Intelligent equipment life cycle information fusion operation and maintenance optimization method

Abstract

The invention discloses an intelligent equipment life cycle information fusion operation and maintenance optimization method, and relates to the technical field of intelligent equipment fusion operation and maintenance, and the method comprises the following steps: based on a judged nested task execution state, extracting state jump features from state data of load, current, vibration and the like collected in an equipment operation process, calculating state change amplitude, duration and frequency by using a time sequence window comparison algorithm, and obtaining a state hopping feature set associated with task switching; and based on the task level mapping table and the state jump feature set, constructing a graph convolutional neural network model, taking a nesting relationship between tasks as a topological structure, inputting each task node and the corresponding state jump feature into the model, and identifying each stage of the life cycle of the intelligent equipment. According to the method, the problem that life cycle division is affected by nested tasks is solved, and accurate state recognition of the intelligent equipment and dynamic optimization of operation and maintenance parameters are realized.

Description

Technical Field

[0001] This invention relates to the field of integrated operation and maintenance technology for intelligent devices, and specifically to a method for optimizing operation and maintenance by integrating lifecycle information of intelligent devices. Background Technology

[0002] Intelligent device lifecycle information fusion and operation optimization refers to the continuous acquisition and fusion of heterogeneous information (such as sensor data, log records, maintenance history, usage behavior, environmental status, etc.) from different sources throughout all stages of the device's lifecycle—including design, manufacturing, installation, operation, maintenance, upgrade, and decommissioning. Leveraging a unified data model and fusion algorithm, this enables holistic perception, accurate diagnosis, and predictive maintenance of the device's status, thereby improving operational efficiency, reliability, and cost control. Existing intelligent device lifecycle information fusion and operation optimization technologies primarily involve deploying sensors and embedded acquisition modules within the device to acquire operational data in real time. This data is then aggregated and cleaned using edge computing or cloud platforms. Subsequently, big data analytics, knowledge graphs, or machine learning are employed to semantically fuse and temporally correlate data from different stages, constructing a global view of the device's lifecycle. Based on the analysis results, optimized operation and maintenance strategies are then developed, including fault warnings, optimized maintenance scheduling, and optimized resource allocation. The entire process typically includes six major stages: data acquisition, data preprocessing, information fusion modeling, status assessment, strategy generation, and intelligent execution. Through the collaboration of these stages, data-driven intelligent operation and maintenance optimization is achieved, reducing downtime due to failures, lowering maintenance costs, and extending equipment lifespan.

[0003] The existing technology has the following shortcomings: During the execution of tasks by intelligent devices, when the device supports nested task execution (i.e., one or more sub-tasks are executed while the main task is not yet completed), the device's operating state will frequently change within the same lifecycle phase. Because existing control systems typically only mark lifecycle phases by the start and end times of tasks and do not synchronize nested task information to the information fusion system, the fusion system cannot perceive the nesting relationships within the task structure. In this situation, the operational data generated by nested sub-tasks will be mistakenly assigned to the lifecycle phase of the main task, resulting in inaccurate lifecycle phase segmentation during state fusion. Therefore, existing intelligent device lifecycle information fusion and operation and maintenance optimization technologies cannot accurately identify and segment lifecycle phases based on the state-change characteristics of the device under nested task execution. The fundamental reason is that these technologies typically rely on the assumption of task temporal continuity to establish the fusion window, lacking the ability to model and recognize the semantics of nested structures, leading to a discrepancy between the lifecycle information fusion results and the actual device operation. This deviation can cause misjudgments of the main task's operating state, thereby affecting the setting of maintenance cycles, the adjustment of control parameters, and the rational allocation of operation and maintenance resources, ultimately reducing the accuracy and reliability of equipment operation and maintenance optimization.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for optimizing operation and maintenance by fusing information throughout the lifecycle of intelligent devices, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing operation and maintenance of intelligent devices through lifecycle information fusion, specifically including the following steps: S1. By obtaining the time information and task type identifier of the task scheduling chain, and combining the timing relationship between the main task and the sub-task, determine whether the smart device is in a nested task execution state, and generate a task hierarchy mapping table to express the nested structure between tasks. S2. Based on the determined nested task execution state, extract state transition features from the load, current, vibration and other state data collected during the device operation, and use the time window comparison algorithm to calculate the state change amplitude, duration and frequency to obtain a set of state transition features associated with task switching. S3. Based on the task hierarchy mapping table and the state transition feature set, construct a graph convolutional neural network model. With the nested relationship between tasks as the topology, input each task node and its corresponding state transition feature into the model to identify each stage of the smart device life cycle. S4. Based on the identified lifecycle stages, construct multiple data windows in the original time series data. Adjust the boundary position and acquisition ratio of the data windows according to the task type and its jump characteristics in each stage to complete the lifecycle stage segmentation process. S5. Based on the segmented lifecycle stage data, construct fusion evaluation indicators, calculate the state offset of each stage, and dynamically adjust the task priority, maintenance cycle and data fusion interval based on the offset to realize the operation and maintenance control optimization of intelligent device lifecycle information fusion.

[0007] Preferably, S1 is as follows: By reading the task execution start time, task execution end time and task type identifier recorded in the task scheduling chain, a data table of time information and task type identifier of the task scheduling chain is constructed, and the task number is used as an index field for unified management. Based on the time information of the task scheduling chain and the task type identifier data table, calculate the start and end time cross relationship between any two tasks, and determine whether the execution time of one task completely includes the execution time of another task. If it does, and the task type identifiers show a master-slave relationship, then the smart device is judged to be in a nested task execution state. After determining that the task is in a nested execution state, a tree structure encoding sequence is constructed based on the temporal inclusion relationship between the main task and the sub-tasks and the task type identifier, and a task hierarchy mapping table is generated to express the nested structure between tasks in the task scheduling chain.

[0008] Preferably, S2 specifically includes the following steps: S201. Based on the determined nested task execution status, continuously collect load data, current data and vibration data in chronological order during equipment operation, and construct the status data sequence corresponding to the task according to the task number. S202. Set a sliding time window in each state data sequence, calculate the state change amplitude by the difference between the maximum and minimum values ​​in the window, record the time period when the signal is continuously higher than the average value to calculate the state change duration, and count the number of drastic state changes per unit time to calculate the state change frequency. S203. The calculated state change amplitude, state change duration and state change frequency are used as state transition features. An aligned time window is constructed with the task switching time point as the center. All state transition features within the aligned time window are selected to form a set of state transition features associated with task switching, which is used to express the state change behavior during task switching.

