A Method and System for Industrial Equipment Anomaly Identification Based on an Industrial Cloud Platform

By defining the state transition range and decomposing seasonal trends through control commands, the problem of inaccurate multidimensional anomaly identification caused by misalignment of state transition timing and local disturbances in industrial equipment was solved, achieving more accurate anomaly identification and stage positioning.

CN123116097APending Publication Date: 2026-11-17SHANDONG TIANJIA NEW ENERGY TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611463474.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-20
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies suffer from inaccurate multidimensional anomaly identification and stage positioning due to misalignment of state transition timing and local disturbances in industrial equipment.

Method used

The state transition range is defined by control commands. Local distribution merging and minimum description length filtering are used to determine the equipment response location and the location of steady state formation, generating a state transition phase sequence. Seasonal trend decomposition is used to distinguish between isolated residuals and continuous residuals, sampling interference is corrected, and the direction of adjacent phase change is encoded to form a multidimensional feature sequence of state transitions. The multidimensional feature sequence of state transitions is phase-paired with the same type of normal historical feature sequence, the dimensional distance of the running parameters is pre-sorted, and the nearest non-overlapping subsequence is retrieved dimension by dimension to form a multidimensional matrix profile of state transitions. The continuity of residuals and the difference in slope direction are combined for verification to determine the abnormal stage.

Benefits of technology

It improves the temporal consistency and stage positioning accuracy of industrial equipment anomaly identification results, reduces comparison misalignment caused by different response times, reduces the masking of local anomalies by normal parameters, and ensures that the anomaly identification results reflect the actual state transition stage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116097A_ABST
    Figure CN123116097A_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for identifying industrial equipment anomalies based on an industrial cloud platform, belonging to the field of anomaly identification technology. The method defines the state transition range according to control commands, determines the equipment response location and stable state formation location through local distribution merging and minimum description length filtering, unifies similar state transition phases, and generates a state transition phase sequence. Seasonal trend decomposition is used to distinguish isolated residuals from continuous residuals, corrects sampling interference, and encodes the direction of adjacent phase changes, obtaining a multi-dimensional feature sequence of state transitions. The multi-dimensional feature sequence of state transitions is then paired with similar normal historical feature sequences, the dimensional distance of operating parameters is pre-sorted, and the nearest non-overlapping subsequence is retrieved dimension by dimension, forming a multi-dimensional matrix profile of state transitions. This invention utilizes residual continuity, slope direction differences, and the nearest normal historical state transitions for consistency verification, enhancing the interpretability and traceability of industrial equipment anomaly identification results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of anomaly identification technology, and in particular to a method and system for anomaly identification of industrial equipment based on an industrial cloud platform. Background Technology

[0002] With the development of industrial cloud platforms, the Industrial Internet, and equipment condition monitoring technologies, the identification of industrial equipment anomalies is gradually shifting from judging single parameter exceedances to the joint analysis of multi-source operational data. Existing technologies typically utilize continuous operational data generated by industrial equipment, such as speed, current, power, temperature, and environmental parameters, and combine this data with historical normal operation records to establish reference features. Methods such as time-series statistical analysis, residual analysis, distance metrics, cluster analysis, and machine learning are then used to describe changes in equipment operating status. For industrial equipment with repetitive operating patterns, methods such as periodic decomposition, sliding windows, and historical nearest neighbor comparisons can be used to characterize local operating features. Industrial cloud platforms can then be used to complete cross-equipment data storage, computation, and anomaly identification to support equipment operation monitoring and fault analysis.

[0003] However, when industrial equipment is driven by control commands, its operating state typically goes through a source operating state, a response process, and a target operating state, with variations in the duration and response speed of these different state transition processes. Existing methods based on fixed time windows or uniform sampling locations for anomaly comparisons are prone to misalignment between different state transition stages, causing normal response time differences to be perceived as abnormal changes. Furthermore, single-point sampling interference and true continuous deviations may coexist in industrial settings. If judgment is based solely on single-point residuals or multi-parameter composite distances, isolated interferences are prone to misjudgment, while persistent anomalies existing only in some operating parameters may be weakened by other normal operating parameters. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an industrial equipment anomaly identification method based on an industrial cloud platform to solve the problem of inaccurate multidimensional anomaly identification and stage positioning caused by misalignment of state transition timing and local disturbances in industrial equipment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for identifying industrial equipment anomalies based on an industrial cloud platform, comprising,

[0008] The state transition range is defined according to the control command. The device response position and the steady state formation position are determined by local distribution merging and minimum description length filtering. The same type of state transition phase is unified and a state transition phase sequence is generated.

[0009] Seasonal trend decomposition is used to distinguish between isolated residuals and continuous residuals, sampling interference is corrected and the direction of adjacent phase change is encoded to obtain a multidimensional feature sequence of state transition;

[0010] Phase pairing is performed between the state transition multidimensional feature sequence and the same type of normal historical feature sequence. The distance of the running parameter dimension is pre-sorted and the nearest non-overlapping subsequence is retrieved dimension by dimension to form the state transition multidimensional matrix outline.

[0011] The continuous deviation segments are determined based on the state transition multidimensional matrix profile. The results are then verified by combining the residual continuity and slope direction differences. The abnormal stage is determined according to the equipment response position and the stable state formation position, thus obtaining the industrial equipment abnormality identification results.

[0012] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform according to the present invention, wherein: the step of defining the state transition range according to control commands specifically includes:

[0013] By matching industrial equipment control records with continuous operation records according to equipment identification and time information, the location where control commands are issued and the source and target operating states corresponding to the control commands are determined.

[0014] The previous stable operating state continuous segment is taken as the operating segment before the control command occurs, the continuous operating changes after the control command occurs are taken as the operating change segment, and the continuous records between the operating change segment and the next control command occurrence position are taken as the subsequent operating segment.

[0015] By connecting the previous stable operating state continuous segment, the operating change segment, and the subsequent operating segment, the state transition analysis range is obtained.

[0016] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform according to the present invention, the generation of the state transition phase sequence specifically includes:

[0017] The equipment operating parameters and environmental impact parameters within the scope of state transition analysis are arranged according to a unified sampling time to form a multivariate operating sequence, and a continuous sliding window is used to divide the multivariate operating sequence into overlapping local operating segments;

[0018] By comparing the data distributions of local running segments with adjacent time intervals, and merging local running segments with continuously overlapping data distribution intervals, a candidate running segment sequence is obtained.

[0019] For the current candidate running segment in the candidate running segment sequence, determine the data description length and segment description length of the complete multivariate running sequence when retaining the current candidate running segment, and determine the data description length and segment description length of the complete multivariate running sequence after deleting the current candidate running segment, and compare the overall description lengths corresponding to the two states;

[0020] When the overall description length is shortened after deleting the current candidate running segment, the current candidate running segment is determined as a mixed running segment. The mixed running segment is deleted, and the data distribution of the remaining running segments before and after is re-compared to determine the new data distribution change location.

[0021] The overall description length comparison is repeated along the occurrence order of the candidate running segment sequence until the deletion of any candidate running segment can no longer shorten the overall description length. The location of the data distribution change between adjacent retained running segments is determined as the running change point.

[0022] According to the direction of change from the source operating state to the target operating state, check the operating sections before and after each operating change point. The operating change point where a continuous operating change first occurs after the source operating state leaves is determined as the equipment response position. The starting position of the operating section where the equipment operating state enters the target operating state and continues until the next state change is determined as the stable state formation position.

[0023] Using the consistency of the source operating state, control command type, and target operating state as the condition for state transition of the same type, multiple state transition processes between the location where the control command occurs and the location where the stable state is formed are mapped to phase positions with the same number, generating a state transition phase sequence.

