Generator unit maintenance plan generation method and system based on multi-scale data fusion

Abstract

The present application relates to the field of power system operation and equipment maintenance, and particularly relates to a generator unit maintenance plan generation method and system based on multi-scale data fusion. The method comprises: through integrating multi-time scale load data, equipment parameters and historical records, data alignment and modeling are performed to generate a two-dimensional prediction structure; through constraint processing and resource allocation, a maintenance scheme draft is formed; machine learning technology is used to correlate and mine historical data and extract rules to generate a multi-version alternative scheme sequence; through real-time monitoring of load changes, rule instantiation and target scheme selection are dynamically triggered, and temporary adjustment and optimization of maintenance procedures and standby equipment calling are performed accordingly, and finally an executable plan configuration is generated. The present application effectively improves the adaptive ability and decision-making scientificity of the maintenance plan in response to power grid load fluctuations, and ensures the stable and efficient operation of the power grid during maintenance.

Description

Technical Field

[0001] This invention relates to the field of power system operation and equipment maintenance, and in particular to a method and system for generating generator set maintenance plans based on multi-scale data fusion. Background Technology

[0002] In the field of power system operation and maintenance, existing generator maintenance plans typically rely on traditional periodic maintenance methods, combining equipment health status with grid load demand. Current methods generally formulate maintenance plans by collecting load forecast data, equipment operating status, and historical maintenance records. However, these methods have limitations in effectively addressing grid load fluctuations and equipment status changes. Existing load data often relies on a single time scale or simple forecasting models, making it difficult to comprehensively consider short-term load fluctuations and long-term trends. Equipment operating status data also often focuses on a single dimension of monitoring, neglecting the dynamic changes in equipment performance degradation. Furthermore, adjustments to maintenance plans often lack real-time capability, making it difficult for the system to flexibly adjust to sudden load changes. Existing methods largely rely on manual experience or fixed rules based on historical data, which can easily lead to unreasonable maintenance arrangements or increased equipment failure risks when facing variable grid loads and equipment operating statuses. Therefore, how to generate accurate and flexible maintenance plans through efficient multi-dimensional data fusion and dynamic adjustment under the complex constraints of grid load fluctuations and equipment status changes is a key technological challenge. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for generating generator set maintenance plans based on multi-scale data fusion, comprising: Acquire short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates, and historical maintenance records. Perform clock synchronization and equipment identifier mapping, scale alignment and missing measurement filling and time anchoring, and two-dimensional modeling processing to generate a two-dimensional prediction structure. The execution constraint archiving and section mapping, feasible time period screening and output reduction calculation, resource allocation and process arrangement and conflict resolution are carried out to generate a draft structure of the maintenance adaptation plan; The process involves correlation mining through comparative analysis of historical maintenance data and current draft maintenance plans, parameter learning using machine learning or deep learning algorithms, rule extraction by analyzing recurring patterns in historical maintenance records, combined evaluation of alternative solutions based on extraction rules and parameters, and version packaging for version management of each solution, resulting in a multi-solution sequence structure including different versions of maintenance solutions. The system performs the following functions: monitoring real-time load changes and matching them with preset load fluctuation patterns; instantiating rules from the strategy library templates to apply to actual load fluctuations and equipment status changes; selecting target schemes based on real-time load data and strategy library template rules; quickly adjusting the order and execution method of maintenance procedures based on the target selection results; dynamically adjusting the activation order of standby equipment; compressing maintenance procedures and execution time; merging adjusted plans; updating parameters in the system; recording and tracking the adjustment process; and generating a plan write-back configuration structure that includes the adjusted maintenance plan and standby equipment call plan.

[0004] Furthermore, the short-term load sequence, weekly / monthly load sequence, equipment operating parameter sequence, real-time load sequence, unit constraint benchmarks, strategy library templates, and historical maintenance records include: Short-term load sequences contain forecast data on load demand over a period of several hours or a day; Weekly / monthly load series contain forecast data for several weeks or months; The equipment operating parameter sequence includes the unit's start-up and shutdown times, output records, equipment efficiency, and performance degradation data; Real-time load sequences contain instantaneous load data obtained from the power grid dispatching system; The unit's constraint benchmarks include minimum technical output and maximum permissible output reduction. The strategy library templates include options for adjusting maintenance plans and activating backup equipment in response to load changes. Historical maintenance records include maintenance duration, maintenance methods, and load impact.

[0005] Furthermore, the process of generating a multi-scheme sequence structure that includes different versions of maintenance plans also includes: Correlation mining is conducted through comparative analysis of historical maintenance data and current draft maintenance plans; Parameter learning is performed using machine learning or deep learning algorithms. By analyzing recurring patterns in historical maintenance records, we can extract relevant rules. And a combined evaluation of the alternatives based on the extraction rules and parameters.

[0006] Furthermore, generating multi-scheme sequence structures also includes: Version packaging is performed on each solution to generate a multi-solution sequence structure that includes different versions of maintenance solutions.

