Digital twin driven unit maintenance optimization method and system

Abstract

The present application relates to the field of unit maintenance, in particular to a kind of digital twin driven unit maintenance optimization method and system;The method comprises the following steps: obtaining the historical task data of unit, and identifying typical task mode, constructing task profile model, extracting the task characteristic parameter of typical task mode, generating load influence label;Load influence label and the obtained unit operating state data are used as input, based on sliding window mechanism is analyzed, outputs the life state trajectory and life abnormal data of unit in different task scenarios;Set up digital twin platform, map life state trajectory and life abnormal data to digital twin platform, construct task scheduling diagram structure and multidimensional association graph;Based on multidimensional association graph, analyze the conflict boundary between task execution and maintenance operation, combine resource availability data and task priority parameter, output maintenance recommended window set, generate maintenance optimization strategy.The present application can improve the safety of unit operation.

Description

A digital twin-driven method and system for optimizing unit maintenance Technical Field

[0001] This invention relates to the field of unit maintenance, and specifically to a digital twin-driven method and system for optimizing unit maintenance. Background Technology

[0002] With the accelerated advancement of intelligentization, digital twin technology is gradually becoming an important supporting means in the fields of manufacturing and operation and maintenance management. Digital twins realize the mapping and tracking of equipment operation by constructing virtual models that are highly consistent with real physical equipment in terms of status, behavior, task execution, etc., and have multiple capabilities such as real-time monitoring, predictive diagnosis, and optimization decision-making. In the field of typical units, digital twin platforms are used for modeling.

[0003] Chinese Patent Publication No. CN115392577A discloses a method, apparatus, and equipment for optimizing the composite maintenance strategy of wind turbines in offshore wind farms. The method includes: predicting the operating status of each wind turbine in the offshore wind farm based on monitoring data; optimizing the maintenance strategy based on the predicted operating status of each wind turbine to obtain the optimal preventive maintenance strategy for each wind turbine; and selecting a target combination of turbines to be maintained and optimizing its maintenance strategy based on the current operating status and actual maintenance resource constraints of each wind turbine to obtain the optimal actual maintenance strategy for the target turbines to be maintained, thereby achieving preventive and actual maintenance of each wind turbine in the offshore wind farm.

[0004] In existing technologies, most methods fail to fully consider the relationship between task load patterns and lifetime changes, resulting in a lack of scenario adaptability in maintenance strategies. Traditional scheduling methods often lack comprehensive coordination between task scheduling and maintenance strategies. In practical applications, it is difficult to form an operable maintenance recommendation window and optimization scheme, which are problems we need to solve. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a digital twin-driven unit maintenance optimization method and system.

[0006] The technical solution of this invention: a digital twin-driven unit maintenance optimization method, comprising the following steps:

[0007] S1. Obtain historical task data of the unit, identify typical task patterns, construct a task profile model, extract task feature parameters of typical task patterns, and generate corresponding load impact labels.

[0008] S2. Using the load impact label and the acquired unit operating status data as input, analyze based on the sliding window mechanism to output the unit's life status trajectory and life anomaly data under different task scenarios.

[0009] S3. Set up a digital twin platform, map the lifespan status trajectory and lifespan anomaly data to the digital twin platform, and construct a task scheduling chart structure and a multi-dimensional correlation graph.

[0010] S4. Based on the analysis of the conflict boundary between task execution and maintenance operations using a multi-dimensional correlation graph, and combined with the resource availability data and task priority parameters within the unit, output a set of maintenance recommendation windows and generate maintenance optimization strategies.

[0011] Preferably, the process of acquiring historical task data of the unit, identifying typical task patterns, and constructing a task profile model includes:

[0012] By acquiring historical task data generated by the unit during historical task execution, the historical task data is constructed into a task data sequence in chronological order. The task data sequence is segmented based on a sliding window mechanism to obtain multiple historical task cycles, and local operation segments corresponding to each historical task cycle are extracted. Multidimensional operation features are extracted for each local operation segment, and task feature vectors are formed by combining the multidimensional operation features extracted from the same local operation segment. A task feature vector set is constructed using the task feature vectors of all local operation segments. Cluster analysis is performed on the task feature vector set to identify representative typical task patterns, and the multidimensional operation features extracted from each typical task pattern are summarized to construct a task profile model.