[0024] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform described in this invention, the step of distinguishing between isolated residuals and continuous residuals using seasonal trend decomposition specifically includes:

[0025] Connect the same operating parameter in the state transition phase sequence according to the order of state transition, so that each complete state transition corresponds to a repeating cycle. Perform seasonal trend decomposition on the connected operating parameter sequence to obtain the residual sequence corresponding to each state transition phase.

[0026] Compare the distribution relationship of each residual with the other residuals in the same repeated running cycle, and compare the continuous change relationship of the residual with the adjacent residuals before and after. The single residual that deviates from the residual distribution of the same cycle and is discontinuous with the adjacent residuals before and after is identified as the isolated residual position.

[0027] The direction of change of adjacent residuals is examined along the phase sequence. The phase range in which multiple adjacent residuals change in the same direction and the adjacent residuals are continuous is determined as the continuous residual segment, thus obtaining the isolated residual position and the continuous residual segment.

[0028] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform described in this invention, the step of obtaining the state transition multidimensional feature sequence specifically includes:

[0029] Interpolation is performed using the effective operating parameter point closest to the isolated residual position, replacing the operating parameter corresponding to the isolated residual position, and retaining the continuous operating changes corresponding to the continuous residual segment to obtain the corrected operating parameter sequence;

[0030] The operating parameter sequence is rearranged and corrected according to the state transition phase sequence. The direction of change of operating parameters between adjacent phase positions is calculated, and the direction interval to which the direction of change belongs is converted into slope direction code in sequence to obtain the slope direction code sequence.

[0031] By maintaining the corresponding state transition phases for the modified operating parameter sequence, slope direction encoding sequence, device response duration, and steady state formation duration, a multidimensional feature sequence of state transitions is obtained.

[0032] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform described in this invention, the step of performing phase pairing of the state transition multidimensional feature sequence with the same type of normal historical feature sequence specifically includes:

[0033] Based on the condition that the source operating state, control command type and target operating state are consistent, the same type of normal historical state transition is determined from the history of normal state transition. The same type of normal historical feature sequence is obtained by using the same phase position as the state transition multidimensional feature sequence.

[0034] By matching the unified state transition phase with the state transition multidimensional feature sequence and the same type of normal historical feature sequence, the continuous phase positions are divided into current subsequences and historical subsequences of the same length, thus obtaining the phase-paired feature subsequence.

[0035] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform according to the present invention, the step of forming a state transition multidimensional matrix profile specifically includes:

[0036] Calculate the standardized Euclidean distance between the current subsequence and each historical subsequence in the phase-paired feature subsequence in each running parameter dimension, and sort the running parameter dimensions in descending order of standardized Euclidean distance to obtain the running parameter dimension distance order;

[0037] Starting from the first position of the sorting order, the dimension of the running parameter is increased sequentially for each sorting position until all dimensions of the running parameter are compared. The subsequence combination distance is calculated for each increase in the dimension of the running parameter, and the dimension-wise combination distance is obtained.

[0038] For each dimension of the operating parameter, candidate subsequences that overlap with the current subsequence in time are excluded from the normal historical feature sequences of the same type, and the historical subsequence with the smallest distance is determined according to the dimension-wise combination distance, so as to obtain the nearest non-overlapping subsequence corresponding to each dimension of the operating parameter.

[0039] Along the state transition phase, the distance to the nearest subsequence, the historical state transition position, and the dimensions of the operating parameters involved in the comparison are saved to form the outline of the state transition multidimensional matrix.

[0040] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform according to the present invention, wherein: determining the continuous deviation segment based on the state transition multidimensional matrix contour specifically includes:

[0041] According to the unified state transition phase, the distance of the nearest subsequence in the current state transition multidimensional matrix profile is matched with the matrix profile distance of the same normal historical state transition at the same phase position, and the normal distance boundary of each phase position is determined by the median level and quartile discrete range of the normal historical matrix profile distance.

[0042] Starting from the phase corresponding to the location where the control command is issued, the distance of the current nearest subsequence is compared sequentially with the normal distance boundary. The phase position where the distance of the current nearest subsequence first exceeds the normal distance boundary is determined as the deviation from the starting position. The continuous phase range that continues to exceed the normal distance boundary after the deviation from the starting position until it first returns to the normal distance boundary is determined as the continuous deviation segment.

[0043] By preserving the dimensions of the operating parameters and the most recent normal historical state transition position corresponding to the continuous deviation segment according to the state transition multidimensional matrix outline, the continuous deviation segment is obtained.

[0044] As a preferred embodiment of the industrial equipment anomaly identification method based on an industrial cloud platform described in this invention, the step of obtaining the industrial equipment anomaly identification result specifically includes:

[0045] The continuous deviation segment is mapped to the isolated residual position, continuous residual segment and slope direction coding sequence within the same phase range. When the continuous deviation segment only corresponds to the isolated residual position and the operation change of the preceding and following phases remains continuous, the abnormal state of the corresponding continuous deviation segment is canceled.

[0046] When a continuous deviation segment coincides with a continuous residual segment, and multiple operating parameter dimensions corresponding to the continuous deviation segment continue to deviate, while the slope direction encoding sequence is inconsistent with the slope direction encoding corresponding to the most recent normal historical state transition, the continuous deviation segment is identified as an abnormal segment, and the position of the first abnormal phase is determined.

[0047] The initial abnormal phase position is compared with the control command occurrence position, equipment response position, and steady state formation position. If the initial abnormal phase position is between the control command occurrence position and the equipment response position, it is determined to be a control response abnormality. If the initial abnormal phase position is between the equipment response position and the steady state formation position, it is determined to be an operational response abnormality. If the initial abnormal phase position is after the steady state formation position and the continuity relationship of the target operational state changes, it is determined to be a stable operational abnormality. The industrial equipment abnormality identification result is obtained.

[0048] Secondly, the present invention provides an industrial equipment anomaly identification system based on an industrial cloud platform, comprising,

[0049] The phase division module defines the state transition range according to the control command, determines the device response position and the stable state formation position through local distribution merging and minimum description length filtering, unifies the same type of state transition phases, and generates a state transition phase sequence.

[0050] The feature encoding module uses seasonal trend decomposition to distinguish between isolated residuals and continuous residuals, corrects sampling interference and encodes the direction of adjacent phase changes to obtain a multidimensional feature sequence of state transition.

[0051] The nearest neighbor retrieval module performs phase pairing between the state transition multidimensional feature sequence and the same type of normal historical feature sequence, pre-sorts the distance of the running parameter dimension and retrieves the nearest non-overlapping subsequence dimension by dimension to form the state transition multidimensional matrix outline.

[0052] The anomaly detection module determines continuous deviation segments based on the state transition multidimensional matrix profile, verifies the deviations by combining residual continuity and slope direction differences, and determines the anomaly stage based on the equipment response position and the stable state formation position, thus obtaining the industrial equipment anomaly identification result.

[0053] The beneficial effects of this invention are as follows: By limiting the scope of state transition analysis through control commands and using local distribution merging and minimum description length to filter operational variables, the equipment response position and steady-state formation position are determined by changes in actual operating data. Furthermore, similar state transitions with different durations are mapped to a unified phase, thereby reducing comparison misalignment caused by different response durations. Seasonal trend decomposition is further employed to distinguish between isolated residuals and continuous residuals, correcting only isolated sampling interference while retaining continuous operational changes, ensuring that subsequent features reflect the true state evolution. For multidimensional operating parameters, historical pairing is performed using the same state transition phase, and the nearest non-overlapping subsequences are retrieved dimension-by-dimensionally after pre-sorting according to the dimensional distance of the operating parameters. This allows abnormal changes in some operating parameters to independently enter the comparison process, reducing the masking of local anomalies by normal parameters. Finally, cross-checking is performed using continuous deviation segments, residual continuity, and slope direction differences, combined with the equipment response position and steady-state formation position to determine control response anomalies, operational response anomalies, and stable operation anomalies. This ensures that the anomaly identification results not only reflect whether a deviation has occurred but also correspond to the actual state transition stage, thereby improving the temporal consistency and stage positioning accuracy of anomaly identification results in complex industrial operations. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of an industrial equipment anomaly identification method based on an industrial cloud platform.

