent's decision success rate in the test scenario improved from 67% to 89% in terms of completing maintenance within the window period without failure.

[0134] When deployed online, the agent runs once every 4 hours, outputting decision actions based on the real-time state space. When the equipment degradation trend changes abruptly or the power grid operation mode is adjusted, the agent can recalculate and output an updated decision scheme within 5 minutes.

[0135] This optimized embodiment achieves global optimization of maintenance decisions under multiple constraints through reinforcement learning, solving the efficiency and accuracy problems of the deterministic optimization algorithm in the basic embodiment in multi-device collaborative scenarios.

[0136] Step S2: Based on Example 1, a maintenance process-level reasoning module based on knowledge graph is introduced to construct an ontology knowledge graph that includes equipment structure, fault modes, maintenance processes, safety measures, and resource requirements. Based on the real root causes of faults reported from the maintenance site, a process-level maintenance plan is generated through graph neural network reasoning, realizing refined decision-making from fault and countermeasure mapping to fault and process path.

[0137] In this embodiment, a scenario in which a fault diagnosis deviation occurs during the maintenance of a high-voltage circuit breaker in a substation is used as an example for illustration.

[0138] In Example 1, the adjustment of the maintenance procedure relies on a preset procedure library and human experience judgment. When the actual root cause of the fault deviates significantly from the diagnosis conclusion, the adjustment logic is rather mechanical.

[0139] This embodiment constructs a circuit breaker maintenance knowledge graph, which includes the following entity types and relationship types:

[0140] Entity types: Equipment components, such as arc-extinguishing chambers, operating mechanisms, springs, coils, auxiliary contacts; Fault modes, such as arc-extinguishing chamber erosion, mechanism jamming, spring fatigue, coil burnout, contact oxidation; Fault characteristics, such as abnormal opening and closing times, current waveform distortion, abnormal mechanical vibration, reduced insulation resistance; Maintenance procedures, such as disassembly, inspection, replacement, adjustment, and testing; Tools, such as torque wrenches, multimeters, characteristic testers; Safety measures, such as grounding wire installation, fencing, and implementation of two-ticket procedures.

[0141] Relationship types: Component-Fault Mode (Occurrence Relationship), Fault Characteristic-Fault Mode (Characteristic Relationship), Fault Mode-Maintenance Procedure (Resolution Relationship), Maintenance Procedure-Tools (Usage Relationship), Maintenance Procedure-Safety Measures (Dependency Relationship), Procedure-Procedure (Pre-process / Post-process / Parallel Relationship).

[0142] When the inspection site reports that the actual root cause of the fault is spring fatigue in the operating mechanism rather than coil burnout as previously diagnosed, the system inputs the spring fatigue entity into the graph neural network (GNN) inference module. The inference module then performs multi-hop reasoning in the knowledge graph using the graph neural network, automatically generating the following process path:

[0143] Disassemble the operating mechanism housing → inspect spring deformation and elasticity → replace the spring → adjust the mechanism stroke → test the opening and closing characteristics → restore the housing.

[0144] Meanwhile, the reasoning module identifies the tools and equipment required for the process path through relational propagation, such as spring-specific disassembly and assembly tools, stroke measuring instruments, and safety measure mechanisms for energy storage and release confirmation and grounding to prevent malfunctions. It also automatically generates a resource application list and safety measure tickets.

[0145] This embodiment uses knowledge graphs and graph neural network reasoning to achieve refined decision-making from the root cause of the fault to the process-level maintenance plan, solving the technical problem in the basic embodiment that process adjustment relies on a preset rule base and is difficult to deal with complex fault combinations.

[0146] Example 3: Step S1: As an alternative to the joint optimization calculation in the basic example, a hybrid algorithm based on particle swarm optimization and constraint satisfaction problem is used to jointly optimize and solve the maintenance window period and process reduction scheme.

[0147] In this embodiment, the annual maintenance plan scheduling scenario of a converter station is used as an example for illustration.

[0148] When the maintenance task is small in scale, with ≤5 units of equipment in a single station and relatively clear constraints, a hybrid algorithm of particle swarm optimization (PSO) and constraint satisfaction problem (CSP) can be used as an alternative to reduce the training overhead required for reinforcement learning.

[0149] The maintenance window selection problem is modeled as CSP, and the process pruning and resource allocation problem is modeled as PSO. The CSP module is responsible for enumerating all feasible window combinations under the constraints of power grid operation, window availability, equipment outage sequence, and safety rules, and outputting candidate solutions. The PSO module uses the candidate solutions as the initial population to perform particle swarm optimization on the process pruning scheme and resource allocation scheme.