[0009] Preferably, S202 specifically refers to: In each state data sequence, a fixed-width sliding time window is set, with a sliding step size smaller than the window width, to ensure that the window continuously covers all data points in the state data sequence, and a timestamp index is created for each window for subsequent comparison. Within each sliding time window, the maximum and minimum values ​​of the state data are obtained, and the state change amplitude is calculated by the difference between the two. At the same time, the arithmetic mean of the state data within the window is calculated, and continuous segments higher than the average value are marked. The length of consecutive segments above the average value in each sliding time window is counted to represent the duration of state change. Cases where the difference between two adjacent data points is greater than a preset threshold are counted to calculate the frequency of state change, thereby constructing a complete set of jump characteristic indicators for each segment of state data.

[0010] Preferably, S3 specifically includes the following steps: S301. Construct a graph structure for a graph convolutional neural network model based on a task hierarchy mapping table. Treat each task in the task hierarchy mapping table as a node in the graph, establish directed connections between nodes according to the nesting relationship between the main task and sub-tasks, and construct a graph data input form with the nesting relationship between tasks as the topology. S302. The graph structure constructed by the task hierarchy mapping table is mapped one-to-one with the state transition features of each task node in the state transition feature set. Each task node and its corresponding state transition features are used as feature vectors and input into the graph convolutional neural network model. Multiple rounds of graph convolution operations are performed to fuse the structural and state features between adjacent tasks. S303. In the output stage of the graph convolutional neural network model, supervised classification training of life cycle stage labels is performed on the high-dimensional feature representations generated by each task node. Based on the feature vector jointly represented by the nested structure and state transition features, the various stages of the smart device life cycle are identified.

[0011] Preferably, S303 is as follows: In the output stage of the graph convolutional neural network model, the high-dimensional feature representations obtained after convolution calculation of each task node are summarized into a feature matrix, and a life cycle stage label is assigned to each task node to construct a dataset corresponding to the task nodes and labels for supervised classification training. During supervised classification training, the cross-entropy loss function is used as the optimization objective. The parameter weights of the graph convolutional layer and the classification layer are continuously adjusted through the backpropagation algorithm to minimize the difference between the lifecycle stage labels and the model output labels. After training, the feature vector jointly represented by the nested structure and state transition features is input into the classification layer. Based on the probability distribution results output by the classification layer, the various stages of the smart device's life cycle are identified, and the life cycle stage attributes of the task nodes are updated through the stage identification results.

[0012] Preferably, S4 is as follows: Based on the identified life cycle stages, time indices corresponding to the start and end points of each life cycle stage are extracted from the original time series data. Multiple data windows are constructed in the original time series data according to the time indices, with each data window corresponding to a life cycle stage, and the continuity and order are maintained on the time axis. Based on the task type and its transition characteristics in each life cycle stage, the boundary position of each data window is refined and adjusted. Specifically, the original boundary is moved forward or backward to the time point when the transition behavior occurs, according to the abrupt change point or frequency peak point in the transition characteristics. The data collection ratio of the data window is adjusted according to the operational characteristics of the task type in each life cycle stage. Different sampling density strategies are set according to the sensitive state dimension of the task. The sampling density is increased in the active areas of abrupt features and decreased in the stable areas to complete the segmentation processing of the life cycle stage and output structured stage data.

[0013] Preferably, S5 is as follows: Based on the segmented life cycle stage data, the mean, standard deviation, and volatility of load data, current data, and vibration data in each life cycle stage are extracted as state features. The numerical differences between each state feature and historical benchmark data are calculated to construct a fusion evaluation index to characterize the operational stability and deviation of each life cycle stage. Based on the fusion evaluation index, the load offset, current offset and vibration offset in each life cycle stage are calculated. The offset scale is unified by normalization and the offset information of multiple state dimensions is integrated through a linear weighted model to output the comprehensive state offset as the basis for adjustment. Based on the comprehensive state offset, a task priority adjustment threshold, a maintenance cycle compression factor, and a data fusion interval scaling factor are set. When the offset is greater than the set threshold, the task priority is increased, the original maintenance cycle is compressed, and the length of the data fusion interval is expanded. The operation and maintenance parameters are dynamically adjusted to achieve the regulation and optimization of intelligent device lifecycle information fusion.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. By constructing a task hierarchy mapping table and a graph structure to express nested task relationships, and combining the jump characteristics of state dimensions such as load, current, and vibration in time-series data, the problem of confusion in lifecycle stage segmentation caused by nested task execution is solved. A graph convolutional neural network model is used to fuse task topology and state behavior data, achieving accurate identification of lifecycle stages of intelligent devices in complex task scheduling scenarios. Furthermore, based on the identified stage segmentation results, the original time-series data is structured, and the data window boundaries and sampling ratios are dynamically adjusted to make the stage data representation more accurate, significantly improving the temporal consistency and semantic integrity of lifecycle segmentation.

[0015] 2. Further, by introducing integrated evaluation indicators and state offset calculation mechanisms, and combining the operational characteristics of each lifecycle stage, dynamic adjustments are made to task priorities, maintenance cycles, and data fusion intervals, establishing a logical closed loop between nested structure identification, state change detection, and operational parameter optimization. Compared to traditional methods that rely solely on task start and end times, this technology not only possesses semantic awareness of nested task structures but also real-time response capabilities to minute state offsets within the lifecycle stage. This improves the accuracy of intelligent device operational status assessment, the rationality of operational resource allocation, and the intelligence of information fusion strategies, significantly enhancing the full lifecycle management level of devices in complex task execution environments. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart illustrating the intelligent device lifecycle information fusion and operation and maintenance optimization method of the present invention. Detailed Implementation

[0018] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0019] This invention provides, for example Figure 1 The intelligent device lifecycle information fusion and operation and maintenance optimization method shown includes the following steps: S1. By obtaining the time information and task type identifier of the task scheduling chain, and combining the timing relationship between the main task and the sub-task, determine whether the smart device is in a nested task execution state, and generate a task hierarchy mapping table to express the nested structure between tasks. In this embodiment, S1 specifically refers to: By reading the task execution start time, task execution end time and task type identifier recorded in the task scheduling chain, a data table of time information and task type identifier of the task scheduling chain is constructed, and the task number is used as an index field for unified management. During implementation, the start time, end time, and associated task type identifier for each task can be extracted by accessing the task scheduling log files of intelligent devices, MES system instruction records, or real-time communication data from the device control bus. Time information can be directly extracted from the timestamp field, while task type identifiers are typically present in the form of operation codes, process types, or task instruction codes. The extracted results are then organized into structured tabular data, using the task number as the primary key index field to create a data table for time information and task type identifiers. This supports subsequent time sequence relationship judgment and task nesting structure identification. Using the task number as the index enables task-level data retrieval and cross-task logical association, facilitating efficient analysis of time sequence overlap between master and slave tasks.