[0056] Figure 2 This is a flowchart of state transition phase partitioning and residual identification.

[0057] Figure 3 This is a flowchart for state transition multidimensional matrix contour retrieval and anomaly stage determination.

[0058] Figure 4 This is a diagram illustrating the comparison of abnormal operating responses of industrial equipment. Detailed Implementation

[0059] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0060] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0062] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for identifying industrial equipment anomalies based on an industrial cloud platform, including the following steps:

[0063] S1. Define the state transition range according to the control command, determine the equipment response position and the stable state formation position through local distribution merging and minimum description length filtering, unify the state transition phases of the same type, and generate a state transition phase sequence.

[0064] S1.1. The industrial cloud platform retrieves industrial equipment control records and continuous operation records according to equipment identifiers, and arranges the industrial equipment control records and continuous operation records corresponding to the same equipment identifier according to time information. The industrial equipment control record records at least the location where the control command occurred, the type of control command, the source operating status, and the target operating status; the continuous operation record continuously records the equipment operating parameters and environmental impact parameters according to the sampling time.

[0065] The control command occurrence location is the sampling position in the continuous operation record corresponding to the moment the control command was issued. When the control command issuance time and the sampling time are not completely consistent, the position corresponding to the nearest sampling time after the control command issuance time is determined as the control command occurrence location. The control command type is the command category in the industrial equipment control record that indicates the mode of adjustment of the current operating state. According to the actual control process, the specific command categories corresponding to start-up, shutdown, speed-up, and speed-down are recorded. The source operating state is the equipment operating state continuously maintained before the control command occurrence location, and the target operating state is the operating state that the equipment should maintain after the execution of the control command. The equipment operating parameters are continuously measured parameters in the continuous operation record that directly reflect the working state of the industrial equipment, including speed, operating current, and operating power. The environmental impact parameters are external environmental parameters recorded at the same sampling time as the equipment operating parameters, including ambient temperature and ambient humidity.

[0066] For a single control command, the system retrieves data backward from the control command's origin to a continuous segment where the industrial equipment maintained its previous stable operating state. This segment of the previous stable operating state is defined as the operating segment preceding the control command's origin. The previous stable operating state is defined as an operating state where the source operating state identifier remains consistent before the control command's origin and no new control commands are issued during this period. The source operating state identifier is checked sequentially from the control command's origin until a change in the source operating state identifier or the previous control command's origin is detected. The sampling segment that maintains the same source operating state identifier after this point is defined as the previous stable operating state's continuous segment. Starting from the control command's origin, the system checks continuous operating records chronologically, identifying the segment where the equipment's operating parameters begin to change continuously as the operating change segment. The system continues chronologically to the next control command's origin, defining the continuous records between the end of the operating change segment and the next control command's origin as the subsequent operating segment.

[0067] By connecting the previous stable operating state continuous segment, the operating change segment, and the subsequent operating segment in chronological order, the state transition analysis range is obtained. The state transition analysis range limits the continuous operating process corresponding to a single control command, ensuring that subsequent operating changes are determined only within the data range corresponding to the same control command, and preventing state changes corresponding to adjacent control commands from entering the same operating change determination process.

[0068] S1.2. Obtain equipment operating parameters and environmental impact parameters according to the state transition analysis scope, and arrange the equipment operating parameters and environmental impact parameters according to a unified sampling time. The same sampling time corresponds to a set of equipment operating parameters and environmental impact parameters, and different sampling times are arranged in sequence according to the occurrence time to form a multivariate operating sequence.

[0069] A continuous sliding window is used to sequentially extract continuous sampling positions along the time direction of the multivariate running sequence. The length of the continuous sliding window is set to 20 continuous sampling positions, and the window movement step size is set to 10 continuous sampling positions. The first continuous sliding window extracts 20 sampling positions continuously starting from the first sampling position of the state transition analysis range. The next continuous sliding window extracts 20 sampling positions starting from the 10th sampling position after the starting position of the previous continuous sliding window, so that the continuous sliding windows with adjacent time overlap by 10 sampling positions, and continue to move along the multivariate running sequence in the same way. When there are less than 20 sampling positions remaining at the end of the state transition analysis range, the remaining sampling positions are merged into the previous local running segment, and no new local running segment is established separately, resulting in mutually overlapping local running segments.

[0070] For each local operating section, the minimum and maximum sampled values ​​of the operating parameters and environmental impact parameters of each device are statistically analyzed within that local operating section. The range between the minimum and maximum sampled values ​​is taken as the data distribution interval for the corresponding operating parameters and environmental impact parameters.

[0071] The corresponding data distribution intervals of adjacent local operating segments are compared in chronological order. When the data distribution intervals of corresponding equipment operating parameters and environmental impact parameters of temporally adjacent local operating segments overlap, and the data distribution intervals of adjacent local operating segments continue to overlap, the consecutive local operating segments are merged into the same candidate operating segment; when the data distribution intervals of any corresponding equipment operating parameter no longer overlap and form a new consecutive distribution interval, the current candidate operating segment ends, and the next candidate operating segment is established starting from the local operating segment where the data distribution changes.

[0072] All candidate running segments are arranged according to their occurrence time to obtain a sequence of candidate running segments. By employing a continuous sliding window and merging of data distribution intervals, the sampling-level changes are first converted into segment-level changes. This reduces overly fine segmentation caused by individual sampling fluctuations and provides continuous candidate running segments for minimum description length screening.

[0073] S1.3. For the current candidate running segment in the candidate running segment sequence, calculate the overall description length required to describe the complete multivariate running sequence when retaining the current candidate running segment and when canceling the independent boundary of the current candidate running segment.

[0074] The minimum description length is calculated using an existing two-part coding method. The data description length encodes the sampled values ​​of equipment operating parameters and environmental impact parameters within each candidate operating segment according to the data distribution, and the total coding length corresponding to all sampled values ​​is accumulated. The segment description length is determined based on the number of candidate operating segments, their start and end positions, and the amount of information required to describe the data distribution of the candidate operating segments. The overall description length is obtained by adding the data description length and the segment description length.

[0075] When retaining the current candidate running segment, the current candidate running segment maintains its independent segment boundary, and the data description length and segment description length corresponding to the complete multivariate running sequence are determined respectively. When canceling the independent boundary of the current candidate running segment, the original sampled data corresponding to the current candidate running segment is not deleted. Instead, the segmentation state of the current candidate running segment as an independent candidate running segment is canceled, and the data distribution range to which the current sampled data belongs is re-determined according to the running segments retained before and after. The data description length and segment description length corresponding to the complete multivariate running sequence are determined again.

[0076] Compare the overall description lengths corresponding to the two states. If the overall description length after removing the independent boundary of the current candidate running segment is less than the overall description length when retaining the current candidate running segment, the current candidate running segment is determined as a mixed running segment, the independent segment boundary of the mixed running segment is removed, and the data distribution of the retained running segments before and after is re-compared to determine the new data distribution change location; if the overall description length after removing the independent boundary of the current candidate running segment is not less than the overall description length when retaining the current candidate running segment, the current candidate running segment is retained.