[0150] The position vector of each particle is represented as [window index, device 1 clipping scheme index, device 2 clipping scheme index, ..., resource allocation scheme index], and the velocity vector represents the step size of change in the corresponding dimension.

[0151] Fitness function: The fitness function includes three sub-items: equipment risk score, window period utilization rate, and resource balance, which are consistent with the optimization objectives in the basic embodiment.

[0152] Convergence condition: Optimization stops when the fitness change of particles over 20 consecutive generations is less than the threshold (1%) or when the maximum number of iterations (500 generations) is reached.

[0153] In a converter station maintenance scenario with 5 devices, the hybrid algorithm converges within 30 seconds, and the output scheme is consistent with the reinforcement learning scheme, but no offline training process is required, making it suitable for scenarios with a small task scale.

[0154] This embodiment provides an alternative implementation of joint optimization computation in the basic embodiment, expanding the applicability of the technical solution.

[0155] Step S2: As an alternative to virtual execution and pre-verification in the basic embodiment, a distributed parallel simulation architecture based on digital twins is adopted. The maintenance process sequence is divided into simulation task blocks that can be executed in parallel. The virtual simulation process is accelerated through multi-node parallel computing to meet the real-time requirements of ultra-large-scale maintenance scenarios.

[0156] In this variant embodiment, the scenario of an annual comprehensive overhaul of an ultra-high voltage substation, involving 23 pieces of equipment including main transformers, high-voltage reactors, GIS, and circuit breakers, with 5 professional teams working simultaneously, is used for illustration.

[0157] In the basic embodiment, the single-machine simulation engine takes more than 2 hours to complete a full simulation when dealing with maintenance scenarios involving more than 15 devices and 200 procedures, which cannot meet the timeliness requirements for dynamic adjustment of maintenance plans.

[0158] This embodiment adopts a distributed parallel simulation architecture, which includes the following components:

[0159] Task decomposer: The maintenance process sequence is divided into independent simulation task blocks according to equipment units, such as main transformer unit, GIS unit, high-voltage reactor unit, circuit breaker unit and work area, such as area A, area B and area C. Each task block contains the timing, resource and safety measures information of all processes in the area.

[0160] Node Allocator: Distributes simulation task blocks to multiple computing nodes, a total of 8 computing nodes. Each node is configured with 4 CPU cores and 16GB of memory. The allocation strategy takes into account the dependencies between task blocks. For example, if resource conflicts in cross-regional operations require synchronous simulation, task blocks with resource dependencies will be allocated to the same node or a communication channel between nodes will be established.

[0161] Parallel simulation engine: Each node runs virtual simulation independently, and nodes synchronize the status of shared resources for cross-regional operations through message queues, such as sharing cranes and experimental instruments. The synchronization frequency is set to once every 10 minutes of simulation to balance computational efficiency and status consistency.

[0162] Results Aggregator: After each node completes its simulation, the results are submitted to the aggregator, which merges them to generate a full-scenario virtual simulation report. The conflict detection module performs joint analysis on the results of all nodes to identify cross-regional job timing conflicts and resource competition issues.

[0163] In the ultra-high voltage maintenance scenario involving 23 devices, the distributed parallel simulation architecture reduced the complete simulation time from 2 hours to 12 minutes, meeting the real-time requirements.

[0164] This variant embodiment provides an alternative to the virtual execution and pre-verification in the basic embodiment, solving the real-time simulation bottleneck problem in large-scale maintenance scenarios through distributed parallel computing.

[0165] Step S3: As an alternative to dynamically adjusting and optimizing the maintenance process in the basic embodiment, a real-time feedback and cloud-edge collaborative decision-making architecture based on edge computing is adopted. Edge computing nodes are deployed at the maintenance site to bring the real-time synchronization and dynamic correction capabilities of the digital twin down to the work site, achieving millisecond-level response and network interruption resume capability.

[0166] In this variant embodiment, the maintenance scenario of a substation in a remote area is used as an example for illustration.

[0167] In the basic embodiment, the digital twin is deployed in the cloud. Real-time feedback data from the maintenance site needs to be transmitted to the cloud via the network for decision-making, which poses the risk of network latency and network outages. This problem is even more prominent in substations in remote areas.