[0007] Furthermore, the process of generating a plan write-back configuration structure that includes the adjusted maintenance plan and the standby equipment call-up plan includes: Perform real-time load monitoring and match event triggering with preset load fluctuation patterns; Instantiate rules that apply rules from the strategy library templates to actual load fluctuations and equipment status changes; The target selection is based on real-time load data and policy library template rules to select the target scheme.

[0008] Furthermore, the process of generating the plan write-back configuration structure, which includes the adjusted maintenance plan and the standby equipment call-up plan, also includes: Based on the target selection results, quickly adjust the temporary arrangement of the maintenance procedure sequence and execution method; Backup equipment call-up with dynamic adjustment of the activation order of backup equipment; The maintenance process is compressed to shorten the execution time of the maintenance process.

[0009] Furthermore, the process of generating the plan write-back configuration structure, which includes the adjusted maintenance plan and the standby equipment call-up plan, also includes: The adjusted and integrated plans will be merged; Write back the adjusted plan to update the parameters in the system; Audit filing process for recording and tracking the adjustment process.

[0010] Furthermore, the association mining process also includes: Analyze the impact of maintenance tasks on equipment status, load fluctuations, and power grid stability, and identify which types of maintenance activities are closely related to load demand fluctuations.

[0011] Furthermore, the rule extraction process also includes: By analyzing recurring patterns in historical maintenance records, a set of rules is generated for evaluating maintenance plans. These rules include maintenance operations for specific equipment, measures to deal with load fluctuations, or optimization suggestions for resource allocation.

[0012] Furthermore, a generator set maintenance plan generation system based on multi-scale data fusion, applied to any of the methods described above, includes: The data acquisition module is used to acquire short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates, and historical maintenance records. The clock synchronization module is used to align time information from different data sources to ensure that all data items are synchronized. The scale alignment module is used to unify short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, and real-time load sequences to the same time scale. The missing data filling module is used to fill in missing data parts to ensure data integrity; The time anchoring module is used to align all data to a uniform point in time or time period to ensure time consistency. The two-dimensional modeling module is used to combine load demand with equipment status information to generate a two-dimensional prediction structure; The rule instantiation module is used to generate executable adjustment plans based on policy library templates and real-time load data; The plan write-back and update module is used to write back the adjusted maintenance plan and record the audit results.

[0013] The following are the beneficial effects of the present invention: (1) By using dual-dimensional modeling, a dual-dimensional prediction structure is generated by innovatively combining load demand and equipment status information. This enables the system to simultaneously consider grid load fluctuations and equipment degradation patterns, accurately predict load changes and equipment health status, thereby providing a scientific and accurate data foundation for maintenance plans. This effectively avoids the prediction errors and unreasonable maintenance plans caused by traditional methods that rely on a single data dimension, and improves the reliability and adaptability of the plan.

[0014] (2) By using clock synchronization and scale alignment techniques, data at different time scales (such as short-term load sequences, weekly and monthly load sequences, etc.) are unified during the acquisition of data from multiple sources, ensuring data consistency and comparability. This not only solves the analysis error caused by data inconsistency in existing methods, but also ensures that the analysis results after multidimensional data fusion have high precision and reliability, providing solid support for subsequent maintenance decisions.

[0015] (3) The event-triggered temporary adjustment mechanism dynamically adjusts the maintenance plan, activates backup equipment, and optimizes the process sequence by monitoring load changes in real time and matching preset fluctuation patterns, combined with rule instantiation and target selection. This enables the system to respond to load fluctuations and equipment status changes in real time, automatically optimize maintenance time and resource allocation, ensure stable operation of the power grid under high load or sudden conditions, improve maintenance efficiency and overall power grid stability, and overcome the shortcomings of insufficient flexibility in traditional static maintenance plans. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the generator set maintenance plan generation method based on multi-scale data fusion provided in this application embodiment; Figure 2 This is a structural block diagram of a generator set maintenance plan generation system based on multi-scale data fusion, provided in an embodiment of this application. Detailed Implementation

[0017] Example 1: Refer to Figure 1This is a flowchart illustrating the generator set maintenance plan generation method based on multi-scale data fusion provided in an embodiment of the present invention. The process may include at least steps S100-S400: S100: Obtain short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates and historical maintenance records; perform clock synchronization and equipment identifier mapping, scale alignment and missing measurement filling and time anchoring, and two-dimensional modeling processing to generate a two-dimensional prediction structure. S200, execution constraint archiving and section mapping, feasible time period screening and output reduction calculation, resource allocation and process arrangement and conflict resolution, generate maintenance adaptation plan draft structure; S300: Perform correlation mining by comparing and analyzing historical maintenance data and current maintenance plan drafts; perform parameter learning by using machine learning or deep learning algorithms; extract rules by analyzing recurring patterns in historical maintenance records; perform combined evaluation of alternative solutions based on extraction rules and parameters; and perform version packaging processing for version management of each solution to generate a multi-solution sequence structure including different versions of maintenance solutions. S400 performs the following functions: monitoring real-time load changes and matching them with preset load fluctuation patterns; instantiating rules from the strategy library templates to apply to actual load fluctuations and equipment status changes; selecting target schemes based on real-time load data and strategy library template rules; quickly adjusting the maintenance procedure sequence and execution method based on the target selection results; dynamically adjusting the standby equipment activation sequence; compressing maintenance procedures and execution time; merging adjusted plans; updating parameters in the system; recording and tracking the adjustment process; and generating a plan write-back configuration structure that includes adjusted maintenance plans and standby equipment call-up plans.