[0013] Preferably, the process of extracting task feature parameters of typical task patterns and generating corresponding load impact labels includes:

[0014] For the various typical task modes included in the task profile model, based on the statistically representative multidimensional operational features in the task feature vector set, task feature parameters are extracted as task load feature description items.

[0015] Each type of task load characteristic description item is input into a preset task rule set. The task rule set is constructed based on the historical operating data and life tag samples within the unit through multidimensional regression statistical analysis or fuzzy logic reasoning, and outputs the task impact intensity index range and its task level. Through the task impact intensity index range that various typical task modes conform to, the task level is bound to the corresponding typical task mode, and a load impact label is generated.

[0016] Preferably, the process of taking load impact labels and acquired unit operating status data as input, analyzing them based on a sliding window mechanism, and outputting the unit's lifespan trajectory and lifespan anomaly data under different task scenarios is as follows:

[0017] The operating status data acquired during unit operation is continuously collected and processed in time synchronization. The operating status data includes equipment temperature rise data, vibration intensity data, speed fluctuation data, energy consumption data, and task trigger records.

[0018] Based on the sliding window mechanism, the running status data is segmented to form status data segments with time series characteristics; each status data segment is associated and bound with the corresponding load impact label to construct a set of status evolution samples.

[0019] Based on the state evolution sample set and each typical task scenario, multiple state data segments are aggregated and compared in chronological order. The continuous change characteristics of state parameters within each task cycle are extracted, state feature nodes are constructed, and they are correlated in chronological order according to the task cycle to form a state change trajectory. The state change trajectory is then analyzed based on deep learning combined with multi-dimensional operational features.

[0020] Preferably, the process of analyzing state change trajectories based on deep learning and multi-dimensional operational features includes:

[0021] Identify abnormal lifespan time intervals, extract the rate of change, range of change, and abrupt change time points of task characteristic parameters within the abnormal lifespan time intervals, and extract the remaining available time within the corresponding task cycle of the abnormal lifespan time intervals, which is recorded as the degradation remaining time. The rate of change, range of change, abrupt change time points, and degradation remaining time are recorded as abnormal lifespan data, and the state change trajectory corresponding to the abnormal lifespan time intervals is recorded as the lifespan state trajectory.

[0022] Preferably, the process of setting up a digital twin platform, mapping lifespan status trajectories and lifespan anomaly data to the digital twin platform, and constructing a task scheduling chart structure and a multi-dimensional correlation graph includes:

[0023] The lifespan status trajectory and lifespan anomaly data are correlated to each virtual task cycle node in the digital twin platform. Based on the time sequence of each task cycle, a multi-time period status mapping chain for the unit operation process is established, and the multi-time period status mapping chain is used as the basic time axis.

[0024] Based on the task trigger records, the corresponding status feature nodes within each task cycle are aggregated to construct task nodes that correspond one-to-one with the task cycle, and lifespan anomaly data is bound to the corresponding task nodes.

[0025] The changing parameters that match the abnormal lifespan data with each task node in the lifespan state trajectory are recorded as the state attribute markers of the task nodes; the remaining degradation time in the abnormal lifespan data is used as the lifespan label in the current task cycle and marked to the corresponding task node.

[0026] The task cycle is analyzed based on the remaining degradation time, the upper threshold and lower threshold of the available scheduling time for task nodes are identified, and the scheduling time boundary is constructed.

[0027] Based on the basic timeline, task nodes are arranged in chronological order, and task edges are constructed between task nodes. Task nodes are then connected through task edges to build a task scheduling graph. Status attribute markers, lifetime labels, and scheduling time boundaries are bound to each corresponding task node and edge, and mapped to the virtual machine group model in the digital twin platform to construct a multi-dimensional association graph.

