Campus integrated energy system preventive maintenance scheduling method

Abstract

This invention discloses a preventive maintenance scheduling method for an integrated energy system in a park, comprising: collecting operating parameters of multiple devices in the HVAC system and energy storage system within the park, the operating parameters including at least heterogeneous characteristic parameters reflecting mechanical degradation, electrical degradation, and thermal degradation of the devices; based on the collected heterogeneous characteristic parameters, obtaining a health index through a health assessment model, and predicting the remaining service life of each device based on a time-series prediction model; triggering maintenance decisions based on the obtained health index and remaining service life, and sorting the devices requiring maintenance based on a multi-dimensional decision matrix including urgency, criticality, and impact to generate a maintenance priority list; obtaining production plan information for a future preset period, and for each maintenance task in the maintenance priority list, optimizing it based on the remaining service life of the corresponding device and the production plan information, and recommending a maintenance window.

Description

Technical Field

[0001] This invention belongs to the field of preventive maintenance technology for energy systems, and specifically relates to a preventive maintenance scheduling method for a comprehensive energy system in a park. Background Technology

[0002] With the development of smart parks and green buildings, integrated energy solutions that combine HVAC and energy storage systems are becoming increasingly popular, serving as the core for ensuring energy efficiency and operational reliability in parks. These systems are complex in structure, including mechanical power equipment such as chillers, pumps, and cooling towers, as well as power electronic equipment such as battery storage cabinets and power conversion systems. Their continuous and stable operation is crucial for the park's production and office activities.

[0003] Traditional maintenance strategies are mainly divided into reactive maintenance and periodic preventive maintenance. Reactive maintenance often leads to unplanned downtime, causing production losses and safety risks; while periodic maintenance may result in over-maintenance or under-maintenance, wasting resources and failing to effectively prevent unexpected failures.

[0004] With the development of IoT sensing technology and artificial intelligence algorithms, predictive maintenance technology based on equipment condition monitoring aims to analyze the degradation trend of equipment by collecting equipment operation data in real time, thereby enabling precise intervention before failure occurs.

[0005] However, applying predictive maintenance technology to integrated energy systems in industrial parks, which include multiple types of equipment across various physical domains such as HVAC and energy storage, still faces a series of pressing technical challenges. Firstly, existing predictive maintenance solutions often target single types of equipment or single physical processes, lacking effective means for cross-domain, unified health status assessments of HVAC mechanical systems and electrochemical energy storage systems. System maintenance personnel face multiple isolated data silos and assessment standards, making it difficult to comprehensively grasp the overall health level of the park's energy system and make collaborative decisions.

[0006] Secondly, even if the remaining service life of a single piece of equipment is predicted through models, effectively translating this prediction into an executable and optimizable systematic maintenance action plan is a weak link in existing technologies. Relying solely on fault warnings may lead to improper allocation of maintenance resources. If maintenance windows are not properly scheduled, especially if equipment maintenance or repairs are carried out during peak production periods, planned production interruptions or efficiency losses may still occur.

[0007] Therefore, the operation and maintenance management of the park's integrated energy system still faces risks such as fragmented decision-making, delayed response, and conflicts with production activities, and the overall operation and maintenance efficiency and economy need to be further improved. Summary of the Invention

[0008] This invention provides a preventive maintenance scheduling method for a park's integrated energy system. By integrating the multi-source heterogeneous degradation characteristics of HVAC and energy storage systems, constructing a unified health assessment and life prediction model across equipment, and performing intelligent priority ranking and window optimization scheduling based on prediction results and production plan information, it achieves intelligent collaboration between preventive maintenance and production operations, solving the problems of unplanned downtime and high operation and maintenance costs caused by conflicts between maintenance activities and production peaks and the lack of systematic optimization in maintenance decisions.

[0009] The technical solution adopted in this invention is as follows: A preventive maintenance scheduling method for an integrated energy system in a park includes: The system collects operating parameters from multiple devices in the HVAC and energy storage systems within the park. These operating parameters include at least heterogeneous characteristic parameters reflecting the mechanical, electrical, and thermal degradation of the devices. Based on the collected heterogeneous feature parameters, a health index is obtained through a preset health assessment model to uniformly characterize the health status of each device, and the remaining service life of each device is predicted based on a time-series prediction model. Based on the obtained health index and remaining service life, a maintenance decision is triggered, and the equipment to be maintained is sorted according to a multi-dimensional decision matrix that includes urgency, criticality and impact, to generate a maintenance priority list. Obtain production plan information for the future preset period, and for each maintenance task in the maintenance priority list, optimize it based on the remaining service life of the equipment corresponding to the task and the production plan information, recommend maintenance windows, and generate a preventive maintenance execution plan that includes specific execution times.