[0018] S100: Obtain short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates and historical maintenance records; perform clock synchronization and equipment identifier mapping, scale alignment and missing measurement filling and time anchoring, and two-dimensional modeling processing to generate a two-dimensional prediction structure. Specifically, short-term load sequences, weekly / monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates, and historical maintenance records will be processed as input data. Short-term load sequences contain power system load demand forecasts over short periods: typically a few hours or a day, providing load fluctuation trends based on real-time market demand. Weekly / monthly load sequences focus on load forecasts over longer timescales, typically covering several weeks or months, used to capture longer-term seasonal variations and cyclical demand. Equipment operating parameter sequences involve unit status data during operation, including start-up and shutdown times, output records, and various operating parameters such as equipment efficiency and performance degradation data, used to reflect the health status and operating mode of the equipment. Real-time load sequences are generated from grid dispatch data. The real-time load data directly acquired by the system can reflect the current load status and changing trends of the power grid in real time; the unit constraint benchmarks include the constraints on equipment operation, such as minimum technical output and maximum allowable output reduction, to ensure the stability and operability of unit operation during maintenance; the strategy library templates contain a series of predetermined operations and strategies to be executed according to load fluctuations and maintenance needs, such as adjusting maintenance plans and activating standby equipment when the load changes; historical maintenance records provide past equipment maintenance data, including maintenance duration, maintenance methods and load impact, which helps to identify the potential impact of maintenance on the load.

[0019] In the data acquisition and preparation phase, the first step is to aggregate and integrate data from different data sources. A key task in this phase is to align load forecasting data, equipment operation data, and historical data from multiple data sources using a unified clock synchronization method, ensuring temporal consistency across all datasets. Specifically, timestamp alignment ensures that the time dimension of each dataset perfectly matches the scheduling cycle, thus avoiding data deviations caused by time mismatches. Furthermore, to address the multi-scale issues of equipment operation status and load forecasting, a scale alignment method is needed to adjust short-term and weekly / monthly load forecasting data, as well as equipment status data from different time periods, to the same time scale, enabling effective comparison and analysis.

[0020] After clock synchronization and scale alignment, the next step is to perform missing data imputation. For missing data encountered during data acquisition, methods such as data interpolation, extrapolation algorithms, or regression models based on historical data can be used to fill in missing values, ensuring the completeness and accuracy of subsequent analysis. For regions exhibiting abnormal fluctuations, outlier detection algorithms can be used to mark them, and data correction strategies, such as mean or median substitution, can be employed to repair these anomalies. The key to this step is ensuring data completeness and accuracy, as erroneous data can negatively impact the model's predictive performance.

[0021] The next step is time anchoring, where all data from various data sources needs to be unified into a global time frame. This process is crucial because the timelines of load data, equipment operation data, and maintenance records need to be precisely aligned to ensure that all data reflect the grid status and equipment health at the same point in time. Time anchoring can be accomplished by remapping all data to a unified time node or time period, ensuring seamless data continuity across the time dimension.

[0022] Finally, by performing two-dimensional modeling on the processed data, combining load demand and equipment status, a two-dimensional prediction structure is generated. Specifically, load demand prediction and equipment status information are integrated into a single model. By analyzing grid load fluctuations and equipment degradation patterns, a scientific basis is provided for subsequent maintenance plan generation. The two-dimensional model, by simultaneously considering load demand and equipment status, can provide more accurate prediction results for each time period, thus providing fundamental data for subsequent system maintenance scheduling and load adaptation optimization.

[0023] Through the above processing, the final generated two-dimensional prediction structure is a composite model based on load demand and equipment status, which can comprehensively and accurately reflect the load fluctuations of the power grid and the health status of equipment. This structure provides an accurate predictive basis for subsequent maintenance plan generation and load adaptation, and is the key to the optimization of generator set maintenance plans in this invention.

[0024] In summary, the technical effects of this step are as follows: By synchronizing, aligning, and correcting multidimensional data, a two-dimensional prediction structure is generated by combining load demand and equipment status. This provides an accurate prediction basis for the generation of subsequent maintenance plans, ensuring the accuracy and reliability of subsequent steps.