[0020] A task scheduling chain refers to a set of tasks recorded sequentially in chronological order during the operation of a smart device. The task execution start time and task execution end time represent the actual start and end times of each task's scheduling, typically recorded as high-precision timestamps. Task type identifiers are used to mark different types of tasks, such as main tasks, detection tasks, and warm-up tasks. The time information and task type identifier data table is a two-dimensional data set storing all task information in a structured format. Time information is used to construct the task time sequence, and task type identifiers are used to distinguish task categories. The task number is a unique identifier assigned by the device when generating a task; managing it as an index field ensures data uniqueness and traceability during subsequent searches, comparisons, and the construction of nested relationships.

[0021] Based on the time information of the task scheduling chain and the task type identifier data table, calculate the start and end time cross relationship between any two tasks, and determine whether the execution time of one task completely includes the execution time of another task. If it does, and the task type identifiers show a master-slave relationship, then the smart device is judged to be in a nested task execution state. In the task scheduling chain's time information and task type identifier data table, all task pairs are iterated sequentially. By comparing the start and end times of each task, it can be determined whether there is a time overlap between two tasks. When one task's start time is earlier than another task's start time, and its end time is later than another task's end time, it can be considered that it includes the latter in time. If there is a master-slave structure between the task type identifiers of the two tasks, such as one being a "main processing task" and the other a "temporary inspection task" or "rapid cleaning task," it can be further inferred that the device is currently in a nested task execution state. This logic is usually implemented programmatically, using nested loops to sequentially compare the time intervals of all tasks, and combining the task type field to filter the judgment results. For example, if task A starts at 08:00 and ends at 10:00, its task type is a main task; task B starts at 08:45 and ends at 09:15, its task type is a sub-task. Then, task B is completely surrounded by the time interval of task A, and the types show a master-slave relationship. In this case, it is determined that the device is in a nested task execution state.

[0022] The task scheduling chain's time information and task type identifier data table is a data structure used to record the entire lifecycle of task execution. It includes fields for the task's unique ID, task type, start time, and end time. Start-end time overlap refers to the partial or complete overlap of two tasks on the timeline, with complete inclusion meaning one task's time interval completely covers the other's. Execution time inclusion is determined by judging the order of start and end times. The master-slave relationship of task type identifiers is determined based on a pre-defined logical lookup table of task type classification rules. For example, the main task is typically a primary process operation such as processing or assembly, while sub-tasks are auxiliary process operations such as testing or debugging. Nested task execution state refers to the intelligent device inserting one or more sub-tasks while running the main task, resulting in logical overlap during task execution. By judging the time inclusion relationship and combining it with the master-slave relationship of task types, it is possible to accurately identify whether the device has entered a nested task state.

[0023] After determining that the task is in a nested execution state, a tree structure encoding sequence is constructed based on the temporal inclusion relationship between the main task and the sub-tasks and the task type identifier, and a task hierarchy mapping table is generated to express the nested structure between tasks in the task scheduling chain.

[0024] After determining the execution status of nested tasks, a tree-structured encoding sequence can be constructed based on the task's start time and task type identifier to represent the subordinate and containment relationships between tasks. Specifically, the time containment relationship is treated as a parent-child hierarchical relationship. For example, if the start and end times of a task completely cover the time intervals of multiple tasks, then that task is set as the parent node, and the contained tasks are its child nodes. Simultaneously, the master-slave logic is further validated based on the task type identifier to ensure that the parent node is a master task type and the child nodes are child task types. A recursive structure can be used during encoding, with each master task and its child tasks forming a first-level branch to construct the entire task nesting tree. After encoding, the number, type, time information, and hierarchical index of all task nodes are written into a task hierarchy mapping table. The purpose of generating this mapping table is to provide a structural basis for subsequent lifecycle phase division, state transition analysis, and multi-task fusion, ensuring that nested information is not lost during the fusion process.

[0025] After determining that a task is in a nested execution state, it is necessary to establish the structural relationships between tasks. The temporal inclusion relationship between the main task and subtasks means that the time range of the main task completely encompasses the time range of the subtask, defining the hierarchical connection between parent and child nodes in the tree structure. The task type identifier is a core field for identifying task roles, used to confirm whether a task possesses the attributes of a main task or belongs to the category of subtasks. The tree structure encoding sequence is a method of recording task hierarchy through node numbers and parent node indices, making the nested relationship computable and searchable. The task hierarchy mapping table is a data structure used to record the position of each task in the nested tree, containing fields such as task number, task type, hierarchy depth, and parent node number. This table, by structurally expressing the nesting logic between tasks, is a prerequisite for subsequent semantic fusion of lifecycle information.

[0026] S2. Based on the determined nested task execution state, extract state transition features from the load, current, vibration and other state data collected during the device operation, and use the time window comparison algorithm to calculate the state change amplitude, duration and frequency to obtain a set of state transition features associated with task switching. In this embodiment, S2 specifically includes the following steps: S201. Based on the determined nested task execution status, continuously collect load data, current data and vibration data in chronological order during equipment operation, and construct the status data sequence corresponding to the task according to the task number. Assuming the device is in a nested task execution state, the task number and its corresponding start and end times must be extracted one by one based on the task scheduling schedule. Subsequently, load data, current data, and vibration data corresponding to the task execution time periods are continuously collected from the device's operation monitoring channel. The data collection process employs a high-precision timestamp alignment mechanism synchronized with the device control system to ensure accurate time-series correspondence between different data types. After collection, different types of status data within the same task number are organized into time series, forming a dataset indexed by the task number. The status data sequence for each task is saved independently, facilitating refined analysis and feature extraction by task dimension during subsequent lifecycle phase segmentation. This approach not only improves data processing efficiency but also provides a fundamental structural support for subsequent transition feature recognition.

[0027] "The determined nested task execution status" refers to the nested structure of master and sub-tasks identified through the task start and end times and type relationships; "Continuously collecting load data, current data, and vibration data in chronological order during equipment operation" means that equipment operating status indicators are collected uninterruptedly throughout the entire task execution process, maintaining temporal consistency; "Constructing a status data sequence corresponding to the task based on the task number" means grouping all status data according to the task number and organizing them into a serialized structure in chronological order, ensuring that each type of status indicator has a complete, continuous, and quantifiable statistical basis within the task time. This process is a necessary prerequisite for subsequent calculations of state transition amplitude, duration, and frequency, and also ensures the accuracy and contextual consistency of lifecycle information fusion.

[0028] S202. Set a sliding time window in each state data sequence, calculate the state change amplitude by the difference between the maximum and minimum values ​​in the window, record the time period when the signal is continuously higher than the average value to calculate the state change duration, and count the number of drastic state changes per unit time to calculate the state change frequency. S203. The calculated state change amplitude, state change duration and state change frequency are used as state transition features. An aligned time window is constructed with the task switching time point as the center. All state transition features within the aligned time window are selected to form a set of state transition features associated with task switching, which is used to express the state change behavior during task switching.