[0010] The preventive maintenance and scheduling method for the integrated energy system of the park adopted in this invention also has the following additional technical features: The heterogeneous characteristic parameters include vibration spectrum characteristics, current harmonic characteristics, and condenser differential pressure characteristics from the chiller unit, as well as internal resistance characteristics and voltage-current imbalance characteristics from the energy storage battery.

[0011] The health assessment model adopts a weighted fusion model, which is implemented as follows:

[0012] HI stands for Health Index. For the first Preset weights for each feature parameter, For the first The normalized degradation deviation of each feature parameter As a nonlinear correction factor related to the total degradation degree, if At that time, ,otherwise .

[0013] The time-series prediction model is a long short-term memory network model based on an encoder-decoder architecture. Input the historical health status sequence and feature parameter sequence into the encoder. The final state of the encoder is input into the decoder for decoding and prediction, and the first health prediction sequence and feature parameter prediction sequence are output. The first health prediction sequence and the feature parameter prediction sequence are input into the fully connected layer to determine the remaining lifespan.

[0014] The first health prediction sequence and the feature parameter prediction sequence are input into the fully connected layer to determine the remaining lifespan, specifically: Based on the predicted sequence of feature parameters, a second health prediction sequence is obtained through the health assessment model; By simultaneously comparing the first health prediction sequence and the second health prediction sequence, and taking the smaller value, the health prediction sequence is determined. The remaining service life is then determined based on the time point when the health prediction sequence first falls below the failure threshold.

[0015] The comprehensive priority score of the multidimensional decision matrix Specifically: , in, The urgency score is calculated based on the health index and the remaining lifespan. The criticality score is determined based on the device's pre-defined importance level within the system. The impact score is determined based on the assessment of the impact of equipment failure on related systems. , , The preset weighting coefficients, and .

[0016] Recommended maintenance windows include: For each maintenance task, determine its theoretical maintainable life. , ],in The result is obtained by subtracting the safety buffer period from the current time and the predicted remaining useful life of the equipment. The production plan information is obtained, which divides future time into peak production unavailable periods, planned downtime ideal available periods, and normal production condition available periods. During the theoretically maintainable period [ , Within the scope of the task, a sliding window traversal algorithm is used to find a continuous time window with a length equal to the estimated working hours of the task, so as to minimize the sum of the comprehensive inconvenience costs of each time period within the window, and to determine the candidate maintenance window. Different inconvenience cost values ​​are assigned to different types of time periods.

[0017] After determining the candidate maintenance window, the following are also included: For each maintenance task in the maintenance priority list, an execution window is determined sequentially from the corresponding candidate maintenance window based on the comprehensive priority score of the multidimensional decision matrix, from highest to lowest. Perform time overlap checks on multiple maintenance tasks, and determine the preventive maintenance execution plan after confirming that there are no time overlaps.

[0018] A second aspect of the present invention provides a preventive maintenance scheduling system for an integrated energy system in a park, used to implement the method described above, the system comprising: The data acquisition and preprocessing module is used to collect operating parameters of multiple devices in the HVAC system and energy storage system within the park. The operating parameters include at least heterogeneous characteristic parameters reflecting the mechanical degradation, electrical degradation and thermal degradation of the equipment. The health assessment and prediction module, based on the collected heterogeneous feature parameters, obtains a health index that uniformly represents the health status of each device through a preset health assessment model, and predicts the remaining service life of each device based on a time-series prediction model. The intelligent decision-making and sorting module is used to trigger maintenance decisions based on the obtained health index and remaining service life, and to sort the equipment to be maintained based on a multi-dimensional decision matrix including urgency, criticality and impact, and generate a maintenance priority list. The production awareness scheduling optimization module is used to obtain production plan information within a future preset period. For each maintenance task in the maintenance priority list, it optimizes the task based on the remaining service life of the equipment corresponding to the task and the production plan information, recommends maintenance windows, and generates a preventive maintenance execution plan that includes specific execution times. The output of the health assessment and prediction module is connected to the input of the intelligent decision-making and ranking module, and the output of the intelligent decision-making and ranking module and the data from the external production system are connected to the input of the production perception scheduling optimization module.

[0019] A third aspect of the present invention provides a computing device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method.

[0020] Due to the adoption of the above technical solution, the beneficial effects achieved by this invention are as follows: 1. In this invention, the method breaks through the limitations of traditional maintenance schemes that target a single device or a single physical process. It realizes cross-system, unified health status perception and life prediction for HVAC and energy storage devices with different physical domains and different degradation mechanisms within the park. This enables operation and maintenance personnel to grasp the overall health level of the entire integrated energy system in real time and quantitatively from an integrated perspective, providing a reliable data foundation for systemic decision-making.

[0021] Secondly, by introducing a multi-dimensional decision matrix that includes urgency, criticality, and impact, this method automatically transforms health prediction results into scientific, reasonable, and prioritized executable maintenance action plans, effectively avoiding the blindness and arbitrariness of maintenance resource allocation and improving the refinement and proactivity of operation and maintenance management.