[0025] S200, execution constraint archiving and section mapping, feasible time period screening and output reduction calculation, resource allocation and process arrangement and conflict resolution, generate maintenance adaptation plan draft structure; In this step, the two-dimensional prediction structure, unit constraint benchmarks, and other input data are used together to generate a draft structure of a maintenance adaptation plan that meets grid load requirements and equipment operating constraints. This structure not only provides a time window for maintenance but also ensures that maintenance activities can be carried out smoothly without affecting grid stability by rationally allocating resources and arranging procedures. The specific process is as follows.

[0026] First, based on the two-dimensional forecast structure generated in the previous steps, and combining the load demand forecast and equipment status forecast results, data processing and analysis begin. The two-dimensional forecast structure includes grid load fluctuations and unit operating status, reflecting the changing trends of load demand and the operating characteristics of the units. On this basis, combined with the operating rules in the unit constraint benchmarks, such as minimum technical output and maximum permissible output reduction, a constraint archiving operation is performed. The main purpose of this step is to uniformly archive the unit's operating constraints and forecast load data, ensuring that subsequent maintenance activities can be carried out within the unit's capacity, while avoiding maintenance operations with significant impact during high load periods. The archiving process matches the unit's load constraints with the forecast load data, ensuring that each maintenance period meets the unit's minimum output requirements and the grid stability requirements.

[0027] Further, segment mapping is performed. Segment mapping refers to mapping load demand forecasts to maintenance periods to determine which times are most suitable for equipment maintenance. Because power grid load fluctuations are periodic and seasonal, it is necessary to combine weekly and monthly load forecasts to identify load troughs. Maintenance during these periods minimizes the impact on the power grid. The mapping process determines the optimal maintenance period by comparing changes in load demand with predetermined maintenance windows. Especially for the maintenance of large-scale equipment, the impact of load fluctuations can be very significant; therefore, it is essential to accurately identify maintenance windows with minimal load and within the equipment's tolerance.

[0028] The next stage is the feasible time period selection phase. In this phase, the feasible time period selection is based on the results of the aforementioned segment mapping, combined with the unit's operating status and grid demand to select the most suitable maintenance time period. Through comprehensive analysis of predicted load fluctuations, equipment operating data, and constraints, the selected time period must not only ensure that equipment maintenance does not cause excessive load fluctuations, but also ensure that the unit's minimum technical output is met. This time period selection considers not only the overall grid load demand, but also the unit's response capability under different operating conditions. Therefore, the selection process requires multiple iterations to determine the optimal maintenance time period.

[0029] After the feasible time period is selected, the output reduction calculation phase begins. Based on the unit's minimum technical output requirements and changes in load demand, the maximum allowable output reduction during maintenance is calculated. Specifically, this process involves analyzing the current load level and the unit's maximum reduction limit to determine the maximum allowable load reduction without affecting the grid's power supply stability. This calculation provides a basis for subsequent resource allocation, ensuring that the grid's power supply capacity is not affected during maintenance. If a sudden increase in load occurs during maintenance, the insufficient power supply can still be supplemented through standby units or other emergency strategies.

[0030] The next steps are resource allocation and work order scheduling. Resource allocation involves arranging various resources required for maintenance, such as equipment, personnel, and spare parts, to ensure the smooth progress of maintenance activities. In this step, resources are rationally allocated based on the specific maintenance needs of the unit, the technical requirements of the equipment, and the length of the maintenance period. This process needs to consider multiple factors, including the special characteristics of the equipment, the scarcity of resources, and potential resource conflicts. Furthermore, work order scheduling involves allocating maintenance operations to different time periods based on the technical requirements of the equipment and the difficulty of maintenance, ensuring that each operational step during the maintenance process can be executed smoothly. Work order scheduling also involves prioritizing different maintenance tasks and figuring out how to complete tasks in the shortest possible time to minimize unit downtime.

[0031] Based on resource allocation and work sequence arrangement, if any conflicts are found in the maintenance plan, such as excessive resource overlap or conflicting maintenance time periods, conflict resolution is required. This process involves analyzing and evaluating the conflicts, combined with optimal scheduling algorithms, to adjust resource allocation or maintenance time, ensuring that maintenance tasks are not delayed due to insufficient resources or time conflicts. This process ensures the efficiency and continuity of maintenance operations, minimizing bottlenecks caused by resource conflicts or overlapping plans.

[0032] Ultimately, all the maintenance plans that have been screened, calculated, configured, and resolved will form a draft structure for the maintenance adaptation plan. This draft structure includes specific maintenance periods, resource allocation, work procedures, and load protection plans. Through this draft, the specific requirements and arrangements for each maintenance task can be clearly identified, providing foundational data for subsequent multi-solution generation and optimization. The generated draft maintenance adaptation plan will then be further processed by the S300's historical maintenance records and solution evaluation module.

[0033] In summary, the technical effects of this step are as follows: By combining the dual-dimensional prediction structure with the unit constraint benchmark, the maintenance window screening, resource allocation, process arrangement and conflict resolution were completed, and a reasonable draft structure of maintenance adaptation plan was generated, which provides a solid foundation for the generation and adjustment of multiple plans in subsequent steps.