[0028] Preferably, the process of generating a maintenance optimization strategy by analyzing the conflict boundary between task execution and maintenance operations based on multi-dimensional correlation graph analysis, combining resource availability data and task priority parameters within the unit, and outputting a set of maintenance recommendation windows includes:

[0029] Obtain the set of task nodes, the set of lifetime labels, and the scheduling time boundary in the multidimensional correlation graph; identify the task node group that has a risk of conflict between the scheduling time boundary and the lifetime state trajectory; and determine the corresponding lifetime conflict time period.

[0030] Based on the task priority parameter, the task nodes are sorted, and the task nodes with lower priority that are within the lifespan conflict period are marked as compressible task groups. The yieldable time segments in their scheduling time boundaries are extracted to construct a set of releaseable time windows.

[0031] Based on resource availability data, identify time segments within the set of releaseable time windows that meet maintenance conditions, output a set of recommended maintenance windows as candidates, send compressible task groups, transferable time segments, and recommended maintenance window sets to maintenance personnel, and combine the compressible task groups, transferable time segments, and recommended maintenance window sets to generate corresponding maintenance optimization scheduling schemes.

[0032] This invention also discloses a digital twin-driven unit maintenance optimization system, including a management center, which is communicatively connected to a unit data acquisition and analysis module, a unit anomaly module, a unit task module, and a unit maintenance optimization module.

[0033] The unit data acquisition and analysis module is used to acquire historical task data of the unit, identify typical task patterns, construct task profile models, extract task feature parameters of typical task patterns, and generate corresponding load impact labels.

[0034] The unit anomaly module takes load impact tags and acquired unit operating status data as input, performs analysis based on a sliding window mechanism, and outputs the unit's life status trajectory and life anomaly data under different task scenarios.

[0035] The unit task module is used to set up a digital twin platform, map the life status trajectory and life anomaly data to the digital twin platform, and construct a task scheduling chart structure and a multi-dimensional correlation map.

[0036] The unit maintenance optimization module is used to analyze the conflict boundary between task execution and maintenance operations based on multi-dimensional correlation graph analysis. Combining the unit's resource availability data and task priority parameters, it outputs a set of maintenance recommendation windows and generates maintenance optimization strategies.

[0037] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: the task profile model and load impact label constructed based on typical task modes can perform scenario-based modeling of the state evolution process of the unit under various operating tasks, realize multi-dimensional analysis of the life consumption trend within the task cycle, and thus provide fine-grained support for state monitoring and task-level life management; by mapping the life state trajectory, life anomaly data and scheduling time boundary to the multi-dimensional correlation graph constructed in the digital twin platform, the dynamic identification of task scheduling chart and maintenance timing conflict is realized, and combined with factors such as task priority and resource availability, a set of maintenance recommendation windows and maintenance optimization strategies are generated to realize dynamic maintenance scheduling and intelligent decision-making oriented towards life evolution. Attached Figure Description

[0038] Figure 1 is a flowchart of an embodiment of the present invention. Detailed Implementation

[0039] Example 1, as shown in Figure 1, a digital twin-driven unit maintenance optimization method proposed in this invention includes the following steps:

[0040] S1. Obtain historical task data of the unit, identify typical task patterns, construct a task profile model, extract task feature parameters of typical task patterns, and generate corresponding load impact labels.

[0041] S2. Using the load impact label and the acquired unit operating status data as input, analyze based on the sliding window mechanism to output the unit's life status trajectory and life anomaly data under different task scenarios.

[0042] S3. Set up a digital twin platform, map the lifespan status trajectory and lifespan anomaly data to the digital twin platform, and construct a task scheduling chart structure and a multi-dimensional correlation graph.

[0043] S4. Based on the analysis of the conflict boundary between task execution and maintenance operations using a multi-dimensional correlation graph, and combined with the resource availability data and task priority parameters within the unit, output a set of maintenance recommendation windows and generate maintenance optimization strategies.