[0077] The overall description length comparison is repeated along the occurrence order of the candidate running segment sequence until the overall description length cannot be shortened further after canceling any independent boundary of the candidate running segment, thus obtaining the retained running segment sequence.

[0078] S1.4. Check the sequence of retained operating segments in order of occurrence. When the data distribution of adjacent retained operating segments no longer satisfies the continuous overlapping relationship, determine the position of the data distribution change between the previous retained operating segment and the next retained operating segment as the operating change point.

[0079] The data distribution change positions between all adjacent retained operating sections within the state transition analysis range are determined sequentially, and all data distribution change positions are arranged according to the sampling time to obtain the operating change points within the state transition analysis range.

[0080] The reserved operating segments before and after each operating change point are checked according to the source operating state and target operating state recorded by the control command. First, the operating change points are checked sequentially from the position where the control command occurred. When the reserved operating segment before the operating change point maintains the source operating state, and the reserved operating segment after the operating change point leaves the source operating state and continues until the next operating change point, the first operating change point after the source operating state leaves is determined as the equipment response position.

[0081] Continue to check subsequent reserved operating sections in chronological order from the equipment response position. When the equipment operating state corresponding to a certain reserved operating section enters the target operating state specified by the control command and remains there from the starting position of the reserved operating section to the next state change position, the starting position of the reserved operating section is determined as the stable state formation position.

[0082] A complete state transition process is determined by the location where the control command is issued, the equipment response location, and the steady-state formation location. The equipment response location indicates the position where the industrial equipment begins to leave the source operating state, and the steady-state formation location indicates the position where the industrial equipment enters and maintains the target operating state.

[0083] S1.5. Compare multiple complete state transition processes according to the source operating state, control command type, and target operating state, and classify complete state transition processes with the same source operating state, the same control command type, and the same target operating state into the same type of state transition process.

[0084] For each similar state transition process, the location where the control command occurs is taken as the starting position of the state transition, and the location where the stable state is formed is taken as the ending position of the state transition. Maintaining the original temporal order of each sampling location, linear interpolation is used to map similar state transition processes of different durations to a consistent number of phase positions. Equipment operating parameters and environmental influence parameters are synchronously mapped to the same phase position according to their corresponding sampling times. The equipment response position and the stable state formation position are determined according to the same mapping relationship to determine their corresponding phase positions.

[0085] Arrange the state transition processes of the same type according to the unified phase order, so that the same phase position corresponds to the same relative process in the process of the source running state to the target running state, and generate a state transition phase sequence.

[0086] S2. Seasonal trend decomposition is used to distinguish between isolated residuals and continuous residuals, sampling interference is corrected and the direction of adjacent phase change is encoded to obtain a multidimensional feature sequence of state transition.

[0087] S2.1. For the state transition phase sequence, maintain the same type of state transition relationship where the source operating state, control command type, and target operating state are consistent, and arrange each complete state transition according to the actual occurrence order of the state transition. Assume that each complete state transition contains the same number of phase positions after mapping. For each device operating parameter, sequentially connect the device operating parameters of each complete state transition under the same phase order, so that each complete state transition corresponds to a repeating operating cycle, forming a continuous operating parameter sequence.

[0088] Seasonal trend decomposition is performed on the continuous operating parameter sequence. The seasonal trend decomposition adopts the existing seasonal trend decomposition method based on local regression. The original method splits the time series into trend component, seasonal component and residual component. This embodiment does not change the operation structure of seasonal trend decomposition, but uses the number of phase positions contained in a determined complete state transition as the length of the repeated operation period, so that the seasonal component corresponds to the phase changes of the same type of state transition that occur repeatedly, and the residual component retains the local deviation of each state transition relative to the repeated phase changes.

[0089] The basic additive relationship for seasonal trend decomposition is:

[0090] ;

[0091] in, Represents the sequence of continuously running parameters. Device operating parameter values ​​for each sequence position, Indicates the position number of the continuously running parameter sequence; Indicates the first Trend components at each sequence position; Indicates the first Seasonal components of each sequence position; Indicates the first The residual components at each sequence position.

[0092] The period length for seasonal trend decomposition is taken as the number of single state transition phase positions after standardization; the seasonal smoothing length is an odd number not less than 7, the trend smoothing length is an odd number greater than the recommended range corresponding to the seasonal period, and the low-pass smoothing length is the smallest odd number not less than the period length. The residual components are re-trunculated according to the phase position corresponding to each complete state transition to obtain a residual sequence that corresponds one-to-one with each state transition phase.

[0093] S2.2. For each complete state transition, check the residuals one by one in phase order. For the current residual, temporarily exclude it from the residual set of the same repeated running cycle, sort the remaining residuals according to their numerical values, obtain the first quartile and the third quartile, and calculate the interquartile range.

[0094] The interquartile range is calculated as follows:

[0095] ;

[0096] in, The first quartile of the remaining residual for the same repeated cycle; The third quartile represents the remaining residual for the same repeated cycle; 1.5 represents the interquartile range; 1.5 represents the interquartile range extension factor used in the box plot fence rule, which is used to extend the range by 1.5 times outside the first and third quartiles respectively, thereby forming the normal value boundary of the residual distribution.

[0097] The boundary of residual distribution within the same period is determined using the box plot inner fence rule, expressed as:

[0098] ;

[0099] ;

[0100] in, Indicates the lower boundary of the residual distribution; This represents the upper boundary of the residual distribution.

[0101] The current residual is at and When in between, maintain the current residual as a normal residual within the same period; when the current residual is greater than When the residual deviation direction is positive, the current residual is less than 1%. When the residual deviation direction is negative, it is recorded as negative.

[0102] Residuals that exceed the residual distribution boundary are continuously checked along the state transition phase sequence. Residuals that deviate in the same direction and appear consecutively in adjacent phase positions are classified into the same residual segment. When a residual segment contains only a single phase position, the corresponding phase position is determined as an isolated residual position. When a residual segment contains at least two consecutive adjacent phase positions, the corresponding phase range is determined as a continuous residual segment.

[0103] Two adjacent positions represent the minimum number of adjacent positions required to form a continuous segment, and are not used as an empirical threshold.

[0104] The residuals at the first and last phase positions of a complete state transition lack complete bilateral adjacent positions within the same repeated cycle. Therefore, the first and last phase positions are not individually determined as isolated residual positions. The first and last phase positions are only incorporated into the continuous residual segment when they form continuous residuals with the same deviation direction as adjacent phase positions. This avoids using unilateral neighborhood corrections to define the state transition initiation boundary and the stable state formation boundary. This yields the isolated residual positions and continuous residual segments.

[0105] S2.3. Retrieve the corresponding equipment operating parameters within the same complete state transition according to the isolated residual position, and determine the equipment operating parameters corresponding to the isolated residual position as the sampling interference position; the equipment operating parameters corresponding to the continuous residual segment retain the original sampling value and do not perform interpolation replacement.

[0106] For each isolated residual position, the nearest equipment operating parameter point that has not been identified as an isolated residual position is searched forward and backward along the state transition phase sequence, and the previous and next valid operating parameter points are used as the endpoints of the linear interpolation.

[0107] Since the first and last phase positions of the complete state transition are not separately determined as isolated residual positions, each isolated residual position that needs interpolation has two valid operating parameter points before and after it.