[0034] S300: Perform correlation mining by comparing and analyzing historical maintenance data and current maintenance plan drafts; perform parameter learning by using machine learning or deep learning algorithms; extract rules by analyzing recurring patterns in historical maintenance records; perform combined evaluation of alternative solutions based on extraction rules and parameters; and perform version packaging processing for version management of each solution to generate a multi-solution sequence structure including different versions of maintenance solutions. In this step, the draft structure of the maintenance adaptation plan and historical maintenance records will serve as input data. Through multiple data processing steps and model learning, a multi-solution sequence structure will be generated that can be used for evaluation and decision-making. The core task of this process is to provide a basis for subsequent scheme evaluation and adjustment by mining and analyzing historical maintenance data and combining it with existing draft maintenance adaptation plans. Specific processes include association mining, parameter learning, rule extraction, combined evaluation, and version packaging. The following is a detailed description of each sub-step.

[0035] First, relevant input data is obtained from the draft structure of the maintenance adaptation plan and historical maintenance records. The draft structure includes information such as the initially planned maintenance periods, resource allocation, and work procedures. Historical maintenance records include data on past maintenance activities, including the duration of each maintenance, operating steps, equipment operating status, and post-maintenance load fluctuations. This historical data provides important references for analyzing equipment performance under different maintenance conditions. By integrating this data, a historical correlation input package can be obtained, providing a necessary foundation for subsequent data mining and learning processes.

[0036] Next, correlation mining is conducted. This process involves comparing historical maintenance data with the current draft maintenance plan to uncover potential relationships between maintenance operations and load fluctuations. Specifically, the impact of maintenance tasks on equipment condition, load fluctuations, and grid stability is analyzed to identify which types of maintenance activities are closely related to load demand fluctuations. For example, some maintenance activities may cause significant load fluctuations, while others have a smaller impact on load changes. By mining these correlations, a scientific basis can be provided for subsequent decision-making, ensuring that the negative impact of maintenance on grid operation is minimized in actual implementation.

[0037] Furthermore, parameter learning is conducted. By performing parameter learning on various data included in historical maintenance records and the current draft plan, a model suitable for predicting grid load and equipment operating status is constructed. In this process, machine learning or deep learning algorithms are used to learn the correlation between maintenance activities and load fluctuations, and the parameters of the relevant models are optimized, enabling the model to accurately predict the specific impact of different types of maintenance activities on load fluctuations. Through continuous training and adjustment, a parameter model with high accuracy and predictive capabilities is generated. This model will provide data support for subsequent adjustments to maintenance plans, allowing for dynamic adjustments to maintenance plans based on load fluctuations during actual maintenance to minimize the impact on the power grid.

[0038] The next step is rule extraction. The goal of rule extraction is to generate a set of rules that can be used to evaluate maintenance plans by analyzing recurring patterns in historical maintenance records. These rules may include specific maintenance operations for certain equipment, measures to cope with load fluctuations, and optimization suggestions for resource allocation. For example, if certain equipment is prone to failure under specific load conditions, the rule extraction process will indicate that maintenance of these devices should be avoided during periods of high load, or that backup equipment should be prioritized. By extracting these rules, an operational guideline can be formed to guide the formulation and adjustment of future maintenance plans.

[0039] After the rules are extracted, a combined evaluation is performed. The main task of this step is to comprehensively evaluate all alternative maintenance plans based on the previously extracted rules and parameters, and select the optimal plan. Specifically, the evaluation criteria include the impact of maintenance on load fluctuations, the speed of equipment recovery, and resource utilization efficiency. By scoring and ranking the advantages and disadvantages of different plans, the most suitable maintenance plan is selected. This step ensures the rationality and feasibility of each maintenance plan, providing a scientific basis for the final plan selection.

[0040] Finally, version packaging is performed. In this step, version management of each solution is implemented to generate a multi-solution sequence structure. Each solution is assigned a version number based on its priority, applicability, and implementation risk. All versions are packaged into a complete maintenance plan alternative solution library for subsequent decision-making and adjustments. The multi-solution sequence structure not only provides alternative maintenance solutions but also offers decision-makers performance forecasts for different solutions under varying loads and equipment conditions, making the decision-making process more flexible and reliable.

[0041] Through the above processing, the generated multi-scheme sequence structure provides multiple alternatives for the implementation of maintenance plans in subsequent steps. This structure provides important support for the effective management of grid load and the optimized maintenance of generating units, enabling maintenance work to be carried out more efficiently and ensuring the stable operation of the grid during the maintenance process.

[0042] In summary, the technical effects of this step are as follows: through association mining, parameter learning, rule extraction, combined evaluation, and version packaging, a multi-scheme sequence structure based on historical data and existing schemes is generated, providing rich alternative schemes and scientific basis for subsequent maintenance scheme optimization and decision-making, ensuring the efficiency of maintenance activities and the stability of the power grid.