[0044] It should be further explained that, in the specific implementation process, the process of acquiring historical task data of the unit, identifying typical task patterns, constructing a task profile model, extracting task characteristic parameters of typical task patterns, and generating corresponding load impact labels includes:

[0045] By acquiring historical task data generated by the unit during the execution of historical tasks, the historical task data includes historical operation logs, scheduling plan information, and task execution records.

[0046] Historical task data is constructed into a task data sequence in chronological order. The task data sequence is segmented using a sliding window mechanism to obtain multiple historical task cycles, and local operation segments corresponding to each historical task cycle are extracted. Multidimensional operation features are extracted for each local operation segment, including average load intensity, load fluctuation index, start / stop frequency, temperature rise rate, and operation rhythm parameters. A task feature vector is formed from the multidimensional operation features extracted from the same local operation segment. A task feature vector set is constructed using the task feature vectors from all local operation segments. Cluster analysis is performed on the task feature vector set, using Euclidean distance or cosine distance combined with K-means or DBSCAN algorithms. Each cluster represents a typical task pattern. Representative typical task patterns are identified, and the multidimensional operation features extracted from each typical task pattern are summarized to construct a task profile model.

[0047] It should be noted that the similarity metric uses Euclidean distance or cosine similarity to measure the degree of difference in multidimensional operational features between different local running segments; the clustering analysis can be implemented using K-means or DBSCAN algorithms, constructing multiple clusters based on the feature similarity between tasks, with each cluster representing a typical task pattern; after clustering is completed, the central feature vector of each task pattern is used as the representative task profile to form a standard task type.

[0048] For the various typical task modes included in the task profile model, based on the statistically representative multidimensional operation features in the task feature vector set, task feature parameters are extracted as task load feature description items; the task feature parameters include, but are not limited to, load feature range, start-stop feature frequency and temperature rise feature range.

[0049] Each type of task load characteristic description item is input into a preset task rule set. The task rule set is constructed based on the historical operating data and lifetime label samples within the unit through multidimensional regression statistical analysis or fuzzy logic reasoning, and outputs the task impact intensity index range and its task level. Through the task impact intensity index range that various typical task modes conform to, the task level is bound to the corresponding typical task mode, and a load impact label is generated.

[0050] Specifically, the historical operating data and lifespan tag samples are recorded data within the unit and can be directly obtained. The historical operating data includes, but is not limited to, historical equipment temperature rise values, historical loads, and historical start-up and shutdown frequencies; the lifespan tag samples include, but are not limited to, service life and number of tasks.

[0051] It should be further explained that, in the specific implementation process, the process of taking the load impact label and the acquired unit operating status data as input, analyzing them based on the sliding window mechanism, and outputting the unit's life status trajectory and life anomaly data under different task scenarios is as follows:

[0052] The operating status data acquired during the operation of the unit is continuously collected and processed in time synchronization. The operating status data includes, but is not limited to, equipment temperature rise data, vibration intensity data, speed fluctuation data, energy consumption data, and task trigger records.

[0053] Based on the sliding window mechanism, the running status data is segmented to form status data segments with time series characteristics; each status data segment is associated and bound with the corresponding load impact label to construct a set of status evolution samples.

[0054] Based on the state evolution sample set and each typical task scenario, multiple state data segments are aggregated and compared in chronological order. The continuous change characteristics of state parameters within each task cycle are extracted, state feature nodes are constructed, and they are correlated in chronological order according to the task cycle to form a state change trajectory. The state change trajectory is then analyzed based on deep learning combined with multi-dimensional operational features.

[0055] Identify abnormal lifespan time intervals, extract the rate of change, range of change, and abrupt change time points of task characteristic parameters within the abnormal lifespan time intervals, and extract the remaining available time within the corresponding task cycle of the abnormal lifespan time intervals, which is recorded as the degradation remaining time. The rate of change, range of change, abrupt change time points, and degradation remaining time are recorded as abnormal lifespan data, and the state change trajectory corresponding to the abnormal lifespan time intervals is recorded as the lifespan state trajectory.