[0108] The replacement value corresponding to the isolated residual position is calculated using the basic linear interpolation relation in numerical analysis. The expression is as follows:

[0109] ;

[0110] in, Indicates the first The interpolation running parameter values ​​corresponding to the state transition phase position being the isolated residual position. This indicates the state transition phase position number after the unified state transition; Indicates the most recent effective phase position before the isolated residual position. The corresponding equipment operating parameter values; Indicates the nearest valid phase position after the isolated residual position. The corresponding equipment operating parameter values; Indicates the position of the previous effective phase The corresponding state transition phase value; Indicates the first The state transition phase value corresponding to each state transition phase position; Indicates the position of the next effective phase The corresponding state transition phase value.

[0111] Replace the equipment operating parameters corresponding to isolated residual positions with the interpolated operating parameter values, while keeping the equipment operating parameters corresponding to continuous residual segments unchanged, to obtain the corrected operating parameter sequence.

[0112] S2.4. Arrange the corrected operating parameter sequence according to the determined state transition phase order, and for each device operating parameter in the same complete state transition, calculate the rate of change of the operating parameter between adjacent phase positions in sequence:

[0113] ;

[0114] in, Indicates the first The state transition phase position is to the first Rate of change of operating parameters between state transition phase positions; Indicates the first The corrected operating parameter values ​​corresponding to each state transition phase position; Indicates the first The corrected operating parameter values ​​corresponding to each state transition phase position; Indicates the first The state transition phase value corresponding to each state transition phase position; Indicates the first The state transition phase value corresponding to each state transition phase position.

[0115] Since S1.5 maps each complete state transition to a phase position that is the same in number and arranged in chronological order, Always positive. The positive and negative directions are consistent with the changing directions of adjacent operating parameters.

[0116] The slope direction interval is divided according to the sign of the rate of change of the operating parameters: Corresponding to the upward direction range, The corresponding direction range is maintained. The corresponding descent direction interval is used; the ascending direction, hold direction and descending direction are encoded in phase order respectively, and the encoding results of each phase position are arranged continuously to obtain the slope direction encoding sequence.

[0117] S2.5. For each complete state transition, the corresponding sampling time is obtained using the determined control command occurrence position, device response position, and steady-state formation position. The time interval between the control command occurrence position and the device response position is determined as the device response duration, and the time interval between the device response position and the steady-state formation position is determined as the steady-state formation duration, expressed as:

[0118] ;

[0119] ;

[0120] in, This indicates the sampling time corresponding to the location where the control command occurred. Sampling time corresponding to the device response position The determined device response duration; Indicates the sampling time corresponding to the device response position. The sampling time corresponding to the location where the steady state is formed The duration of a defined steady state; Indicates the sampling time corresponding to the location where the control command occurred; Indicates the sampling time corresponding to the device's response position; This indicates the sampling time corresponding to the location where the steady state is formed.

[0121] According to the state transition phase sequence, the modified operating parameter sequence and the slope direction encoding sequence are kept in the same phase position correspondence, and the device response duration and steady state formation duration are kept in the corresponding complete state transition record; the device response position and steady state formation position continue to maintain the phase position relationship determined in S1.5, and are arranged according to the phase order of each complete state transition to obtain the state transition multidimensional feature sequence.

[0122] S3. Pair the state transition multidimensional feature sequence with the same type of normal historical feature sequence in phase, pre-sort the distance of the running parameter dimension and search for the nearest non-overlapping subsequence dimension by dimension to form the state transition multidimensional matrix outline.

[0123] S3.1. Based on the source operating state, control command type, and target operating state, check the completed state transition processes from the historical state transition records stored on the industrial cloud platform. If the source operating state, control command type, and target operating state corresponding to the historical state transition record are consistent with the current state transition, and if the historical state transition record continues from the device response position to the target operating state until the stable state is formed, or maintains the target operating state after the stable state is formed until the next state change, then the corresponding historical state transition record is identified as a normal historical state transition of the same type.

[0124] The normal historical state transitions of the same type are processed according to the phase mapping method in S1.5, so that the normal historical state transitions of the same type have the same number and the same order of state transition phases as the current state transitions; the normal historical state transitions of the same type are processed according to the same residual processing, sampling interference correction and slope direction encoding method in S2.1 to S2.5, so that the corrected operating parameter sequence, slope direction encoding sequence, equipment response duration and steady state formation duration in the historical record maintain the same field relationship with the current state transition multidimensional feature sequence, so as to obtain the normal historical feature sequence of the same type.

[0125] Normal historical feature sequences and current state transition multidimensional feature sequences of the same type use the same state transition phase and the same operating parameter dimension. Subsequent distance comparisons are no longer based on the original sampling time, thus limiting historical state transitions with different durations to the same state transition process.

[0126] S3.2. Divide the current subsequence and the historical subsequence according to the unified state transition phase. For the current state transition multidimensional feature sequence, start from the state transition start phase and extract continuous phase positions in phase order, and form the current subsequence by moving phase positions one by one; for each current subsequence, extract continuous phase ranges with the same start phase position and the same end phase position from each normal historical feature sequence of the same type to form the historical subsequence.

[0127] The phase span covered by the subsequence is converted to a unified state transition phase using the linear interpolation relationship of S1.5, based on the local running time span corresponding to the continuous sliding window. The number of converted phase positions is rounded to an integer; if the number of phase positions is less than 2, it is rounded to 2; if the number of phase positions exceeds the total number of phase positions included in a complete state transition, it is rounded to the total number of phase positions included in the complete state transition. A phase position count of 2 is the minimum data length required to form adjacent phase changes and is not used as an anomaly detection threshold. During a state transition comparison, all current subsequences and historical subsequences use the same length.

[0128] The current subsequence is matched with the normal historical feature sequences of the same type sequentially along the state transition phase to obtain the phase-paired feature subsequence. The phase pairing method restricts the current subsequence to be compared only with the historical subsequence corresponding to the same state transition process, avoiding cross-stage nearest neighbor matching between data in the control response phase and data in the stable operation phase.

[0129] S3.3. For the phase-paired feature subsequence, extract the dimensions of each device operating parameter within the state transition multidimensional feature sequence. The slope direction encoding sequence, device response duration, and steady-state formation duration maintain the original state transition phase relationship.

[0130] For any current subsequence and any historical subsequence, calculate the subsequence mean and standard deviation under the same equipment operating parameter dimension, and use the standard score method to make the current subsequence and historical subsequence dimensionless respectively.

[0131] For including The current subsequence at the i-th phase position, the i-th The standardized relationships among the dimensions of device operating parameters are as follows:

[0132] ;

[0133] The normalized relation corresponding to the historical subsequence is:

[0134] ;

[0135] in, Indicates the first subsequence in the current subsequence. The dimension of device operating parameters in the first Standardized operating parameter values ​​for the phase positions of each subsequence; Represents the first subsequence in the history sequence The dimension of device operating parameters in the first Standardized operating parameter values ​​for the phase positions of each subsequence; Indicates the first subsequence in the current subsequence. The dimension of device operating parameters in the first Corrected operating parameter values ​​for the phase position of each subsequence; Represents the first subsequence in the history sequence The dimension of device operating parameters in the first Corrected operating parameter values ​​for the phase position of each subsequence; Indicates the current subsequence number. The mean of each device's operating parameters; Indicates the first subsequence of history The mean of each device's operating parameters; Indicates the current subsequence number. Standard deviation of each device operating parameter dimension; Indicates the first subsequence of history Standard deviation of each device operating parameter dimension; For the dimension number of the equipment operating parameters, This represents the phase position index within the subsequence. This represents the number of phase positions contained in the current subsequence and the historical subsequence.