[0043] S400 performs the following functions: monitoring real-time load changes and matching them with preset load fluctuation patterns; instantiating rules from the strategy library templates to apply to actual load fluctuations and equipment status changes; selecting target schemes based on real-time load data and strategy library template rules; quickly adjusting the maintenance procedure sequence and execution method based on the target selection results; dynamically adjusting the standby equipment activation sequence; shortening maintenance procedures and execution time; merging adjusted plans; updating parameters in the system; recording and tracking the adjustment process; and generating a plan write-back configuration structure that includes adjusted maintenance plans and standby equipment call plans. Specifically, the multi-scheme sequence structure, strategy library template, and real-time load sequence are used as inputs to begin event-triggered matching and rule instantiation. At this stage, the previously generated multi-scheme sequence structure is first used as input. This structure includes different maintenance schemes and their corresponding load fluctuation characteristics and equipment status predictions, demonstrating the potential impact of different maintenance plans on the grid load. The strategy library template provides strategies for dealing with grid load fluctuations and maintenance requirements, including how to adjust maintenance plans to avoid grid instability during sudden load increases. The real-time load sequence provides real-time load change data of the grid during actual operation. This data helps the system determine the current load status and guide the adjustment of maintenance plans.

[0044] During the event-triggered matching phase, the system first monitors real-time load changes in the power grid and matches them against preset load fluctuation patterns. When abnormal fluctuations or significant increases or decreases occur in the real-time load sequence, the system automatically triggers the corresponding response mechanism based on the adjustment rules stored in the strategy library templates. For example, if real-time load data indicates that the load is about to reach its peak or surge, the system will immediately activate the preset load adjustment plan, dispatching backup equipment or shortening non-critical maintenance procedures to prevent excessive impact on the normal power supply of the power grid.

[0045] Furthermore, rule instantiation involves applying rules extracted from the strategy library templates to actual load fluctuations and equipment status changes. Specifically, based on real-time load changes and known maintenance procedures, the system generates immediately executable adjustment plans based on rule instantiation. These plans may include adjusting maintenance periods, activating standby units, and increasing output capacity, thereby enabling real-time responses to grid load fluctuations and avoiding disruption to normal maintenance work. By instantiating rules, the system ensures greater adaptability of each maintenance task to the grid's operating status, reducing interference with the grid.

[0046] During the target selection phase, the system selects the most suitable target scheme based on real-time load data and rules in the strategy library templates. This scheme is typically a dynamic adjustment plan used to cope with emergencies or load fluctuations. For example, when a sharp increase in load occurs, the system will select a temporary adjustment scheme to ensure stable grid operation while ensuring that equipment maintenance tasks can be completed in the shortest possible time. The core of target selection is to comprehensively consider the urgency of the maintenance task, the severity of load fluctuations, and the availability of backup equipment, thereby achieving a balance among multiple factors.

[0047] Temporary adjustments to the scheduling occur after the target solution is selected. The core task of this process is to quickly adjust the sequence and execution method of maintenance procedures based on the target selection results. Specifically, in the event of a sudden load surge, the system will adjust the maintenance sequence, prioritizing tasks with minimal impact on grid power supply, or postponing some tasks to avoid maintenance during high-load periods. Furthermore, the system will incorporate the use of backup equipment into the adjustment plan. If a device is under maintenance and the load needs immediate restoration, the backup equipment will be activated to provide temporary supplementary power, maintaining a stable grid supply.

[0048] In the standby equipment activation and maintenance procedure compression phase, the system dynamically adjusts the activation sequence of standby equipment and optimizes the execution time of maintenance procedures based on current load fluctuations and the specific circumstances of the maintenance tasks. If a certain procedure does not affect the basic load supply of the power grid, its execution time can be shortened or its operation process can be temporarily adjusted to speed up the maintenance progress. In addition, if the standby units can support the load demand, the system will activate the standby units when necessary to ensure the stability of the power grid supply.

[0049] After all temporary adjustments are applied and implemented, all data and processing results are integrated and a plan write-back is performed. This process involves updating the system with the adjusted maintenance plan, standby equipment call-up plan, process compression information, and load scheduling, ensuring that all changes to the maintenance plan are accurately recorded and provide a basis for subsequent operations. Auditing and disk entry records the adjustment process, ensuring that each step is tracked and audited to avoid potential operational errors or omissions. Finally, the generated plan write-back configuration structure contains all adjusted information for use in subsequent maintenance record updates and plan optimization.

[0050] The technical effects of this step can be summarized as follows: By monitoring real-time load changes, instantiating rules, and selecting targets, combined with temporary adjustment scheduling and backup equipment mobilization, the maintenance plan is dynamically adjusted to ensure that the power grid can still operate stably under load fluctuations, while ensuring that maintenance work is carried out as planned. Finally, an operable maintenance plan write-back configuration structure is generated to support subsequent steps.