[0022] Finally, the technical constraint of the remaining service life of the equipment is deeply coordinated and dynamically optimized with the business constraint of the upper-level production plan. This ensures that the recommended maintenance window can proactively avoid production peaks, avoid the risk of interruption to core production operations caused by planned maintenance activities, minimize the impact of all downtime, and significantly improve the overall efficiency and economy of park operation. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the preventive maintenance and scheduling method for the integrated energy system of a park according to one embodiment of the present invention. Detailed Implementation

[0024] To more clearly illustrate the overall concept of the present invention, a detailed description will be provided below with reference to the accompanying drawings and examples.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0026] like Figure 1 As shown, a preventive maintenance scheduling method for a park's integrated energy system includes: S100: Collect operating parameters of multiple devices in the HVAC and energy storage systems within the park. The operating parameters include at least heterogeneous characteristic parameters reflecting the mechanical, electrical, and thermal degradation of the devices.

[0027] The main purpose of this step is to establish a comprehensive and unified data foundation for subsequent health assessments and decision-making.

[0028] The park's integrated energy system comprises equipment with vastly different physical principles and degradation mechanisms, such as chillers and pumps belonging to mechanical / fluid systems, and energy storage batteries belonging to electrochemical systems. To achieve unified management across these systems, it is essential to collect heterogeneous characteristic parameters that reflect the core degradation processes of various equipment types.

[0029] In practice, raw signals are acquired in real time or periodically through a sensor network (such as vibration accelerometers, current transformers, differential pressure transmitters, and battery management systems) deployed in key parts of the equipment. These parameters cover at least the dimensions of mechanical degradation (such as vibration amplitude and spectrum), electrical degradation (such as current harmonic distortion rate and battery internal resistance), and thermal degradation (such as condenser pressure difference).

[0030] By incorporating the heterogeneous operating parameters of all devices into the same data acquisition and preprocessing pipeline, the data silos in traditional operation and maintenance are broken down, system-level health awareness is achieved, and a unified health indicator system across devices is constructed, enabling horizontal comparison and collaborative management of different types of devices.

[0031] S200: Based on the collected heterogeneous feature parameters, a health index is obtained through a preset health assessment model to uniformly characterize the health status of each device, and the remaining service life of each device is predicted based on a time-series prediction model.

[0032] The core objective of this step is to transform the collected diverse and heterogeneous raw data into two intuitive and crucial decision indicators: a unified health index and quantified remaining lifespan.

[0033] The strategy involves using a pre-defined health assessment model to fuse and calculate the normalized feature parameters, outputting a score between 0 and 100. This allows for the quantitative evaluation of the real-time health status of drastically different devices, such as chiller units and energy storage batteries, using the same benchmark.

[0034] Meanwhile, by using time-series prediction models to analyze the historical sequence trends of health index and characteristic parameters, the remaining time from the current state of the equipment to failure is extrapolated and predicted, thus achieving lifespan prediction.

[0035] It provides predictive judgment basis for operation and maintenance decisions, and realizes proactive planning and preventive maintenance.

[0036] S300: Based on the obtained health index and remaining service life, trigger maintenance decisions, and based on a multi-dimensional decision matrix including urgency, criticality and impact, sort the equipment to be maintained and generate a maintenance priority list.

[0037] The main purpose of this step is to transform the output technical forecast indicators into a management decision list with a clear priority that can guide actual operation and maintenance actions.

[0038] The strategy involves constructing a multi-dimensional decision matrix for comprehensive scoring and ranking, based not only on the equipment's own health and remaining lifespan (urgency), but also on the equipment's functional importance in the system (criticality) and the chain reactions that a failure may cause (impact).

[0039] In reality, maintenance resources (manpower, spare parts, time) are always limited, and not all equipment predicted to require maintenance can be addressed immediately. Ranking equipment solely by remaining lifespan may overlook the support of critical production equipment or underestimate the impact of localized failures on the overall system.

[0040] Therefore, this step, by introducing multi-dimensional decision-making logic, ensures that the generated maintenance priority list, while based on technical rationality, also takes into account the economic efficiency of operation and maintenance management and systemic risk control. This upgrades the maintenance plan from a simple ranking based on a single technical parameter to intelligent decision-making based on multi-objective optimization, improving the scientific nature of maintenance resource allocation and overall operational efficiency.

[0041] S400: Obtain production plan information for a future preset period, optimize each maintenance task in the maintenance priority list with the remaining service life of the equipment corresponding to the task as a constraint, and combine the production plan information to recommend maintenance windows, so as to generate a preventive maintenance execution plan containing specific execution times.

[0042] The ultimate goal of this step is to ensure that the optimized maintenance plan can be seamlessly integrated with the core production activities of the park during actual implementation, so as to achieve maintenance without interfering with production.