[0029] During task execution, to extract state response features closely related to task switching, the calculated amplitude, duration, and frequency of state changes need to be used as state transition features. A fixed-width aligned time window is set around each task switching point, for example, extending forward and backward by a certain period centered on the task switching point to form a symmetrical time segment. Within this time window, all state transition features falling within this time range are filtered and grouped into the state transition feature set of the current task switching event. This focused filtering operation effectively isolates local state responses triggered during the switching between the main task and sub-tasks, thereby enhancing the correspondence between features and structural changes and providing accurate support for subsequent identification of lifecycle stage division points.

[0030] The magnitude of state changes is a crucial indicator for measuring the intensity of equipment operational fluctuations, the duration of state changes reflects the stability of these fluctuations, and the frequency of state changes captures the dynamic characteristics of these fluctuations. By unifying these three types of indicators within a time window constructed with task switching as a reference, the instantaneous response features of the equipment during nested task alternations can be accurately extracted. The alignment of the time window ensures consistency of feature extraction along the timeline, while the filtering operation ensures a direct correlation between features and task events, forming a set of transition features oriented towards nested switching semantics. This set possesses both timeliness and task dependence, providing sufficient input for accurately dividing lifecycle stages.

[0031] In this embodiment, S202 specifically refers to: In each state data sequence, a fixed-width sliding time window is set, with a sliding step size smaller than the window width, to ensure that the window continuously covers all data points in the state data sequence, and a timestamp index is created for each window for subsequent comparison. To continuously monitor trends in state data sequences, a fixed-width sliding time window can be defined within each segment of the sequence, representing the data coverage for each analysis. The sliding process uses a step size smaller than the window width, such as half the window width, to ensure continuous coverage and overlapping monitoring of data points. During window division, the start and end times within each window are numbered, and a corresponding timestamp index table is created. This time index serves as the positioning benchmark for comparing subsequent state features with task switching points. This sliding structure not only ensures that no possible state abrupt changes are missed but also enhances the sensitivity to short-term sudden fluctuations through window overlap. The fixed width design makes state sequences comparable across different tasks, while the step size ensures continuity and responsiveness in change capture. This sliding structure exhibits good time-series preservation capabilities in most dynamic monitoring algorithms, providing stable data support for subsequent extraction of change amplitude, duration, and frequency.

[0032] Within each sliding time window, the maximum and minimum values ​​of the state data are obtained, and the state change amplitude is calculated by the difference between the two. At the same time, the arithmetic mean of the state data within the window is calculated, and continuous segments higher than the average value are marked. Within each sliding time window, all state data points covered by the window are first traversed, and the maximum and minimum values ​​are extracted. The difference between these two values ​​is used to calculate the amplitude of the state change, which measures the intensity of fluctuations in the equipment's operating state during that time period. Then, the values ​​of all state data within the window are summed and divided by the number of data points to calculate the arithmetic mean of the window. Each data point is then checked sequentially from beginning to end, and segments continuously exceeding this average value are marked as time periods where significant state responses may exist. For example, in a current monitoring window, if the maximum current is 12 amps and the minimum current is 4 amps, the amplitude of the change is 8 amps. Simultaneously, if the average value is 7 amps, and three consecutive sampling points are 8 amps, 9 amps, and 8.5 amps, these three points constitute a continuous segment exceeding the average value. This processing method can accurately capture the amplitude and high-response persistence of state changes, reflecting whether the equipment experiences drastic fluctuations or stable load anomalies during that time period, thus providing a valid data foundation for the subsequent accurate extraction of duration and frequency. This computational method is highly interpretable, easily reproducible, and has good engineering feasibility, making it suitable for continuous feature analysis of most sensor-type state signals.

[0033] The length of consecutive segments above the average value in each sliding time window is counted to represent the duration of state change. Cases where the difference between two adjacent data points is greater than a preset threshold are counted to calculate the frequency of state change, thereby constructing a complete set of jump characteristic indicators for each segment of state data.

[0034] Within each sliding time-series window, the length of the marked, continuously above-average state data segments is first calculated. The length is measured by the number of sampling points or the corresponding time span, representing the duration of the state change and reflecting the persistence of equipment malfunctions or load shifts. Subsequently, the numerical difference between any two adjacent data points within the window is calculated one by one, and a specific jump detection threshold is set, such as a 5-amp current fluctuation or a 2-millimeter vibration change. When the difference between adjacent data points exceeds this threshold, it is considered a severe jump event and counted. This preset threshold is reasonably selected based on the equipment's operating characteristics and sensing accuracy; different threshold standards can be used for different equipment types or task stages. For example, in a processing task, if the vibration signal is continuously above the average for 5 seconds, and there are 3 jumps with amplitudes exceeding the set threshold, the duration is recorded as 5 seconds and the frequency as 3 times. Finally, the duration and jump frequency in each window are combined with the previously extracted change amplitudes to construct a complete set of jump feature indicators. This set provides multi-dimensional state basis for subsequent analysis of accurate identification of lifecycle stage changes and can effectively distinguish state response differences under nested sub-tasks.

[0035] S3. Based on the task hierarchy mapping table and the state transition feature set, construct a graph convolutional neural network model. With the nested relationship between tasks as the topology, input each task node and its corresponding state transition feature into the model to identify each stage of the smart device life cycle. In this embodiment, S3 specifically includes the following steps: S301. Construct a graph structure for a graph convolutional neural network model based on a task hierarchy mapping table. Treat each task in the task hierarchy mapping table as a node in the graph, establish directed connections between nodes according to the nesting relationship between the main task and sub-tasks, and construct a graph data input form with the nesting relationship between tasks as the topology. When constructing the graph structure of the graph convolutional neural network model, the process begins by reading all task numbers listed in the task hierarchy mapping table and mapping each task number to an independent node in the graph. Then, based on the nesting relationships recorded in the task hierarchy mapping table, the hierarchical relationship chains between the main task and subtasks are extracted, with the main task as the parent node and the subtasks as child nodes, establishing unidirectional connections between nodes. If a main task contains multiple subtasks, a directed edge is created between each subtask and the main task. This construction method forms a directed graph with a task hierarchy as its logical structure, explicitly expressing the nesting and subordinate relationships of each task in graph topology. Finally, all nodes and their connecting edges in the graph are organized into the input structure of the graph convolutional neural network for subsequent graph feature propagation and convolution computation.