[0056] It should be further explained that, in the specific implementation process, the process of setting up a digital twin platform, mapping the lifespan status trajectory and lifespan anomaly data to the digital twin platform, and constructing the task scheduling chart structure and multi-dimensional correlation graph is as follows:

[0057] The lifespan status trajectory and lifespan anomaly data are correlated to each virtual task cycle node in the digital twin platform. The virtual task cycle node is set in the digital twin platform according to the task cycle of the unit. Based on the time sequence of each task cycle, a multi-time period status mapping chain of the unit operation process is established, and the multi-time period status mapping chain is used as the basic time axis.

[0058] Based on the task trigger records, the corresponding status feature nodes within each task cycle are aggregated to construct task nodes that correspond one-to-one with the task cycle, and lifespan anomaly data is bound to the corresponding task nodes.

[0059] The changing parameters that match the abnormal lifespan data with each task node in the lifespan status trajectory are recorded as the status attribute markers of the task nodes. The status attribute markers are used to describe the set of equipment status features that describe the current operating status of the equipment. The remaining degradation time in the abnormal lifespan data is used as the lifespan label in the current task cycle and marked to the corresponding task node.

[0060] The task cycle is analyzed based on the remaining degradation time, the upper threshold and lower threshold of the available scheduling time for task nodes are identified, and the scheduling time boundary is constructed.

[0061] Task nodes are arranged in chronological order, and task edges are constructed between them. Task nodes are then connected through these task edges to build a task scheduling graph. Status attribute markers, lifetime labels, and scheduling time boundaries are bound to the corresponding task nodes and edges, and mapped to the virtual machine group model in the digital twin platform to construct a multidimensional association graph.

[0062] It should be further explained that, in the specific implementation process, the process of generating maintenance optimization strategies by analyzing the conflict boundary between task execution and maintenance operations based on multi-dimensional correlation graph analysis, combined with resource availability data and task priority parameters within the unit, outputting a set of maintenance recommendation windows, is as follows:

[0063] Obtain the set of task nodes, the set of lifetime labels, and the scheduling time boundary in the multidimensional correlation graph; identify the task node group that has a risk of conflict between the scheduling time boundary and the lifetime state trajectory; and determine the corresponding lifetime conflict time period.

[0064] Based on the task priority parameter, the task nodes are sorted, and the task nodes with lower priority that are within the lifespan conflict period are marked as compressible task groups. The yieldable time segments in their scheduling time boundaries are extracted to construct a set of releaseable time windows.

[0065] The resource availability data includes equipment downtime conditions, standby equipment status, and available personnel resource information. Based on the resource availability data, time segments that meet the maintenance conditions in the set of release time windows are identified. The maintenance conditions are the standard maintenance duration and task level classification parameters preset in the digital twin platform. The set of recommended maintenance windows is output as a candidate. The set of recommended maintenance windows includes the standard maintenance operation duration and maintenance level requirements.

[0066] The compressible task group, the transferable time segment, and the maintenance recommendation window set are sent to the maintenance personnel. Based on the compressible task group, the transferable time segment, and the maintenance recommendation window set, a corresponding maintenance optimization schedule plan is generated.

[0067] Example 2: The digital twin-driven unit maintenance optimization system proposed in this invention is applied to the digital twin-driven unit maintenance optimization method described in Example 1. Specifically, it includes a management center, which is communicatively connected to a unit data acquisition and analysis module, a unit anomaly module, a unit task module, and a unit maintenance optimization module.

[0068] The unit data acquisition and analysis module is used to acquire historical task data of the unit, identify typical task patterns, construct task profile models, extract task feature parameters of typical task patterns, and generate corresponding load impact labels.