[0136] Calculate the standardized Euclidean distance between the current subsequence and the historical subsequence under the same device operating parameter dimension according to the classic definition of Euclidean distance:

[0137] ;

[0138] in, Indicates the first The current subsequence and the... The historical subsequence at the _ ... Standardized Euclidean distance across the dimensions of each device's operating parameters; Indicates the current subsequence number; Indicates the sequence number of the historical subsequence.

[0139] Calculate the standardized Euclidean distance for each of the running parameter dimensions to obtain the set of running parameter dimension distances between each current subsequence and the historical subsequence.

[0140] S3.4. For the set of distances between operating parameter dimensions, arrange the device operating parameter dimensions in descending order according to the standardized Euclidean distance; when the standardized Euclidean distances are the same, keep the original order of the device operating parameters in the state transition multidimensional feature sequence unchanged, and obtain the determined order of distances between operating parameter dimensions.

[0141] Suppose a comparison includes Each device operating parameter dimension This represents the total number of device operating parameter dimensions involved in the standardized Euclidean distance calculation for the state transition multidimensional feature sequence. The dimensions are expanded sequentially from the first position in the sorted order, based on the distance between the first and second positions. The first step retains only the device operating parameter dimension at the first position; the second step retains the dimensions from the first to the second position; and so on, increasing the dimensions of the device operating parameters at the next sorted position until all dimensions are included. Each device's operating parameter dimension is used for comparison.

[0142] For each retention For each device operating parameter dimension, a multi-dimensional Euclidean distance relationship is used to merge the corresponding standardized Euclidean distances:

[0143] ;

[0144] in, Indicates the first The current subsequence and the... The historical subsequence is in front Subsequence combination distance when each sorting parameter dimension is used for comparison; This indicates the number of dimensions of the runtime parameters currently involved in calculating the subsequence combination distance, with values ​​increasing from 1 to... ; Indicates the order of the distance dimension in the running parameters. The device operating parameter dimension corresponding to each sorting position Indicates the sorting position number; Indicates the first The current subsequence and the... The historical subsequence at the _ ... Standardized Euclidean distance across the dimensions of each device's operating parameters.

[0145] The combined distances of subsequences corresponding to different numbers of operating parameter dimensions are stored separately, and their numerical values ​​are not compared between different numbers of operating parameter dimensions. Subsequent nearest subsequence searches only compare the combined distances of subsequences with the same number of operating parameter dimensions. The operating parameter dimension distances are sorted first, and then the operating parameter dimensions of the devices involved in the comparison are increased one by one. This ensures that changes that occur only in some device operating parameters can first enter the lower-dimensional combined distance, and will not be masked by the approximate changes of other operating parameters when all device operating parameters are involved at the same time.

[0146] S3.5. For each current subsequence and each dimension of operating parameters, examine all historical subsequences corresponding to the same state transition phase in all similar normal historical feature sequences. If the original sampling time interval corresponding to the historical subsequence does not intersect with the original sampling time interval corresponding to the current subsequence, retain the historical subsequence as a non-overlapping candidate subsequence; if the original sampling time interval corresponding to the historical subsequence intersects with the original sampling time interval corresponding to the current subsequence, exclude the historical subsequence from the current nearest neighbor search range.

[0147] Compare all non-overlapping candidate subsequences according to the subsequence combination distance corresponding to the same number of operating parameter dimensions, and determine the historical subsequence with the smallest subsequence combination distance as the nearest non-overlapping subsequence under the corresponding number of operating parameter dimensions; when the same minimum subsequence combination distance occurs, select the historical subsequence with the earlier historical state transition time, so that the method of determining the position of the nearest non-overlapping subsequence remains unique.

[0148] For each dimension of the operating parameters, the nearest non-overlapping subsequence, the nearest subsequence distance, and the historical state transition position are stored. The historical state transition position records the complete state transition record of the nearest non-overlapping subsequence and the starting state transition phase position of the nearest non-overlapping subsequence, thus obtaining the nearest neighbor comparison results for each dimension of the operating parameters.

[0149] Arrange the nearest neighbor comparison results in phase order of the current state transition. For each current subsequence starting phase position, save the nearest subsequence distance corresponding to one running parameter dimension up to all running parameter dimensions. At the same time, save the historical state transition position and the running parameter dimension involved in the comparison corresponding to each nearest subsequence distance, forming the outline of the state transition multidimensional matrix.

[0150] S4. Determine the continuous deviation segment based on the state transition multidimensional matrix profile, verify it by combining the residual continuity and slope direction difference, and determine the abnormal stage according to the equipment response position and the stable state formation position to obtain the industrial equipment abnormality identification result.

[0151] S4.1. Based on the condition that the source running state, control command type and target running state are consistent, obtain the state transition multidimensional matrix profile corresponding to the normal historical state transition of the same type, and keep the number of dimensions of the running parameters separate for distance statistics to avoid the nearest subsequence distances corresponding to different numbers of dimensions of running parameters entering the same distance distribution.

[0152] For any state transition phase position and any number of operating parameters, the distances of the nearest subsequences corresponding to the same phase position of similar normal historical state transitions are arranged in ascending order of value, forming a normal historical matrix contour distance sequence. The first quartile, median, and third quartile are determined according to this normal historical matrix contour distance sequence. When the normal historical matrix contour distance sequence contains an odd number of distances, the middle distance after sorting is determined as the median, and the medians of the distance sequences on either side of the middle distance are calculated as the first and third quartiles. When the normal historical matrix contour distance sequence contains an even number of distances, the average of the two middle distances after sorting is determined as the median, and the medians of the first and second halves of the distance sequence are calculated as the first and third quartiles.

[0153] The discrete range of the normal historical matrix contour distance is calculated using the interquartile range, and the expression is as follows:

[0154] ;

[0155] in, Indicates the first The state transition phase positions are at The interquartile range of the normal historical matrix contour distance when each running parameter dimension is compared. Indicates the first The state transition phase positions are at The first quartile of the normal historical matrix contour distance sequence when each running parameter dimension is used for comparison; Indicates the first The state transition phase positions are at The third quartile of the normal historical matrix contour distance sequence when each running parameter dimension is involved in the comparison.

[0156] The normal distance boundary for the current state transition phase is determined by combining the median level and the interquartile range, and the expression is:

[0157] ;

[0158] in, Indicates the first The state transition phase positions are at Normal distance boundaries when comparing multiple operating parameter dimensions; Indicates the first The state transition phase positions are at The median of the normal historical matrix contour distance sequence when each running parameter dimension is involved in the comparison.

[0159] S4.2. Starting from the state transition phase corresponding to the location where the control command occurs, along the unified state transition phase sequence, compare the nearest subsequence distance of the current state transition multidimensional matrix profile with the normal distance boundary corresponding to the same state transition phase for each dimension of the operating parameter.

[0160] When the distance of the current nearest subsequence does not exceed the normal distance boundary, the corresponding state transition phase is maintained at the normal distance position; when the distance of the current nearest subsequence exceeds the normal distance boundary for the first time, the corresponding state transition phase is determined to be deviated from the starting position, and the distances of subsequent nearest subsequences are compared sequentially along the state transition phase from the deviated starting position.

[0161] When the distance of the nearest subsequence after the deviation from the starting position continuously exceeds the corresponding normal distance boundary, the corresponding state transition phase is continuously recorded. When the distance of the nearest subsequence returns to within the normal distance boundary for the first time, the state transition phase before the first return to within the normal distance boundary is determined as the deviation from the end position, and the continuous phase range from the deviation from the starting position to the deviation from the end position is determined as the continuous deviation segment. When the distance of the nearest subsequence continues to exceed the normal distance boundary until the current complete state transition ends, the last phase position of the complete state transition is determined as the deviation from the end position.