[0036] The task hierarchy mapping table represents the nested structure between tasks, task numbers represent node elements in the graph, and directed connections represent topological edges between nodes. In the nested relationship, the "main task" is the parent node that contains other tasks in time, and the "child task" is the subordinate node nested within its time period. Directed connections between nodes point from the main task to the child task to maintain the dependency order of the nested structure. The topological structure is the graphical representation of all nodes and edges in the graph, reflecting the hierarchical nature of task organization. The graph data input format involves inputting the task node set and edge set into a graph convolutional neural network framework to support the convolutional propagation computation of graph structural features, providing structural prior information for subsequent lifecycle identification.

[0037] S302. The graph structure constructed by the task hierarchy mapping table is mapped one-to-one with the state transition features of each task node in the state transition feature set. Each task node and its corresponding state transition features are used as feature vectors and input into the graph convolutional neural network model. Multiple rounds of graph convolution operations are performed to fuse the structural and state features between adjacent tasks. When matching the graph structure with the set of state transition features, the graph structure constructed in the task hierarchy mapping table is first read, and a state transition feature item matching its task number is searched in each task node. Using the task number as a key index, each task node and its corresponding state transition feature triple are combined into a feature vector input. The feature vector includes the magnitude, duration, and frequency of the state change. Subsequently, the feature vectors of all task nodes are synchronously input into the graph convolutional neural network model. The model uses the task topology as the propagation path and achieves information propagation between adjacent nodes through multiple rounds of graph convolution operations, enabling the structural and state information to be fused layer by layer in the graph, thereby enhancing each node's perception of the states of upstream and downstream tasks. This approach effectively improves the global contextual understanding during lifecycle stage identification.

[0038] Task nodes represent the basic units in the graph structure, with each node representing a specific task. In the state transition feature set, each triple represents the task's behavioral pattern during execution, including the magnitude of the state change (indicating intensity), the duration of the state change (indicating temporal features), and the frequency of the state change (indicating volatility). A one-to-one correspondence means that the task number serves as a unique primary key, binding the state behavioral features to the graph node. The graph convolutional neural network model is used to perform feature propagation within the topological structure. Through graph convolution operations, nodes consider the state features of their neighboring nodes during computation, achieving a coupled representation of local structure and global state. Multi-round graph convolution operations represent a gradual expansion of the receptive field in multi-layer propagation, allowing features to be influenced not only by directly adjacent nodes but also by integrating a wider range of upstream and downstream structural information, thus providing a highly fused input feature foundation for classification in subsequent lifecycle stages.

[0039] S303. In the output stage of the graph convolutional neural network model, supervised classification training of life cycle stage labels is performed on the high-dimensional feature representations generated by each task node. Based on the feature vector jointly represented by the nested structure and state transition features, the various stages of the smart device life cycle are identified.

[0040] The reason for using high-dimensional feature representations generated for each task node for supervised classification training of lifecycle stage labels in the output stage of the graph convolutional neural network model is to enable the model to accurately learn the lifecycle stage affiliation of task nodes under different nesting structures and state transition feature combinations. Because nested task structures lead to complex interleaving of temporal and state behaviors between tasks, traditional time-series or state analysis methods alone are insufficient for effectively dividing lifecycle stages. By constructing joint feature vectors and performing classification training under the supervision of labeled data, the model can establish a mapping relationship between feature distribution and lifecycle stages. This not only identifies obvious task stage transitions but also handles the ambiguity of stage boundaries caused by task nesting, thereby improving the accuracy of lifecycle identification and adaptability to complex task scheduling scenarios, providing a stable and reliable basis for subsequent stage division and operational decisions.

[0041] In this embodiment, S303 specifically refers to: In the output stage of the graph convolutional neural network model, the high-dimensional feature representations obtained after convolution calculation of each task node are summarized into a feature matrix, and a life cycle stage label is assigned to each task node to construct a dataset corresponding to the task nodes and labels for supervised classification training. In the output stage of the graph convolutional neural network model, the high-dimensional feature representations obtained after graph convolution operations for each task node are first extracted. These representations are then sorted by node number and aggregated into a feature matrix. Each row of this matrix corresponds to a task node and contains information about its nested structural relationships and the fusion of state transition features. Subsequently, based on manually labeled or historical data, a lifecycle stage label is assigned to each task node. These labels can represent state categories such as startup, steady-state, overload, or shutdown during device operation. By constructing sample pairs between the feature vectors of each node and their lifecycle stage labels, a standard dataset is formed. This dataset is used for supervised classification training to optimize the parameters in the graph convolutional neural network model, enabling the model to identify lifecycle stages on unknown data. The high-dimensional feature representation carries the joint expression of structural information and state features, the feature matrix provides an overall aligned input format, and the lifecycle stage labels provide a supervisory signal for the model output. The supervised classification training process optimizes the model's discrimination boundary, thereby improving the accuracy of lifecycle segmentation.

[0042] During supervised classification training, the cross-entropy loss function is used as the optimization objective. The parameter weights of the graph convolutional layer and the classification layer are continuously adjusted through the backpropagation algorithm to minimize the difference between the lifecycle stage labels and the model output labels. In supervised classification training, to optimize the prediction accuracy of the graph convolutional neural network model for lifecycle stage labels, the cross-entropy loss function is selected as an indicator to measure the difference between the model's predicted output and the true labels. The cross-entropy loss function measures the deviation between probability distributions and is suitable for multi-class classification tasks; therefore, it can be used to evaluate the model's prediction confidence for each stage label in lifecycle stage segmentation. In each training round, the predicted results are compared with the actual labels, and the loss value between the current model output and the target output is calculated, using this loss as the optimization objective. Subsequently, the backpropagation algorithm is used to calculate the gradient for each parameter in the graph convolutional and classification layers based on the loss value, updating the parameter weights in the gradient direction, allowing the model to gradually converge towards the optimal solution. The backpropagation algorithm automatically propagates the error back layer by layer in the neural network structure using a chain rule, thus efficiently completing the model training process. The cross-entropy loss function provides accurate error signals, while the backpropagation algorithm optimizes the propagation of this error throughout the entire model, enabling the model to gradually acquire the ability to accurately identify the lifecycle stages of task nodes.

[0043] After training, the feature vector jointly represented by the nested structure and state transition features is input into the classification layer. Based on the probability distribution results output by the classification layer, the various stages of the smart device's life cycle are identified, and the life cycle stage attributes of the task nodes are updated through the stage identification results.