[0069] The unit anomaly module takes load impact tags and acquired unit operating status data as input, performs analysis based on a sliding window mechanism, and outputs the unit's life status trajectory and life anomaly data under different task scenarios.

[0070] The unit task module is used to set up a digital twin platform, map the life status trajectory and life anomaly data to the digital twin platform, and construct a task scheduling chart structure and a multi-dimensional correlation map.

[0071] The unit maintenance optimization module is used to analyze the conflict boundary between task execution and maintenance operations based on multi-dimensional correlation graph analysis. Combining the unit's resource availability data and task priority parameters, it outputs a set of maintenance recommendation windows and generates maintenance optimization strategies.

[0072] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A digital twin-driven method for optimizing unit maintenance, characterized in that, Includes the following steps: S1. Obtain historical task data of the unit, identify typical task patterns, construct a task profile model, extract task feature parameters of typical task patterns, and generate corresponding load impact labels. S2. Using the load impact label and the acquired unit operating status data as input, analyze the data based on the sliding window mechanism to output the unit's life status trajectory and life anomaly data under different task scenarios; S3. Set up a digital twin platform to map the life status trajectory and life anomaly data to the digital twin platform, and construct a task scheduling chart structure and a multi-dimensional correlation graph. S4. Based on the multi-dimensional correlation graph analysis, the conflict boundary between task execution and maintenance operations is analyzed. Combined with the resource availability data and task priority parameters within the unit, a set of maintenance recommendation windows is output to generate maintenance optimization strategies. The process of acquiring historical task data of the generator set, identifying typical task patterns, and constructing a task profile model includes: acquiring historical task data generated by the generator set during historical task execution, constructing a task data sequence in chronological order; segmenting the task data sequence based on a sliding window mechanism to obtain multiple historical task cycles, and extracting local operation segments corresponding to each historical task cycle; extracting multi-dimensional operation features for each local operation segment, forming a task feature vector from the multi-dimensional operation features extracted from the same local operation segment, and constructing a task feature vector set from the task feature vectors of all local operation segments; and then processing the task... Cluster analysis is performed on the feature vector set to identify representative typical task patterns. Based on the multi-dimensional operational features extracted from each typical task pattern, a task profile model is constructed. The process of extracting task feature parameters for typical task patterns and generating corresponding load impact labels includes: for each typical task pattern included in the task profile model, task feature parameters are extracted as task load feature descriptions based on statistically representative multi-dimensional operational features from the task feature vector set; each type of task load feature description is input into a preset task rule set, which is based on historical operational data and lifetime label samples within the unit, and uses multiple... The process involves constructing a task impact intensity index range and its level using dimensional regression statistical analysis or fuzzy logic reasoning. By identifying the task impact intensity index ranges that correspond to various typical task patterns, the task level is bound to the corresponding typical task pattern, generating a load impact label. Using the load impact label and the acquired unit operating status data as input, an analysis is performed based on a sliding window mechanism to output the unit's lifespan trajectory and lifespan anomaly data under different task scenarios. This process involves continuously collecting and time-synchronizing the operating status data acquired during unit operation. The operating status data includes equipment temperature rise data, vibration intensity data, and speed fluctuation data. Energy consumption data and task trigger records are collected. Based on a sliding window mechanism, the operating status data is segmented to form state data fragments with time series characteristics. Each state data fragment is associated and bound with the corresponding load impact label to construct a state evolution sample set. Based on the state evolution sample set and each typical task scenario, multiple state data fragments are aggregated and compared in chronological order to extract the continuous change characteristics of state parameters within each task cycle, construct state feature nodes, and associate them in chronological order of the task cycle to form a state change trajectory. The state change trajectory is then analyzed based on deep learning combined with multi-dimensional operating features.The process of analyzing state change trajectories based on deep learning and multi-dimensional operational features includes: identifying abnormal lifetime time intervals; extracting the rate of change, range of change, and abrupt change time points of task feature parameters within these intervals; extracting the remaining available time within the corresponding task cycle, denoted as the degradation remaining time; recording the rate of change, range of change, abrupt change time points, and degradation remaining time as abnormal lifetime data; and recording the state change trajectory corresponding to the abnormal lifetime time intervals as the lifetime state trajectory.