[0168] This embodiment adopts a cloud-edge collaborative architecture:

[0169] An industrial-grade edge computing server is deployed at the maintenance site, equipped with an 8-core ARM processor, 32GB of memory, and 1TB of solid-state storage. It is pre-installed with a lightweight digital twin engine containing core inference modules for equipment state evolution models and maintenance operation process models. Edge nodes are connected to on-site sensors, mobile terminals, and smart tools via wired / wireless networks to collect operation data in real time.

[0170] The edge node and the cloud digital twin adopt an incremental synchronization mechanism. When the network is normal, the edge node compresses and synchronizes the field data to the cloud every 5 minutes. When the network is interrupted, the edge node caches the data locally and automatically resumes the transmission after the network is restored. The local cache capacity of the edge node supports 48 hours of continuous operation data storage.

[0171] When there is a discrepancy between the actual root cause of the fault reported on-site and the diagnostic conclusion, the edge node makes dynamic adjustment decisions locally with a response time of less than 5ms. The decision content includes: process adjustment instructions, resource allocation instructions, and safety measure update instructions. The instructions are pushed to the mobile terminals of operators and smart tools via on-site industrial Wi-Fi.

[0172] After receiving synchronized data from edge nodes, the cloud-based digital twin performs global optimization analysis and distributes the optimized model parameters, such as process dependency weights and resource allocation strategies, to the edge nodes to achieve continuous model evolution.

[0173] In the maintenance practice of substations in remote areas, the cloud-edge collaborative architecture ensured that the maintenance work was not affected even after three network outages with a total outage duration of 45 minutes, and the response time of dynamic adjustment commands remained stable at 3-5ms.

[0174] This embodiment provides an alternative to the dynamic adjustment and optimization of the maintenance process in the basic embodiment, which solves the problem of real-time decision reliability in unstable network scenarios by sinking edge computing.

[0175] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment, characterized in that, Includes the following steps: S1: Construct a multi-source state perception and maintenance process digital twin for primary substation equipment, and determine the multi-dimensional modeling direction of the digital twin. The multi-dimensional modeling direction includes equipment state evolution modeling dimension, maintenance operation procedure modeling dimension, and power grid operation constraint modeling dimension. Set an initial mirror node for each modeling dimension. The initial mirror node of the equipment state evolution modeling dimension carries the online monitoring data of the core primary equipment and its degradation rate assessment model; the initial mirror node of the maintenance operation procedure modeling dimension carries the standard procedure library, procedure time and procedure dependency relationship; the initial mirror node of the power grid operation constraint modeling dimension carries the power grid operation mode during the power supply guarantee period, power outage window time period, and multi-equipment collaborative maintenance constraints. S2: Based on the initial mirror node, perform node expansion and data fusion according to the modeling direction to obtain equipment state evolution data, maintenance operation process data, and power grid operation constraint data. Generate mirror sub-nodes for each modeling dimension, and construct state awareness rules and process optimization rules for each sub-node. S3: Establish a dynamic matching relationship between equipment status evolution trend and maintenance window period. By constructing and optimizing the comprehensive optimization index R of maintenance process, perform joint optimization calculation with integrated power grid supply guarantee constraints, and generate differentiated maintenance window period optimization schemes. S4: Based on the duration constraint of the window period optimization scheme, adaptive process trimming and process merging are performed on the mirror sub-nodes of the maintenance operation process to generate a process-level maintenance process optimization scheme. The adaptive process trimming and process merging are specifically as follows: Based on the window duration constraint in the maintenance window period optimization scheme generated in S3, unnecessary processes in the standard process library that do not match the deterioration trend prediction results are identified and trimmed; based on the process dependencies carried by the initial mirror nodes of the maintenance operation process modeling dimension in S1, processes that can be executed in parallel are identified and parallel timing sequences are established; based on the maintenance operation process data in S2, the spare parts inventory data and personnel team configuration data are obtained to generate the trimmed process sequence, ensuring that the total process duration does not exceed the window duration constraint. S5: Before the maintenance is carried out, a full-process virtual simulation and pre-verification is performed based on the digital twin to identify and avoid problems such as process adaptability deviation, work sequence conflict, resource mismatch, and safety blind spot. S6: During the maintenance process, based on the real-time synchronization of the digital twin, the maintenance process and procedures are dynamically adjusted and optimized according to the root causes of on-site faults, and real-time optimization scheduling instructions are generated. The dynamic adjustment and optimization of the maintenance process and procedures are specifically as follows: During the maintenance process, the actual root causes of faults and equipment status deviation data fed back by on-site operators are received in real time through mobile terminals and synchronized to the digital twin; the digital twin generates an adjusted procedure sequence based on the actual root causes of faults and with reference to the procedure dependencies carried by the initial mirror nodes of the maintenance operation procedure modeling dimension in S1; the remaining window time is calculated based on the window duration in the maintenance window period optimization scheme generated in S3 and the time already completed for the current procedure, the subsequent procedure sequence is automatically rearranged, real-time optimization scheduling instructions are generated and pushed to the on-site operation terminal; S7: Integrate post-maintenance evaluation feedback data and iteratively optimize digital twin parameters.