[0044] After supervised classification training is completed, the feature vector, composed of the nested structure and state transition features corresponding to each task node, is input into the trained classification layer. The classification layer outputs the probability distribution of each lifecycle stage based on the parameter weights learned by the model. By selecting the stage with the highest probability from this probability distribution, the specific lifecycle stage label of each task node can be determined. In this process, the classification layer, as the final decision-making unit of the model, can comprehensively consider the task hierarchy dependencies expressed by the nested structure and the operational dynamics reflected by the state transition features to achieve accurate classification of lifecycle stages. Subsequently, lifecycle stage labels are assigned to the corresponding task nodes based on the classification results, and the lifecycle stage attributes of the task nodes are updated to form a complete lifecycle structure annotation map, providing a high-precision stage identification basis for subsequent data window partitioning and fusion evaluation. This process ensures that the model's classification decisions are highly consistent with the actual task nesting semantics and device operating status, improving the accuracy and reliability of the entire information fusion process.

[0045] S4. Based on the identified lifecycle stages, construct multiple data windows in the original time series data. Adjust the boundary position and acquisition ratio of the data windows according to the task type and its jump characteristics in each stage to complete the lifecycle stage segmentation process. In this embodiment, S4 specifically refers to: Based on the identified life cycle stages, time indices corresponding to the start and end points of each life cycle stage are extracted from the original time series data. Multiple data windows are constructed in the original time series data according to the time indices, with each data window corresponding to a life cycle stage, and the continuity and order are maintained on the time axis. When constructing multiple data windows, the process begins by searching for the corresponding timestamp index positions in the original time-series data, using the start and end times of each lifecycle stage output by the intelligent device lifecycle identification model. These time indices are used to accurately determine the position range of each stage on the timeline. Based on these indices, the original state data stream is segmented, generating data windows corresponding one-to-one with each lifecycle stage. Each data window consists of continuous data from the start index to the end index, arranged according to the natural order of the lifecycle stages on the timeline. This processing method ensures data integrity and the accuracy of stage boundaries, while laying the foundation for time alignment for subsequent boundary adjustments and feature analysis.

[0046] The start and end points represent the time boundaries of each lifecycle stage and are key anchor points for lifecycle segmentation. The time index serves as a location reference in the raw time-series data, used to quickly locate the target segment from continuous state data. Constructing multiple data windows means logically dividing the raw data according to lifecycle stages, with each window corresponding to a unique lifecycle stage. Maintaining continuity on the timeline avoids overlapping or gaps in data segments, ensuring temporal consistency in subsequent analysis; while sequentiality requires each window to be naturally arranged according to stages on the timeline, ensuring that lifecycle analysis conforms to the actual temporal logic of device operation. This processing flow provides crucial support for segmentation accuracy and analysis efficiency during the data preprocessing stage.

[0047] Based on the task type and its transition characteristics in each life cycle stage, the boundary position of each data window is refined and adjusted. Specifically, the original boundary is moved forward or backward to the time point when the transition behavior occurs, according to the abrupt change point or frequency peak point in the transition characteristics, so that the data window can more accurately cover the state change range of the life cycle stage. To improve the accuracy of lifecycle phase segmentation, based on the initially constructed data window, the window boundary positions need to be refined and adjusted by considering the task type and transition characteristics within each lifecycle phase. Specifically, this involves traversing the set of transition characteristics corresponding to the lifecycle phase, focusing on identifying abrupt changes where the magnitude of change jumps dramatically or when the frequency of state changes reaches its peak. These points typically correspond to critical moments when the device transitions from one stable state to another, thus representing the actual lifecycle phase transition nodes. Based on the location of these key points, the boundary indexes originally set based on task time are shifted forward or backward to ensure that the adjusted data window accurately covers the actual state change range of the lifecycle phase, avoiding data from other phases within the same phase, thereby improving the accuracy of subsequent fusion analysis.

[0048] Task types within a lifecycle phase are used to predict behavior along state evolution paths, with different task types corresponding to different state transition patterns. Transition characteristics include the magnitude and frequency of state changes, serving as the primary basis for identifying task switching nodes. Mutation points are locations where state data undergoes drastic changes within a short period, while frequency peaks are locations where state changes are most frequent per unit time; these two types of characteristics typically represent boundary points where substantial state changes occur. Moving the data window boundary forward or backward is a dynamic positioning method aimed at breaking the original static window division mechanism based on task time, ensuring that window division is synchronized with the actual state changes of the device. This boundary adjustment strategy improves the purity and temporal alignment of lifecycle phase data, forming the foundation for achieving high-precision information fusion.

[0049] The data collection ratio of the data window is adjusted according to the operational characteristics of the task type in each life cycle stage. Different sampling density strategies are set according to the sensitive state dimension of the task. The sampling density is increased in the active areas of abrupt features and decreased in the stable areas to complete the segmentation processing of the life cycle stage and output structured stage data.

[0050] To ensure more representative data acquisition across lifecycle stages, the data window's acquisition ratio needs to be adjusted differentiated based on the operational characteristics of the task types within each stage. First, the sensitive state dimensions of the corresponding task in each lifecycle stage are analyzed; for example, the importance of load fluctuations to cutting tasks or the responsiveness of current changes to assembly tasks. After identifying regions with active state transition characteristics, the sampling frequency within these regions is increased to meticulously capture rapidly changing dynamic processes. Conversely, for stable regions with slow state changes, the sampling density is appropriately reduced to minimize redundant data. This adaptive sampling density strategy not only preserves the complete information of key feature points but also effectively reduces the overall data volume, achieving fine segmentation of lifecycle stages while outputting structured stage data for subsequent fusion and optimization.

[0051] The operational characteristics of a task type determine the device's responsiveness to different state parameters during execution, while sensitive state dimensions refer to state indicators highly correlated with task performance. Areas with frequent transition features refer to segments where state data changes frequently within a short period; these areas often conceal crucial lifecycle stage transition information. The sampling density strategy improves data analysis efficiency and quality by increasing temporal resolution in critical areas and decreasing sampling frequency in non-critical areas. This regional and task-specific sampling mechanism breaks the homogeneity limitations of traditional equal-interval sampling, making stage segmentation more accurate and consistent with the actual operating characteristics of the device.

[0052] S5. Based on the segmented lifecycle stage data, construct fusion evaluation indicators, calculate the state offset of each stage, and dynamically adjust the task priority, maintenance cycle and data fusion interval based on the offset to realize the operation and maintenance control optimization of intelligent device lifecycle information fusion.