[0043] The implementation strategy involves obtaining detailed production plan information for future cycles and coupling the theoretical maintainable period (determined by the remaining service life) of each maintenance task with peak / off-peak production periods for optimization. This allows for the recommendation of a maintenance window for each task, ensuring that the window meets the time constraints for safe equipment operation while also falling within periods of low production load or planned downtime.

[0044] If scientifically prioritized maintenance conflicts with peak production times, it may lead to new planned downtime losses. Therefore, this step dynamically matches and optimizes technical maintenance needs with business production rhythms, enabling preventative maintenance scheduling, resolving the inherent contradiction between maintenance activities and production operations, and achieving predictive maintenance to reduce downtime and ensure production.

[0045] It should be noted that the heterogeneous characteristic parameters include vibration spectrum characteristics, current harmonic characteristics, and condenser differential pressure characteristics from the chiller unit, as well as internal resistance characteristics and voltage-current imbalance characteristics from the energy storage battery.

[0046] The main purpose of this implementation method is to select and integrate key indicators that can directly and sensitively reflect the core degradation process of HVAC equipment and energy storage equipment in the integrated energy system of the park, which have completely different physical principles and degradation mechanisms, so as to lay a cross-domain comparable data foundation for building a unified health model.

[0047] For rotating machinery and fluid equipment like chillers, degradation mainly manifests in three aspects: mechanical wear, decreased electrical performance, and declining heat exchange efficiency. Therefore, specific data collection is required: Vibration spectrum characteristics are obtained by installing vibration sensors on key rotating components such as compressors and motors, collecting vibration signals, and performing spectrum analysis (such as Fast Fourier Transform) to extract characteristic frequency amplitudes (such as first harmonic, second harmonic, and high-frequency energy) directly related to faults such as rotor imbalance, misalignment, and bearing wear. This characteristic is used to indicate early degradation of mechanical components.

[0048] Current harmonic characteristics are determined by monitoring the current in the motor drive circuit and calculating its total harmonic distortion (THD) and the content of specific harmonics (such as the 5th and 7th harmonics). Abnormal increases in current harmonics often indicate aging of motor winding insulation, deterioration of power supply quality, or performance degradation of power electronic components.

[0049] The condenser differential pressure characteristic monitors the pressure difference between the inlet and outlet of the condenser cooling water. An abnormal increase in this differential pressure is usually directly related to scaling, blockage, or increased fluid system resistance in the condenser tube bundles, and is a key indicator of heat exchange performance degradation.

[0050] For electrochemical devices like energy storage batteries, their health status is primarily reflected in the consistency between electrochemical impedance and cell performance. Therefore, specific data collection is required: Internal resistance characteristics are measured by the battery management system during charge / discharge transients or using the AC injection method, measuring the DC internal resistance or AC impedance of individual battery cells. The monotonic increase in internal resistance is a direct quantitative reflection of electrochemical aging processes such as the loss of active materials and electrolyte decomposition.

[0051] Voltage and current imbalance characteristics are assessed by monitoring the terminal voltage and charge / discharge current of each individual cell within a battery cluster, and calculating statistical parameters such as standard deviation and range. Increased imbalance is a significant signal of performance differentiation among individual cells within the battery pack, aging of connection points, or failure of the management system's balancing function, accelerating overall capacity decay and posing safety risks.

[0052] This set of characteristic parameters covers multiple physical domains from mechanical, electrical, thermal to electrochemical, encompassing the key degradation paths of major equipment in the target system and ensuring the validity of the source data for health assessment. Secondly, although these characteristics originate from different equipment and reflect different physical processes, they can all be processed into normalized time-series numerical signals. This heterogeneous yet uniformly quantifiable characteristic is used to construct a universal health index model applicable to HVAC and energy storage equipment.

[0053] If only characteristics from a single domain are used (such as monitoring only vibration or only voltage), a unified health assessment across systems cannot be achieved. Therefore, the characteristic parameter selection scheme of this implementation successfully integrates the traditionally fragmented equipment monitoring data silos into a unified report that can comprehensively and fairly reflect the overall health status of the integrated energy system, for subsequent intelligent and collaborative maintenance decisions.

[0054] In a preferred embodiment of the present invention, the health assessment model adopts a weighted fusion model, which is implemented as follows:

[0055] HI stands for Health Index. For the first Preset weights for each feature parameter, For the first The normalized degradation deviation of each feature parameter As a nonlinear correction factor related to the total degradation degree, if At that time, ,otherwise .

[0056] The main purpose of this model is to scientifically, fairly, and intuitively integrate multiple heterogeneous feature parameters into a single health index, thereby enabling horizontal comparison and unified quantitative management of the health status of equipment with vastly different physical principles, such as chillers and energy storage batteries.

[0057] First, for each characteristic parameter (such as vibration characteristic amplitude, current harmonic rate, internal resistance, etc.), calculate its normalized degradation deviation. . This reflects the degree to which the current value of this parameter deviates from its health baseline value. It is usually calculated using the formula (current value - health baseline value) / (fault threshold - health baseline value) and limited to the range [0,1]. The closer the value is to 1, the more severe the degradation of the parameter.