2. The method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment according to claim 1, characterized in that: In S3, the joint optimization calculation adopts a multi-objective optimization algorithm, specifically: the degradation trend prediction results output by the mirror sub-node of the equipment state evolution modeling dimension are used as the equipment risk input, and the power outage window constraints output by the mirror sub-node of the power grid operation constraint modeling dimension are used as the boundary of the feasible solution space. The optimization objectives are to minimize the equipment fault risk integral, maximize the window period utilization rate, and minimize the impact on power grid operation, and to solve for the optimal maintenance window period and multi-equipment collaborative maintenance scheme.

3. The method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment according to claim 1, characterized in that: In S3, the joint optimization calculation adopts a maintenance decision optimization module based on reinforcement learning. Specifically, it uses the degradation trend prediction results output by the mirror sub-node of the equipment state evolution modeling dimension, the power outage window constraints output by the mirror sub-node of the power grid operation constraint modeling dimension, the maintenance resource status data obtained in S2, and historical decision feedback as the state space, and the maintenance window period selection, process pruning scheme, and resource allocation scheme as the action space. The reward function is to minimize the equipment failure risk and the impact of power grid operation and maximize the window period utilization rate. The maintenance decision scheme is generated by combining offline training and online decision-making.

4. The method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment according to claim 1, characterized in that: In S5, the full-process virtual simulation and pre-verification specifically involves: inputting the process information from the process-level maintenance process optimization scheme generated in S4 into the virtual simulation engine; based on the three-dimensional model of the substation primary equipment and personnel operation flow simulation technology, fully simulating the maintenance process during virtual execution and pre-verification; and automatically identifying adaptation deviations, timing conflicts, resource mismatches, and safety blind spots during the simulation process, and generating adjustment schemes.

5. The method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment according to claim 1, characterized in that: In S5, the full-process virtual simulation and pre-verification adopts a distributed parallel simulation architecture, specifically: the maintenance process sequence in the process-level maintenance process optimization scheme generated in S4 is divided into independent simulation task blocks according to equipment units and work areas; the simulation task blocks are allocated to multiple computing nodes for parallel simulation. Nodes synchronize and share resource status through message queues; after each node completes its simulation, the data is merged to generate a full-scenario virtual simulation report.

6. The method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment according to claim 1, characterized in that: In S6, the dynamic adjustment adopts a cloud-edge collaborative decision-making architecture instead of a centralized cloud decision-making architecture. Specifically, edge computing nodes are deployed at the maintenance site to bring the real-time synchronization and dynamic correction capabilities of the digital twin down to the work site; the edge nodes collect work data in real time and complete dynamic adjustment decisions locally; the cloud-based digital twin periodically receives synchronized data from the edge nodes, performs global optimization analysis, and sends the optimized model parameters to the edge nodes.

7. The method for optimizing the status perception and intelligent digital maintenance process of primary substation equipment according to claim 1, characterized in that: In S7, the iterative optimization specifically involves: collecting the actual time consumption, actual resource usage, on-site fault and diagnostic deviation records, and comparison data of virtual simulation and actual execution deviations for each process during maintenance implementation; based on the collected feedback data, updating the fault mode library weight parameters of the mirror sub-nodes in the equipment state evolution modeling dimension, updating the process time library and process dependency parameters of the mirror sub-nodes in the maintenance operation process modeling dimension, and updating the window period selection strategy weight parameters of the mirror sub-nodes in the power grid operation constraint modeling dimension.

Citation Information

Patent Citations

  • Cloud platform-based intelligent control system and method for the entire process of substation maintenance implementation

    CN120317861B

  • Transformer substation operation and maintenance method and system based on digital twinning

    CN118505203A

  • Transformer substation equipment defect automatic identification and positioning method based on digital twinning

    CN120691587A