[0053] In this embodiment, S5 specifically refers to: Based on the segmented life cycle stage data, the mean, standard deviation, and volatility of load data, current data, and vibration data in each life cycle stage are extracted as state features. The numerical differences between each state feature and historical benchmark data are calculated to construct a fusion evaluation index to characterize the operational stability and deviation of each life cycle stage. To construct a fusion evaluation index for characterizing the operational stability and deviation of intelligent devices across their lifecycle stages, a multi-dimensional statistical analysis of the segmented data for each lifecycle stage is first required. This can be achieved by traversing the load, current, and vibration data collected in each stage, calculating the mean, standard deviation, and volatility of these three types of state data within the current stage. The mean reflects the normal operating level of the device, the standard deviation reveals the amplitude of state fluctuations, and the volatility characterizes the frequency of state changes. Subsequently, the statistical results are compared with the stable benchmark values ​​of the corresponding tasks in historical data. The greater the difference, the more the device's operating state deviates from normal conditions, thus reflecting potential operational anomalies or task interference in that stage. The fusion evaluation index constructed in this way can be used to quantitatively characterize the degree of operational deviation at each stage of the lifecycle, providing a reliable basis for subsequent adjustments to maintenance parameters. For example, when the vibration volatility is significantly higher than the historical value in a certain stage, and the current standard deviation also increases simultaneously, it indicates that the device may be under abnormal load conditions, and this stage requires focused attention and optimization.

[0054] Extracting the mean values ​​of load, current, and vibration data is to obtain the basic operating level of the equipment at a specific stage of its life cycle. The standard deviation measures the dispersion of data within that stage, reflecting the stability of the operating state. Volatility quantifies the activity of the data by the frequency of change per unit time. These three statistical indicators depict the equipment state from different dimensions. Historical benchmark data can be derived from a dataset accumulated under stable operation of the equipment performing similar tasks. Difference calculations typically use absolute difference, relative difference, or Z-scores to measure the degree of state deviation. The fusion evaluation indicator combines the difference results from multiple state dimensions as vectors or scalars to form a quantifiable evaluation quantity, used to uniformly assess the level of operational deviation at different life cycle stages. The overall design logic is to use a data-driven approach to assist in judging the performance stability of the equipment during task execution.

[0055] Based on the fusion evaluation index, the load offset, current offset and vibration offset in each life cycle stage are calculated. The offset scale is unified by normalization and the offset information of multiple state dimensions is integrated through a linear weighted model to output the comprehensive state offset as the basis for adjustment. To enhance the quantification of deviations in equipment operating status, load offset, current offset, and vibration offset can be calculated for each lifecycle stage based on integrated evaluation metrics. The calculation typically involves the numerical difference between the current stage's state statistics (such as mean or standard deviation) and historical reference values. Since the units and fluctuation ranges of physical quantities differ across state dimensions, each offset must first be normalized to ensure comparability within the same order of magnitude. After normalization, a linear weighted model is introduced, setting weight coefficients based on the impact of different state dimensions on operational stability. For example, vibration offset may have a greater weight than current offset in high-precision equipment. The normalized offsets are multiplied by their corresponding weights and summed to obtain the comprehensive state offset for each lifecycle stage, serving as the core basis for judging the degree of deviation in the current stage's operating status. This approach takes into account information from multiple state data sources, ensuring that the offset reflects the overall performance of the equipment. For example, if the vibration offset is normalized to 0.6 and the current offset is 0.3 at a certain stage, and the vibration and current weights are assigned to 0.7 and 0.3 respectively, the final output comprehensive state offset will be 0.51, which can be used to drive subsequent control logic.

[0056] Load offset, current offset, and vibration offset represent the differences of three types of state variables relative to historical benchmarks during the current lifecycle stage of the equipment. Normalization often uses min-max scaling or Z-score standardization methods to compress data from different units and ranges to a comparable, uniform scale. A linear weighted model is a structure that combines multiple variables into a single result value according to their weights; these weights can be set through expert experience, training with historical fault data, or based on system stability impact metrics. The comprehensive state offset is a scalar result obtained by summing the products of the normalized offset dimensions and their weights. It is used to uniformly measure the overall offset during the lifecycle stage, providing a quantitative reference for task priority adjustments and maintenance strategy updates. The overall strategy achieves the fusion measurement of multi-dimensional state data, featuring high accuracy, fast response, and ease of implementation.

[0057] Based on the comprehensive state offset, a task priority adjustment threshold, a maintenance cycle compression factor, and a data fusion interval scaling factor are set. When the offset is greater than the set threshold, the task priority is increased, the original maintenance cycle is compressed, and the length of the data fusion interval is expanded. The operation and maintenance parameters are dynamically adjusted to achieve the regulation and optimization of intelligent device lifecycle information fusion.

[0058] After obtaining the comprehensive status offset, to achieve dynamic control of equipment operation status, it is necessary to set task priority adjustment thresholds, maintenance cycle compression factors, and data fusion interval scaling factors based on this offset. In practice, multiple offset threshold intervals can be predefined based on equipment type and historical operating experience, each interval corresponding to a different control strategy. When the comprehensive status offset exceeds a certain threshold, the system determines that there is a high operational risk in that lifecycle stage. At this time, the task priority is adjusted upwards, allowing tasks in high-risk stages to receive more timely resource allocation; simultaneously, based on the product of the offset and the compression factor, the maintenance cycle is shortened proportionally, enabling potential problems to be identified in advance; furthermore, to improve the accuracy of subsequent information fusion, the time range of the data fusion interval is extended according to the scaling factor to capture richer status features. Through this offset-driven dynamic adjustment mechanism, coordinated optimization of task scheduling, maintenance, and data fusion can be achieved. For example, when the offset is 0.75 and the threshold is 0.6, the system automatically raises the task priority by one level, shortens the maintenance cycle by 20%, and extends the data fusion window to 1.3 times the original length in order to more comprehensively assess the health status of the equipment.

[0059] The comprehensive state offset is a unified quantitative indicator of the degree of offset of multi-dimensional state characteristics of equipment during its life cycle, and it is the core input of the control logic. The task priority adjustment threshold is used to determine at what degree of offset the task scheduling order should be changed; its value is usually designed based on equipment importance and fault tolerance level. The maintenance cycle compression factor is used to adjust the planned maintenance interval according to the offset ratio, so that maintenance behavior dynamically matches the actual health status of the equipment. The data fusion interval scaling factor controls the length of the data period covered by the fusion analysis; the higher the offset, the larger the scaling factor, so as to collect a wider time range and improve data relevance. By linking these three parameters with the comprehensive state offset, multi-dimensional operation and maintenance strategies can be dynamically and adaptively adjusted, enabling the life cycle information fusion and operation and maintenance optimization process of intelligent equipment to have self-learning and responsive capabilities, ultimately improving the security and stability of system operation.