[0058] Secondly, a preset weight is assigned to each feature parameter. The determination of weights can be based on the experience of domain experts or statistical analysis of historical failure data (such as using the analytic hierarchy process). The core principle is that features that have a greater impact on the overall health of the equipment and are more closely related to critical failure modes should be given higher weights. This objectively reflects the differences in the impact of different failure symptoms on overall health, making the assessment results more consistent with engineering reality. The sum of all weights should be 1.

[0059] Then, calculate the weighted degeneracy sum of all features. This value, ranging from 0 to 1, intuitively reflects the overall degradation rate of the equipment.

[0060] Finally, a nonlinear correction factor K is introduced. A degradation threshold is set (e.g., 0.7). At that time, ,otherwise The model exhibits a linear relationship. It aims to accelerate equipment failure in the late stages of degradation: when equipment health deteriorates significantly, its remaining reliable lifespan tends to decline more rapidly. By introducing a K-factor, the health index HI decreases more significantly in the late stages, thus providing a more ample safety margin for early warning and decision-making.

[0061] Furthermore, the time-series prediction model is a long short-term memory network model based on an encoder-decoder architecture. Input the historical health status sequence and feature parameter sequence into the encoder. The final state of the encoder is input into the decoder for decoding and prediction, and the first health prediction sequence and feature parameter prediction sequence are output. The first health prediction sequence and the feature parameter prediction sequence are input into the fully connected layer to determine the remaining lifespan.

[0062] The core objective of this model is not only to predict the future trend of equipment health index, but also to accurately and reliably estimate the remaining service life of equipment by simultaneously predicting the evolution of underlying characteristic parameters, thus providing a forward-looking time basis for preventive maintenance.

[0063] First, the model's input is the device's historical time-series data, including historical health status sequences and corresponding historical heterogeneous feature parameter sequences. These two related sequences are simultaneously input into the encoder. The encoder consists of multiple layers of LSTM units, whose function is to read and compress the information of the input sequence step by step, ultimately forming a fixed-dimensional context vector. This vector condenses all the key features and patterns of the device's historical degradation process.

[0064] Next, the final state output by the encoder (i.e., the context vector) is used as the initial state of the decoder. The decoder is also composed of LSTM units, which predict the sequence at multiple future time steps in an autoregressive manner based on the historical memory provided by the encoder. The decoder outputs two prediction sequences simultaneously: a future first health prediction sequence and a future feature parameter prediction sequence.

[0065] This dual-output design allows the model to not only directly predict the macroscopic health status, but also simultaneously predict the evolution of microscopic characteristic parameters. The two can be mutually verified, enhancing the reliability of the prediction.

[0066] The first health prediction sequence and the feature parameter prediction sequence are input into the fully connected layer to determine the remaining lifespan, specifically: Based on the predicted sequence of feature parameters, a second health prediction sequence is obtained through the health assessment model; By simultaneously comparing the first health prediction sequence and the second health prediction sequence, and taking the smaller value, the health prediction sequence is determined. The remaining service life is then determined based on the time point when the health prediction sequence first falls below the failure threshold.

[0067] First, the predicted sequence of feature parameters output from the decoder is input again into the previously established health assessment model to recalculate a new second health prediction sequence. Then, using the trained health assessment model, the predicted future feature parameters are inversely evaluated to obtain an independent health prediction trajectory.

[0068] The first health prediction sequence directly output by the time series model is compared point-by-point simultaneously with the second health prediction sequence obtained in the above manner. For each moment, the smaller value is selected, meaning the more pessimistic (lower) health value predicted by the two sequences at that moment is chosen. Finally, these two sequences are merged to form a single, definitive final health prediction sequence. This sequence will be used to determine the remaining useful life (RUL), which is the time elapsed from the current moment until the final health prediction sequence first falls below a predefined equipment failure threshold.

[0069] By employing a fusion strategy that takes the smaller value, the system automatically tends to adopt the more conservative result (i.e., the one that assumes faster equipment degradation and worse health) between the two prediction paths. This effectively reduces the risk of prediction failure caused by the optimistic bias that may exist in a single prediction model, making the final remaining service life estimate more robust and reliable.

[0070] As a preferred embodiment of the present invention, the comprehensive priority score of the multidimensional decision matrix... Specifically: , in, The urgency score is calculated based on the health index and the remaining lifespan. The criticality score is determined based on the device's pre-defined importance level within the system. The impact score is determined based on the assessment of the impact of equipment failure on related systems. , , The preset weighting coefficients, and .

[0071] The main objective is to build a decision-making framework that quantitatively assesses maintenance needs from multiple dimensions, thereby organically integrating technical predictive indicators (Health Index HI, Remaining Useful Life RUL) with managerial and systemic factors such as equipment importance and the potential impact of failures, ultimately generating a comprehensive and actionable maintenance priority list.