[0051] Example 2: Figure 2A structural block diagram of a generator set maintenance plan generation system based on multi-scale data fusion according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include: Data acquisition module 01 is used to acquire short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates, and historical maintenance records. Specifically, it receives short-term load sequences, weekly and monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates, and historical maintenance records from external data sources and performs data acquisition tasks. This data is provided by multiple monitoring systems and databases and is formatted and preprocessed by the system to ensure data accuracy and integrity. The data acquisition module periodically collects and stores this data through its interface with the data sources, forming datasets and saving them in a temporary storage area for subsequent processing. The datasets are recorded as collected data and transmitted to the clock synchronization module for further synchronization and alignment operations.

[0052] The clock synchronization module 02 is used to align time information from different data sources, ensuring synchronization of all data items. Specifically, it receives data acquired from the data acquisition module and performs clock synchronization. By aligning the timestamps of all input data sources, it ensures that the time information of each data item matches the standard time of the power grid dispatching system. Based on a unified time frame, the module synchronizes data at different time scales, ensuring consistency in time between short-term and weekly / monthly load data, equipment operating parameters, and real-time load data. After synchronization, the time-synchronized data is recorded as synchronized data and transmitted to the scale alignment module for use in subsequent scale alignment operations.

[0053] The scale alignment module 03 is used to unify short-term load sequences, weekly / monthly load sequences, equipment operating parameter sequences, and real-time load sequences to the same time scale. Specifically, it receives synchronization data from the clock synchronization module and performs scale alignment according to a preset time frame. The module adjusts the short-term load sequences, weekly / monthly load sequences, equipment operating parameter sequences, and real-time load sequences to the same time scale, ensuring that data with different time granularities can be effectively compared and analyzed. The scale alignment process involves compressing or expanding time periods to ensure that the time dimension of all data items is consistent. The scale-aligned data is recorded as aligned data and transmitted to the missing measurement filling module for use in the missing measurement filling operation.

[0054] The missing data completion module 04 is used to fill in missing data portions to ensure data integrity. Specifically, it receives aligned data from the scale alignment module and performs completion operations on the missing parts. Missing data may be caused by sensor failure, data loss, or abnormal interruption. The missing data completion module fills in the missing parts using methods such as interpolation, extrapolation, or regression models. The method intelligently fills in the missing data based on the temporal relationship and historical patterns of the data, maintaining data consistency and coherence as much as possible. The filled data is recorded as filled data and transmitted to the time anchoring module for subsequent time alignment and calibration processing.

[0055] The Time Anchoring Module 05 aligns all data to a unified point in time or time period, ensuring time consistency. Specifically, it receives supplementary data from the Missing Data Filling Module and performs time anchoring processing on various data types. This module aligns the timestamps of short-term load sequences, weekly / monthly load sequences, equipment operating parameter sequences, and real-time load sequences to a global point in time, ensuring that all data items are processed consistently within the same time period. Through time anchoring, the module achieves precise synchronization of data across different time periods, enabling seamless data integration across multiple time scales. Time-anchored data is recorded as anchored data and transmitted to the two-dimensional modeling module for use in subsequent modeling operations.

[0056] The dual-dimensional modeling module 06 combines load demand with equipment status information to generate a dual-dimensional prediction structure. Specifically, it receives anchored data from the time anchoring module, combines load demand forecasts with equipment operating status data, and generates a prediction structure using dual-dimensional modeling technology. This module analyzes and models load demand and equipment status as two dimensions, integrating information from short-term load sequences, weekly / monthly load sequences, equipment operating parameter sequences, and real-time load sequences to generate a dual-dimensional prediction structure for subsequent decision-making. The dual-dimensional model can capture changes in grid load fluctuations and equipment degradation patterns, providing a scientific basis for the generation and optimization of maintenance plans. The generated dual-dimensional prediction structure is recorded as a prediction model and passed to the rule instantiation module for subsequent rule generation and adjustment.

[0057] Rule instantiation module 07 is used to generate executable adjustment plans based on strategy library templates and real-time load data. Specifically, it receives the prediction model from the dual-dimensional modeling module, combines it with real-time load data and strategy library templates, and performs rule instantiation. This module generates adjustment plans that can be executed immediately by applying rules from the strategy library. The rule instantiation process dynamically adjusts the maintenance plan based on real-time load fluctuations, equipment operating status, and maintenance needs, generating corresponding emergency adjustment plans. The adjustment plans are recorded and passed to the plan write-back and update module for use in subsequent maintenance plan write-back operations.

[0058] The Plan Write-back and Update Module 08 is used to write back the adjusted maintenance plan and record it for auditing. Specifically, it receives the adjustment scheme from the rule instantiation module, records and writes it back. The module writes back the generated maintenance plan, adjustment scheme, and related resource configurations, updating the system's database. Simultaneously, all adjustment records are recorded in detail through the audit log system, ensuring that each operation is traceable and auditable. The written-back maintenance plan will be output as the final plan and archived in the system for subsequent execution and monitoring. The plan write-back data will be recorded as a write-back plan for future system updates and use.