[0060] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0061] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0062] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0063] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0064] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0065] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0066] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for optimizing operation and maintenance by integrating lifecycle information of intelligent devices, characterized in that, Specifically, the following steps are included: S1. By obtaining the time information and task type identifier of the task scheduling chain, and combining the timing relationship between the main task and the sub-task, determine whether the smart device is in a nested task execution state, and generate a task hierarchy mapping table to express the nested structure between tasks. S2. Based on the determined nested task execution state, extract state transition features from the load, current, vibration and other state data collected during the device operation, and use the time window comparison algorithm to calculate the state change amplitude, duration and frequency to obtain a set of state transition features associated with task switching. S3. Based on the task hierarchy mapping table and the state transition feature set, construct a graph convolutional neural network model. With the nested relationship between tasks as the topology, input each task node and its corresponding state transition feature into the model to identify each stage of the smart device life cycle. S4. Based on the identified lifecycle stages, construct multiple data windows in the original time series data. Adjust the boundary position and acquisition ratio of the data windows according to the task type and its jump characteristics in each stage to complete the lifecycle stage segmentation process. S5. Based on the segmented lifecycle stage data, construct fusion evaluation indicators, calculate the state offset of each stage, and dynamically adjust the task priority, maintenance cycle and data fusion interval based on the offset to realize the operation and maintenance control optimization of intelligent device lifecycle information fusion.

2. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 1, characterized in that, S1 specifically refers to: By reading the task execution start time, task execution end time and task type identifier recorded in the task scheduling chain, a data table of time information and task type identifier of the task scheduling chain is constructed, and the task number is used as an index field for unified management. Based on the time information of the task scheduling chain and the task type identifier data table, calculate the start and end time cross relationship between any two tasks, and determine whether the execution time of one task completely includes the execution time of another task. If it does, and the task type identifiers show a master-slave relationship, then the smart device is judged to be in a nested task execution state. After determining that the task is in a nested execution state, a tree structure encoding sequence is constructed based on the temporal inclusion relationship between the main task and the sub-tasks and the task type identifier, and a task hierarchy mapping table is generated to express the nested structure between tasks in the task scheduling chain.

3. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 1, characterized in that, S2 specifically includes the following steps: S201. Based on the determined nested task execution status, continuously collect load data, current data and vibration data in chronological order during equipment operation, and construct the status data sequence corresponding to the task according to the task number. S202. Set a sliding time window in each state data sequence, calculate the state change amplitude by the difference between the maximum and minimum values ​​in the window, record the time period when the signal is continuously higher than the average value to calculate the state change duration, and count the number of drastic state changes per unit time to calculate the state change frequency. S203. The calculated state change amplitude, state change duration and state change frequency are used as state transition features. An aligned time window is constructed with the task switching time point as the center. All state transition features within the aligned time window are selected to form a set of state transition features associated with task switching, which is used to express the state change behavior during task switching.

4. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 3, characterized in that, S202 specifically refers to: In each state data sequence, a fixed-width sliding time window is set, with a sliding step size smaller than the window width, to ensure that the window continuously covers all data points in the state data sequence, and a timestamp index is created for each window for subsequent comparison. Within each sliding time window, the maximum and minimum values ​​of the state data are obtained, and the state change amplitude is calculated by the difference between the two. At the same time, the arithmetic mean of the state data within the window is calculated, and continuous segments higher than the average value are marked. The length of consecutive segments above the average value in each sliding time window is counted to represent the duration of state change. Cases where the difference between two adjacent data points is greater than a preset threshold are counted to calculate the frequency of state change, thereby constructing a complete set of jump characteristic indicators for each segment of state data.

5. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 1, characterized in that, S3 specifically includes the following steps: S301. Construct a graph structure for a graph convolutional neural network model based on a task hierarchy mapping table. Treat each task in the task hierarchy mapping table as a node in the graph, establish directed connections between nodes according to the nesting relationship between the main task and sub-tasks, and construct a graph data input form with the nesting relationship between tasks as the topology. S302. The graph structure constructed by the task hierarchy mapping table is mapped one-to-one with the state transition features of each task node in the state transition feature set. Each task node and its corresponding state transition features are used as feature vectors and input into the graph convolutional neural network model. Multiple rounds of graph convolution operations are performed to fuse the structural and state features between adjacent tasks. S303. In the output stage of the graph convolutional neural network model, supervised classification training of life cycle stage labels is performed on the high-dimensional feature representations generated by each task node. Based on the feature vector jointly represented by the nested structure and state transition features, the various stages of the smart device life cycle are identified.

6. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 5, characterized in that, S303 specifically refers to: In the output stage of the graph convolutional neural network model, the high-dimensional feature representations obtained after convolution calculation of each task node are summarized into a feature matrix, and a life cycle stage label is assigned to each task node to construct a dataset corresponding to the task nodes and labels for supervised classification training. During supervised classification training, the cross-entropy loss function is used as the optimization objective. The parameter weights of the graph convolutional layer and the classification layer are continuously adjusted through the backpropagation algorithm to minimize the difference between the lifecycle stage labels and the model output labels. After training, the feature vector jointly represented by the nested structure and state transition features is input into the classification layer. Based on the probability distribution results output by the classification layer, the various stages of the smart device's life cycle are identified, and the life cycle stage attributes of the task nodes are updated through the stage identification results.

7. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 1, characterized in that, S4 specifically refers to: Based on the identified life cycle stages, time indices corresponding to the start and end points of each life cycle stage are extracted from the original time series data. Multiple data windows are constructed in the original time series data according to the time indices, with each data window corresponding to a life cycle stage, and the continuity and order are maintained on the time axis. Based on the task type and its transition characteristics in each life cycle stage, the boundary position of each data window is refined and adjusted. Specifically, the original boundary is moved forward or backward to the time point when the transition behavior occurs, according to the abrupt change point or frequency peak point in the transition characteristics. The data collection ratio of the data window is adjusted according to the operational characteristics of the task type in each life cycle stage. Different sampling density strategies are set according to the sensitive state dimension of the task. The sampling density is increased in the active areas of abrupt features and decreased in the stable areas to complete the segmentation processing of the life cycle stage and output structured stage data.

8. The intelligent device lifecycle information fusion and operation and maintenance optimization method according to claim 1, characterized in that, S5 specifically refers to: Based on the segmented life cycle stage data, the mean, standard deviation, and volatility of load data, current data, and vibration data in each life cycle stage are extracted as state features. The numerical differences between each state feature and historical benchmark data are calculated to construct a fusion evaluation index to characterize the operational stability and deviation of each life cycle stage. Based on the fusion evaluation index, the load offset, current offset and vibration offset in each life cycle stage are calculated. The offset scale is unified by normalization and the offset information of multiple state dimensions is integrated through a linear weighted model to output the comprehensive state offset as the basis for adjustment. Based on the comprehensive state offset, a task priority adjustment threshold, a maintenance cycle compression factor, and a data fusion interval scaling factor are set. When the offset is greater than the set threshold, the task priority is increased, the original maintenance cycle is compressed, and the length of the data fusion interval is expanded. The operation and maintenance parameters are dynamically adjusted to achieve the regulation and optimization of intelligent device lifecycle information fusion.

Images