[0072] It should be noted that urgency directly reflects the degree to which the equipment needs maintenance from a technical perspective. Preferably, it is calculated based on a combination of the percentage decrease in the Health Index (HI) and the remaining useful life (RUL).

[0073] For example, if a baseline RUL is set (e.g., 60 days), E can be calculated using the following formula: The lower the health level and the shorter the RUL (Right to Limit) of a device, the higher its urgency score will be.

[0074] Criticality is a relatively static, domain-knowledge-based preset value. During system initialization or equipment entry, a criticality level is assigned to each device based on its functional positioning within the park's energy system. For example, the main chiller unit, which ensures the overall production of the plant, is assigned C=10, the standby pump serving a specific area is assigned C=5, and so on. This score reflects the inherent importance of the equipment itself.

[0075] Impact assessment determines the potential cascading effects or losses on the entire system's operation should a device malfunction. This can be analyzed based on the system's physical topology or logical dependencies. For example, it might assess whether a device malfunction would lead to the shutdown of related production lines, or whether it would affect the peak regulation capacity of the energy storage system, resulting in electricity cost losses. Appropriate scores are assigned based on the severity and scope of the impact.

[0076] Weighting coefficient , , These three coefficients determine the relative weights of technological urgency, inherent importance, and systemic impact in the final decision. Their settings can be dynamically adjusted according to the park's operational strategies at different times. For example, during peak production seasons, the criticality level can be appropriately increased to improve reliability. and influence The weight of [something]; during the maintenance season, the urgency level can be increased to address backlogged issues. The weight.

[0077] First, this multi-dimensional decision-making model deeply integrates technical parameters and management factors, making the ranking results more aligned with the comprehensive goals of actual operation and maintenance. By introducing configurable weight coefficients, the system possesses strategic adaptability and management flexibility. Maintenance personnel can adjust the weights according to the work priorities at different stages (such as ensuring production, controlling costs, and eliminating potential hazards), ensuring that the generated list always remains consistent with current management objectives. Finally, the calculated comprehensive priority score P provides a fair, comparable, and quantifiable ranking basis for all equipment to be maintained. This scientifically guides the prioritization of limited maintenance resources (manpower, time, spare parts) on equipment with the greatest impact on the overall reliability and economy of the system, maximizing the return on investment of preventative maintenance and effectively avoiding one-sidedness and arbitrariness in maintenance decisions.

[0078] As a preferred embodiment of the present invention, a maintenance window is recommended, specifically including: For each maintenance task, determine its theoretical maintainable life. , ],in The result is obtained by subtracting the safety buffer period from the current time and the predicted remaining useful life of the equipment. The production plan information is obtained, which divides future time into peak production unavailable periods, planned downtime ideal available periods, and normal production condition available periods. During the theoretically maintainable period [ , Within the scope of the task, a sliding window traversal algorithm is used to find a continuous time window with a length equal to the estimated working hours of the task, so as to minimize the sum of the comprehensive inconvenience costs of each time period within the window, and to determine the candidate maintenance window. Different inconvenience cost values ​​are assigned to different types of time periods.

[0079] The core objective of this implementation method is to intelligently match and optimize the generated maintenance requirement list based on the technical status of the equipment with the actual production and operation rhythm of the park, aiming to find the maintenance execution time with the least interference to continuous production activities, thereby achieving synergy between technical maintenance and operational production.

[0080] First, for each task in the maintenance priority list, a theoretical maintainable period is determined based on the predicted remaining useful life (RUL) of the corresponding equipment. , ].in, It is calculated by adding the predicted RUL to the current time and subtracting a preset safety buffer period. This safety buffer period is set to address the uncertainty of the prediction itself, reserving a necessary safe time window for maintenance operations to ensure that maintenance can be completed before a failure occurs.

[0081] Secondly, detailed production plan information for the future preset period is obtained from upper-level management platforms such as Manufacturing Execution System (MES). This information pre-divides future time resources into different types of time periods, typically including: peak production unavailable periods (such as full-load day shifts), ideally available planned downtime periods (such as statutory holidays or scheduled maintenance days), and periods available under normal production conditions (such as night shifts or low-load operation periods).

[0082] Finally, in the above-determined [ , Within the theoretically maintainable period, a sliding window traversal algorithm is used for optimization search. The algorithm simulates a time window of length equal to the estimated working hours of the maintenance task, sliding daily (or at smaller time granularities) within the theoretically maintainable period. For each possible window position, the sum of the comprehensive inconvenience costs corresponding to each time period covered by the window is calculated. Here, the inconvenience cost is a pre-defined quantified value based on the time period type. For example, assign extremely high cost values ​​(e.g., 100) to unavailable periods during peak production times, zero cost to ideally available periods during planned downtime, and medium cost (e.g., 1) to available periods under normal production conditions. The algorithm will automatically select the continuous time window with the minimum total inconvenience cost as the candidate maintenance window for the maintenance task.