2. The unit maintenance optimization method driven by digital twin according to claim 1, characterized in that, The process of setting up a digital twin platform and mapping lifespan status trajectories and lifespan anomaly data to the platform, constructing a task scheduling chart structure and a multi-dimensional correlation graph, includes: associating lifespan status trajectories and lifespan anomaly data with virtual task cycle nodes in the digital twin platform; establishing a multi-period state mapping chain for the unit's operation process based on the time sequence of each task cycle, and using this multi-period state mapping chain as the basic time axis; aggregating the corresponding state feature nodes within each task cycle based on task trigger records to construct task nodes that correspond one-to-one with the task cycle, and binding lifespan anomaly data to the corresponding task nodes; and mapping the lifespan anomaly data to the task nodes within the lifespan status trajectories. The corresponding changing parameters are recorded as the status attribute markers of the task nodes; the remaining degradation time in the lifetime anomaly data is used as the lifetime label within the current task cycle and marked to the corresponding task nodes; the task cycle is analyzed based on the remaining degradation time to identify the upper and lower thresholds of the available scheduling time for task nodes and to form scheduling time boundaries; task nodes are constructed in chronological order according to the basic time axis, and task edges are constructed between task nodes. The task nodes are connected through the task edges to construct a task scheduling graph, and the status attribute markers, lifetime labels, and scheduling time boundaries are bound to each corresponding task node and edge, mapped to the virtual machine group model in the digital twin platform, and a multi-dimensional association graph is constructed.

3. The unit maintenance optimization method driven by digital twin according to claim 2, characterized in that, Based on multidimensional correlation graph analysis of the conflict boundary between task execution and maintenance operations, and combined with resource availability data and task priority parameters within the unit, the process of generating a maintenance recommendation window set and a maintenance optimization strategy includes: acquiring the task node set, lifetime label set, and scheduling time boundary from the multidimensional correlation graph; identifying task node groups with conflict risks between scheduling time boundaries and lifetime status trajectories, and determining their corresponding lifetime conflict time periods; sorting task nodes based on task priority parameters, prioritizing and marking lower-priority task nodes within lifetime conflict time periods as compressible task groups, extracting the yieldable time segments from their scheduling time boundaries, and constructing a set of releaseable time windows; identifying time segments in the set of releaseable time windows that meet maintenance conditions based on resource availability data, outputting a set of candidate maintenance recommendation windows, sending the compressible task groups, yieldable time segments, and maintenance recommendation window sets to maintenance personnel, and generating corresponding maintenance optimization scheduling schemes by combining the compressible task groups, yieldable time segments, and maintenance recommendation window sets.

4. A digital twin-driven unit maintenance optimization system, specifically applied to the digital twin-driven unit maintenance optimization method according to any one of claims 1 to 3, comprising a management center, characterized in that, The management center communication connection includes a unit data acquisition and analysis module, a unit anomaly module, a unit task module, and a unit maintenance and optimization module. The unit data acquisition and analysis module is used to acquire historical task data of the unit, identify typical task patterns, construct task profile models, extract task feature parameters of typical task patterns, and generate corresponding load impact labels. The unit anomaly module is used to take the load impact labels and the acquired unit operating status data as input, perform analysis based on a sliding window mechanism, and output the unit's life status trajectory and life anomaly data under different task scenarios. The unit task module is used to set up a digital twin platform, mapping life status trajectories and life anomaly data to the digital twin platform, and constructing a task scheduling chart structure and a multi-dimensional correlation graph. The unit maintenance optimization module is used to analyze the conflict boundary between task execution and maintenance operations based on the multi-dimensional correlation graph, and combined with the unit's resource availability data and task priority parameters, output a set of maintenance recommendation windows and generate maintenance optimization strategies.

Images