Claims

1. A method for generating generator set maintenance plans based on multi-scale data fusion, characterized in that, include: S100: Obtain short-term load sequence, weekly and monthly load sequence, equipment operating parameter sequence, real-time load sequence, unit constraint benchmark, strategy library template and historical maintenance record, perform clock synchronization and equipment identification mapping, scale alignment and missing measurement filling and time anchoring and two-dimensional modeling processing, and perform two-dimensional modeling based on load demand and equipment status to generate a two-dimensional prediction structure. S200. Based on the dual-dimensional prediction structure, perform constraint archiving and segment mapping, feasible time period screening, output reduction calculation, resource allocation, process arrangement and conflict resolution, and generate a draft structure of maintenance adaptation plan. The constraint archiving includes matching the operating rules in the unit constraint benchmark with the predicted load data; The segment mapping includes comparing load demand changes with scheduled maintenance windows; The selection of feasible time periods is based on the segment mapping results and combines the unit's operating status with the grid demand to select maintenance time periods; The output reduction calculation includes calculating the maximum allowable output reduction during maintenance based on the unit's minimum technical output requirements and changes in load demand. The resource allocation includes the arrangement of equipment, personnel and spare parts required for maintenance; The process arrangement and conflict resolution include prioritizing different maintenance tasks and making adjustments for resource overlap and maintenance time period conflicts; S300. Based on the draft structure of the maintenance adaptation scheme, generate a multi-scheme sequence structure including different versions of maintenance schemes; specifically, perform the following processing: Correlation mining is performed by comparing and analyzing historical maintenance data with current draft maintenance plans. This correlation mining includes analyzing the impact of maintenance tasks on equipment status, load fluctuations, and power grid stability, and identifying maintenance activity types that are closely related to load demand fluctuations. Parameter learning using machine learning or deep learning algorithms; the parameter learning includes using machine learning or deep learning algorithms to learn the correlation between maintenance activities and load fluctuations, and optimizing relevant model parameters; Rule extraction is achieved by analyzing recurring patterns in historical maintenance records. This rule extraction includes generating a set of rules for evaluating maintenance plans, which includes maintenance operations for specific equipment, measures to address load fluctuations, and optimization suggestions for resource allocation. Based on the extraction rules and parameters, a comprehensive evaluation of the candidate solutions is conducted, and the versions are packaged and processed. S400. Based on the multi-scheme sequence structure, generate a plan write-back configuration structure containing the adjusted maintenance plan and the standby equipment call-up plan; specifically, perform the following processing: Monitor real-time load changes and match real-time load changes with preset load fluctuation patterns for event triggering; Instantiate rules from the strategy library templates to apply to actual load fluctuations and equipment status changes, and select the target scheme based on real-time load data and strategy library template rules. Based on the target selection results, quickly adjust the temporary arrangement of the maintenance procedure sequence and execution method; Dynamically adjusting the activation order of standby equipment and shortening maintenance procedures; The execution time of maintenance procedures was reduced, and the adjusted plans were merged. The adjusted plan is updated to the parameters and the adjustment process is recorded and tracked through an audit process.

2. The method according to claim 1, characterized in that, Short-term load sequences, weekly / monthly load sequences, equipment operating parameter sequences, real-time load sequences, unit constraint benchmarks, strategy library templates, and historical maintenance records include: Short-term load sequences contain forecast data on load demand over a period of several hours or a day; Weekly / monthly load series contain forecast data for several weeks or months; The equipment operating parameter sequence includes the unit's start-up and shutdown times, output records, equipment efficiency, and performance degradation data; Real-time load sequences contain instantaneous load data obtained from the power grid dispatching system; The unit's constraint benchmarks include minimum technical output and maximum permissible output reduction. The strategy library templates include options for adjusting maintenance plans and activating backup equipment in response to load changes. Historical maintenance records include maintenance duration, maintenance methods, and load impact.

3. The method according to claim 1, characterized in that, Combined evaluation and version packaging processing includes: The combined evaluation includes a comprehensive evaluation of the alternative solutions based on the extraction rules and the model parameters; The version packaging includes version management of each scheme and generating the multi-scheme sequence structure.

4. The method according to claim 1, characterized in that, The process of instantiating execution rules and selecting target schemes includes: The rule instantiation includes applying the rules in the strategy library template to actual load fluctuations and equipment status changes to generate an executable adjustment plan; The target scheme selection includes selecting a target scheme from the multi-scheme sequence structure based on the real-time load sequence and the rules in the strategy library template.

5. The method according to claim 1, characterized in that, The process of adjusting maintenance procedures and standby equipment mobilization based on the target scheme selection results, and generating a plan write-back configuration structure containing the adjusted maintenance plan and standby equipment mobilization plan, includes: The adjustment of maintenance procedures and the call-up of standby equipment based on the target scheme selection results includes quickly adjusting the order and execution method of maintenance procedures, as well as dynamically adjusting the activation order of standby equipment. The process of generating a plan write-back configuration structure that includes the adjusted maintenance plan and the standby equipment call-up plan also includes integrating the adjusted plan, updating the adjusted plan to the system, and recording and tracking the adjustment process.

Images