[0083] This step couples and optimizes the remaining useful life of the equipment—a rigid time budget determined by its technical condition—with the production system's operational schedule—a flexible time resource determined by business needs.

[0084] Without considering production planning, even the optimal maintenance priority ranking can lead to actual production disruptions if scheduled for execution at the wrong time, significantly diminishing the value of predictive maintenance. By introducing an optimization algorithm based on inconvenience cost, maintenance windows located during production slumps or planned downtime can be identified and recommended. This resolves the conflict between maintenance needs and production assurance, ensuring long-term reliable equipment operation while minimizing disruption to core production processes.

[0085] Specifically, after determining the candidate maintenance window, the following also applies: For each maintenance task in the maintenance priority list, an execution window is determined sequentially from the corresponding candidate maintenance window based on the comprehensive priority score of the multidimensional decision matrix, from highest to lowest. Perform time overlap checks on multiple maintenance tasks, and determine the preventive maintenance execution plan after confirming that there are no time overlaps.

[0086] After identifying candidate maintenance windows for each maintenance task, key resource coordination and conflict resolution steps must be performed to ensure the final plan is practically executable. The main objective is to integrate multiple independent, optimized candidate windows into a globally feasible execution plan that is time-independent and respects the priorities of maintenance tasks.

[0087] First, the tasks are processed according to the generated maintenance priority list, which is sorted in descending order based on the overall priority score P. The system starts with the task with the highest overall priority score in the list and selects a maintenance window from among several candidate maintenance windows as its pre-execution time. The selection strategy can be to directly choose the window with the lowest inconvenience cost, or to combine it with other strategies (such as earlier-arriving windows).

[0088] Next, the next highest priority task is processed. Before assigning a window to it, an execution time overlap check must be performed: that is, to determine whether the execution window selected for this task overlaps with the execution times of all identified higher priority tasks. If there is an overlap, it means that maintenance resources (such as technicians of specific skills, specialized tools, or systems that must be shut down) will be conflict-occupied within the same time period. In this case, another non-overlapping window will be automatically selected for this task from its candidate window set.

[0089] If all candidate windows conflict with higher-priority tasks, the system can trigger an alarm, prompting manual intervention or adjustments to the production plan. This process iterates until all tasks in the list are assigned to a single, non-overlapping execution window. Finally, all tasks and their assigned execution times are summarized to form a complete, immediately deployable preventative maintenance execution plan.

[0090] First, it ensures that high-priority maintenance needs receive priority in time and resource allocation, maintaining consistency in decision-making logic. Second, through mandatory time overlap verification and conflict resolution mechanisms, it eliminates maintenance issues caused by scheduling conflicts between multiple tasks due to planning oversights. The generated plans are pre-coordinated and highly operable, enabling preventative maintenance work to be carried out smoothly and systematically.

[0091] The second invention provides a preventive maintenance scheduling system for a park integrated energy system, used to implement the method described above, the system comprising: The data acquisition and preprocessing module is used to collect operating parameters of multiple devices in the HVAC system and energy storage system within the park. The operating parameters include at least heterogeneous characteristic parameters reflecting the mechanical degradation, electrical degradation and thermal degradation of the equipment. The health assessment and prediction module, based on the collected heterogeneous feature parameters, obtains a health index that uniformly represents the health status of each device through a preset health assessment model, and predicts the remaining service life of each device based on a time-series prediction model. The intelligent decision-making and sorting module is used to trigger maintenance decisions based on the obtained health index and remaining service life, and to sort the equipment to be maintained based on a multi-dimensional decision matrix including urgency, criticality and impact, and generate a maintenance priority list. The production awareness scheduling optimization module is used to obtain production plan information within a future preset period. For each maintenance task in the maintenance priority list, it optimizes the task based on the remaining service life of the equipment corresponding to the task and the production plan information, recommends maintenance windows, and generates a preventive maintenance execution plan that includes specific execution times. The output of the health assessment and prediction module is connected to the input of the intelligent decision-making and ranking module, and the output of the intelligent decision-making and ranking module and the data from the external production system are connected to the input of the production perception scheduling optimization module.

[0092] Therefore, it can achieve any effect in the preventive maintenance scheduling method of the park's integrated energy system, which will not be elaborated here.

[0093] A third aspect of the present invention provides a computing device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method.

[0094] Therefore, it can achieve any effect in the preventive maintenance scheduling method of the park's integrated energy system, which will not be elaborated here.

[0095] For any parts not mentioned in this invention, existing technologies can be used or referenced.

[0096] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0097] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A preventive maintenance scheduling method for an integrated energy system in a park, characterized in that, include: The system collects operating parameters from multiple devices in the HVAC and energy storage systems within the park. These operating parameters include at least heterogeneous characteristic parameters reflecting the mechanical, electrical, and thermal degradation of the devices. Based on the collected heterogeneous feature parameters, a health index is obtained through a preset health assessment model to uniformly characterize the health status of each device, and the remaining service life of each device is predicted based on a time-series prediction model. Based on the obtained health index and remaining service life, a maintenance decision is triggered, and the equipment to be maintained is sorted according to a multi-dimensional decision matrix that includes urgency, criticality and impact, to generate a maintenance priority list. Obtain production plan information for the future preset period, and for each maintenance task in the maintenance priority list, optimize it based on the remaining service life of the equipment corresponding to the task and the production plan information, recommend maintenance windows, and generate a preventive maintenance execution plan that includes specific execution times.

2. The method according to claim 1, characterized in that, The heterogeneous characteristic parameters include vibration spectrum characteristics, current harmonic characteristics, and condenser differential pressure characteristics from the chiller unit, as well as internal resistance characteristics and voltage-current imbalance characteristics from the energy storage battery.

3. The method according to claim 1, characterized in that, The health assessment model adopts a weighted fusion model, which is implemented as follows: HI stands for Health Index. For the first Preset weights for each feature parameter, For the first The normalized degradation deviation of each feature parameter As a nonlinear correction factor related to the total degradation degree, if At that time, ,otherwise .

4. The method according to claim 3, characterized in that, The time-series prediction model is a long short-term memory network model based on an encoder-decoder architecture. Input the historical health status sequence and feature parameter sequence into the encoder. The final state of the encoder is input into the decoder for decoding and prediction, and the first health prediction sequence and feature parameter prediction sequence are output. The first health prediction sequence and the feature parameter prediction sequence are input into the fully connected layer to determine the remaining lifespan.

5. The method according to claim 4, characterized in that, The first health prediction sequence and the feature parameter prediction sequence are input into the fully connected layer to determine the remaining lifespan, specifically: Based on the predicted sequence of feature parameters, a second health prediction sequence is obtained through the health assessment model; By simultaneously comparing the first health prediction sequence and the second health prediction sequence, and taking the smaller value, the health prediction sequence is determined. The remaining service life is then determined based on the time point when the health prediction sequence first falls below the failure threshold.

6. The method according to claim 1, characterized in that, The comprehensive priority score of the multidimensional decision matrix Specifically: ， in, The urgency score is calculated based on the health index and the remaining lifespan. The criticality score is determined based on the device's pre-defined importance level within the system. The impact score is determined based on the assessment of the impact of equipment failure on related systems. , , The preset weighting coefficients, and .

7. The method according to claim 1, characterized in that, Recommended maintenance windows include: For each maintenance task, determine its theoretical maintainable life. , ],in The result is obtained by subtracting the safety buffer period from the current time and the predicted remaining useful life of the equipment. The production plan information is obtained, which divides future time into peak production unavailable periods, planned downtime ideal available periods, and normal production condition available periods. During the theoretically maintainable period [ , Within the scope of the task, a sliding window traversal algorithm is used to find a continuous time window with a length equal to the estimated working hours of the task, so as to minimize the sum of the comprehensive inconvenience costs of each time period within the window, and to determine the candidate maintenance window. Different inconvenience cost values ​​are assigned to different types of time periods.

8. The method according to claim 7, characterized in that, After determining the candidate maintenance window, the following are also included: For each maintenance task in the maintenance priority list, an execution window is determined sequentially from the corresponding candidate maintenance window based on the comprehensive priority score of the multidimensional decision matrix, from highest to lowest. Perform time overlap checks on multiple maintenance tasks, and determine the preventive maintenance execution plan after confirming that there are no time overlaps.

9. A preventive maintenance and dispatching system for an integrated energy system in a park, characterized in that, The system for implementing the method of any one of claims 1 to 8 comprises: The data acquisition and preprocessing module is used to collect operating parameters of multiple devices in the HVAC system and energy storage system within the park. The operating parameters include at least heterogeneous characteristic parameters reflecting the mechanical degradation, electrical degradation and thermal degradation of the equipment. The health assessment and prediction module, based on the collected heterogeneous feature parameters, obtains a health index that uniformly represents the health status of each device through a preset health assessment model, and predicts the remaining service life of each device based on a time-series prediction model. The intelligent decision-making and sorting module is used to trigger maintenance decisions based on the obtained health index and remaining service life, and to sort the equipment to be maintained based on a multi-dimensional decision matrix including urgency, criticality and impact, and generate a maintenance priority list. The production awareness scheduling optimization module is used to obtain production plan information within a future preset period. For each maintenance task in the maintenance priority list, it optimizes the task based on the remaining service life of the equipment corresponding to the task and the production plan information, recommends maintenance windows, and generates a preventive maintenance execution plan that includes specific execution times. The output of the health assessment and prediction module is connected to the input of the intelligent decision-making and ranking module, and the output of the intelligent decision-making and ranking module and the data from the external production system are connected to the input of the production perception scheduling optimization module.

10. A computing device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8.

Images