[0162] Based on the operating parameter dimensions and historical state transition positions stored together with the nearest subsequence distance in the state transition multidimensional matrix profile, the number of operating parameter dimensions corresponding to each continuous deviation segment, the operating parameter dimensions involved in the comparison, and the nearest normal historical state transition position are retained to obtain the continuous deviation segments. The continuous deviation segments are limited by the two actual phase positions of the distance first crossing the boundary and the first returning to the boundary, and the fixed number of consecutive phases is no longer used as the deviation judgment condition.

[0163] S4.3. Each continuous deviation segment is assigned a unified state transition phase and corresponding to the isolated residual position and continuous residual segment, and the slope direction encoding sequence stored in the same state transition phase is obtained.

[0164] When a continuous deviation segment contains only isolated residual positions and does not coincide with the phase of a continuous residual segment, check the slope direction codes corresponding to the phase before the isolated residual position and the phase after the isolated residual position. When the slope direction codes corresponding to two adjacent change intervals are consistent, it is determined that the two sides of the isolated residual position maintain the same running change direction. Combined with the processing results of interpolation replacement using the effective running parameter points before and after, the abnormal state of the corresponding continuous deviation segment is canceled.

[0165] The phase overlap relationship between continuous deviation segments and continuous residual segments is determined according to the state transition phase position sequence. For a continuous deviation segment and a continuous residual segment, the starting and ending phase positions of the continuous deviation segment and the continuous residual segment are obtained respectively. The larger of the two starting phase position sequences is determined as the overlapping starting phase position, and the smaller of the two ending phase position sequences is determined as the overlapping ending phase position. When the overlapping starting phase position sequence is not greater than the overlapping ending phase position sequence, the continuous phase range between the overlapping starting phase position and the overlapping ending phase position is determined as the overlapping phase range, and the number of overlapping phase positions is equal to the overlapping ending phase position sequence minus the overlapping starting phase position sequence plus 1; when the overlapping starting phase position sequence is greater than the overlapping ending phase position sequence, there is no overlapping phase range between the continuous deviation segment and the continuous residual segment. When one continuous deviation segment corresponds to multiple continuous residual segments, the overlapping phase ranges corresponding to each continuous residual segment and the continuous deviation segment are determined in the same way.

[0166] When the continuous deviation segment and the continuous residual segment have phase overlap, the continuous deviation segment is retained for slope direction difference verification.

[0167] S4.4. For the retained continuous deviation segments, obtain the number of saved operating parameter dimensions, the operating parameter dimensions participating in the comparison, and the position of the most recent normal historical state transition, and obtain the slope direction encoding sequence of the corresponding historical state transition within the same phase range according to the position of the most recent normal historical state transition.

[0168] When the number of operating parameter dimensions is greater than 1, the current slope direction code corresponding to each operating parameter dimension involved in the comparison within the continuous deviation segment is compared with the slope direction code corresponding to the most recent normal historical state transition. For each determined overlapping phase range, the current slope direction code is compared with the slope direction code corresponding to the most recent normal historical state transition in the order from the overlapping start phase position to the overlapping end phase position. If the current slope direction code corresponding to each phase position within the overlapping phase range is inconsistent with the slope direction code corresponding to the most recent normal historical state transition, and the distance of the nearest subsequence corresponding to the same number of operating parameter dimensions continuously exceeds the normal distance boundary within the corresponding overlapping phase range, the corresponding continuous deviation segment is determined as an abnormal segment.

[0169] The abnormal sections are checked sequentially along the state transition phases, and the starting position of the abnormal section is determined as the first abnormal phase position.

[0170] According to the determined phase mapping relationship, obtain the state transition phases corresponding to the control command occurrence position, equipment response position, and stable state formation position, and compare the first abnormal phase position with the three state transition phases in sequence.

[0171] If the first abnormal phase position is located between the phase corresponding to the control command occurrence position and the phase corresponding to the equipment response position, but has not reached the phase corresponding to the equipment response position, the abnormal phase is determined to be a control response abnormality. If the first abnormal phase position starts from the phase corresponding to the equipment response position and is located before the phase corresponding to the stable state formation position, the abnormal phase is determined to be an operational response abnormality. If the first abnormal phase position starts from the phase corresponding to the stable state formation position, the operation records between the stable state formation position and the next state change position are checked. When the equipment operation state no longer maintains the determined target operation state, the abnormal phase is determined to be a stable operation abnormality.

[0172] The state transition type is recorded based on the correspondence between the source operating state, control command type, and target operating state, and mapped back to the sampling position corresponding to the continuous operating record according to the position of the first abnormal phase. Combining the saved equipment identifier, the identified abnormal segment, the saved dimensions of the operating parameters involved in the comparison, and the most recent normal historical state transition position, the equipment identifier, state transition type, first abnormal position, abnormal occurrence stage, corresponding abnormal operating parameters, most recent normal state transition record, and related operating segment are recorded to obtain the industrial equipment abnormality identification result.

[0173] This embodiment also provides an industrial equipment anomaly identification system based on an industrial cloud platform, including:

[0174] The phase division module defines the state transition range according to the control command, determines the device response position and the stable state formation position through local distribution merging and minimum description length filtering, unifies the same type of state transition phases, and generates a state transition phase sequence.

[0175] The feature encoding module uses seasonal trend decomposition to distinguish between isolated residuals and continuous residuals, corrects sampling interference and encodes the direction of adjacent phase changes to obtain a multidimensional feature sequence of state transition.

[0176] The nearest neighbor retrieval module performs phase pairing between the state transition multidimensional feature sequence and the same type of normal historical feature sequence, pre-sorts the distance of the running parameter dimension and retrieves the nearest non-overlapping subsequence dimension by dimension to form the state transition multidimensional matrix outline.

[0177] The anomaly detection module determines continuous deviation segments based on the state transition multidimensional matrix profile, verifies the deviations by combining residual continuity and slope direction differences, and determines the anomaly stage based on the equipment response position and the stable state formation position, thus obtaining the industrial equipment anomaly identification result.

[0178] In summary, this invention maintains the correspondence between industrial equipment control records, operational changes, residual changes, slope direction encoding, and the multi-dimensional state transition matrix profile according to a unified state transition phase, thus forming a continuous evidence chain from control events and operational responses to multi-dimensional deviation features in the anomaly determination process. Isolated residuals, after interpolation correction, no longer directly participate in anomaly determination, while continuous residual segments retain their original operational changes, reducing the impact of occasional factors such as sampling spikes and communication disturbances on the identification results. Simultaneously, operational parameter dimensions participate in nearest-neighbor comparisons according to distance differences, preserving local anomaly features that only appear in some operational parameters. Consistency verification is performed using residual continuity, slope direction differences, and the most recent normal historical state transitions, ensuring that anomaly results simultaneously correspond to specific operational parameters, anomaly phases, and normal historical references. This enhances the interpretability and traceability of industrial equipment anomaly identification results and improves the stability of anomaly determination under complex operating conditions.

[0179] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying anomalies in industrial equipment based on an industrial cloud platform, characterized in that: include, The state transition range is defined according to the control command. The device response position and the steady state formation position are determined by local distribution merging and minimum description length filtering. The same type of state transition phase is unified and a state transition phase sequence is generated. Seasonal trend decomposition is used to distinguish between isolated residuals and continuous residuals, sampling interference is corrected and the direction of adjacent phase change is encoded to obtain a multidimensional feature sequence of state transition; Phase pairing is performed between the state transition multidimensional feature sequence and the same type of normal historical feature sequence. The distance of the running parameter dimension is pre-sorted and the nearest non-overlapping subsequence is retrieved dimension by dimension to form the state transition multidimensional matrix outline. The continuous deviation segments are determined based on the state transition multidimensional matrix profile. The results are then verified by combining the residual continuity and slope direction differences. The abnormal stage is determined according to the equipment response position and the stable state formation position, thus obtaining the industrial equipment abnormality identification results.

2. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 1, characterized in that: The process of defining the state transition range according to control commands specifically includes: By matching industrial equipment control records with continuous operation records according to equipment identification and time information, the location where control commands are issued and the source and target operating states corresponding to the control commands are determined. The previous stable operating state continuous segment is taken as the operating segment before the control command occurs, the continuous operating changes after the control command occurs are taken as the operating change segment, and the continuous records between the operating change segment and the next control command occurrence position are taken as the subsequent operating segment. By connecting the previous stable operating state continuous segment, the operating change segment, and the subsequent operating segment, the state transition analysis range is obtained.

3. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 2, characterized in that: The generation of the state transition phase sequence specifically includes: The equipment operating parameters and environmental impact parameters within the scope of state transition analysis are arranged according to a unified sampling time to form a multivariate operating sequence, and a continuous sliding window is used to divide the multivariate operating sequence into overlapping local operating segments; By comparing the data distributions of local running segments with adjacent time intervals, and merging local running segments with continuously overlapping data distribution intervals, a candidate running segment sequence is obtained. For the current candidate running segment in the candidate running segment sequence, determine the data description length and segment description length of the complete multivariate running sequence when retaining the current candidate running segment, and determine the data description length and segment description length of the complete multivariate running sequence after deleting the current candidate running segment, and compare the overall description lengths corresponding to the two states; When the overall description length is shortened after deleting the current candidate running segment, the current candidate running segment is determined as a mixed running segment. The mixed running segment is deleted, and the data distribution of the remaining running segments before and after is re-compared to determine the new data distribution change location. The overall description length comparison is repeated along the occurrence order of the candidate running segment sequence until the deletion of any candidate running segment can no longer shorten the overall description length. The location of the data distribution change between adjacent retained running segments is determined as the running change point. According to the direction of change from the source operating state to the target operating state, check the operating sections before and after each operating change point. The operating change point where a continuous operating change first occurs after the source operating state leaves is determined as the equipment response position. The starting position of the operating section where the equipment operating state enters the target operating state and continues until the next state change is determined as the stable state formation position. Using the consistency of the source operating state, control command type, and target operating state as the condition for state transition of the same type, multiple state transition processes between the location where the control command occurs and the location where the stable state is formed are mapped to phase positions with the same number, generating a state transition phase sequence.

4. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 3, characterized in that: The method of using seasonal trend decomposition to distinguish between isolated residuals and continuous residuals specifically includes: Connect the same operating parameter in the state transition phase sequence according to the order of state transition, so that each complete state transition corresponds to a repeating cycle. Perform seasonal trend decomposition on the connected operating parameter sequence to obtain the residual sequence corresponding to each state transition phase. Compare the distribution relationship of each residual with the other residuals in the same repeated running cycle, and compare the continuous change relationship of the residual with the adjacent residuals before and after. The single residual that deviates from the residual distribution of the same cycle and is discontinuous with the adjacent residuals before and after is identified as the isolated residual position. The direction of change of adjacent residuals is examined along the phase sequence. The phase range in which multiple adjacent residuals change in the same direction and the adjacent residuals are continuous is determined as the continuous residual segment, thus obtaining the isolated residual position and the continuous residual segment.

5. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 4, characterized in that: The obtained state transition multidimensional feature sequence specifically includes: Interpolation is performed using the effective operating parameter point closest to the isolated residual position, replacing the operating parameter corresponding to the isolated residual position, and retaining the continuous operating changes corresponding to the continuous residual segment to obtain the corrected operating parameter sequence; The operating parameter sequence is rearranged and corrected according to the state transition phase sequence. The direction of change of operating parameters between adjacent phase positions is calculated, and the direction interval to which the direction of change belongs is converted into slope direction code in sequence to obtain the slope direction code sequence. By maintaining the corresponding state transition phases for the modified operating parameter sequence, slope direction encoding sequence, device response duration, and steady state formation duration, a multidimensional feature sequence of state transitions is obtained.

6. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 5, characterized in that: The step of performing phase pairing between the state transition multidimensional feature sequence and the same type of normal historical feature sequence specifically includes: Based on the condition that the source operating state, control command type and target operating state are consistent, the same type of normal historical state transition is determined from the history of normal state transition. The same type of normal historical feature sequence is obtained by using the same phase position as the state transition multidimensional feature sequence. By matching the unified state transition phase with the state transition multidimensional feature sequence and the same type of normal historical feature sequence, the continuous phase positions are divided into current subsequences and historical subsequences of the same length, thus obtaining the phase-paired feature subsequence.

7. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 6, characterized in that: The formation of the state transition multidimensional matrix contour specifically includes: Calculate the standardized Euclidean distance between the current subsequence and each historical subsequence in the phase-paired feature subsequence in each running parameter dimension, and sort the running parameter dimensions in descending order of standardized Euclidean distance to obtain the running parameter dimension distance order; Starting from the first position of the sorting order, the dimension of the running parameter is increased sequentially for each sorting position until all dimensions of the running parameter are compared. The subsequence combination distance is calculated for each increase in the dimension of the running parameter, and the dimension-wise combination distance is obtained. For each dimension of the operating parameter, candidate subsequences that overlap with the current subsequence in time are excluded from the normal historical feature sequences of the same type, and the historical subsequence with the smallest distance is determined according to the dimension-wise combination distance, so as to obtain the nearest non-overlapping subsequence corresponding to each dimension of the operating parameter. Along the state transition phase, the distance to the nearest subsequence, the historical state transition position, and the dimensions of the operating parameters involved in the comparison are saved to form the outline of the state transition multidimensional matrix.

8. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 7, characterized in that: The determination of continuous deviation segments based on the state transition multidimensional matrix profile specifically includes: According to the unified state transition phase, the distance of the nearest subsequence in the current state transition multidimensional matrix profile is matched with the matrix profile distance of the same normal historical state transition at the same phase position, and the normal distance boundary of each phase position is determined by the median level and quartile discrete range of the normal historical matrix profile distance. Starting from the phase corresponding to the location where the control command is issued, the distance of the current nearest subsequence is compared sequentially with the normal distance boundary. The phase position where the distance of the current nearest subsequence first exceeds the normal distance boundary is determined as the deviation from the starting position. The continuous phase range that continues to exceed the normal distance boundary after the deviation from the starting position until it first returns to the normal distance boundary is determined as the continuous deviation segment. By preserving the dimensions of the operating parameters and the most recent normal historical state transition position corresponding to the continuous deviation segment according to the state transition multidimensional matrix outline, the continuous deviation segment is obtained.

9. The industrial equipment anomaly identification method based on an industrial cloud platform as described in claim 8, characterized in that: The obtained industrial equipment anomaly identification results specifically include: The continuous deviation segment is mapped to the isolated residual position, continuous residual segment and slope direction coding sequence within the same phase range. When the continuous deviation segment only corresponds to the isolated residual position and the operation change of the preceding and following phases remains continuous, the abnormal state of the corresponding continuous deviation segment is canceled. When a continuous deviation segment coincides with a continuous residual segment, and multiple operating parameter dimensions corresponding to the continuous deviation segment continue to deviate, while the slope direction encoding sequence is inconsistent with the slope direction encoding corresponding to the most recent normal historical state transition, the continuous deviation segment is identified as an abnormal segment, and the position of the first abnormal phase is determined. The initial abnormal phase position is compared with the control command occurrence position, equipment response position, and steady state formation position. If the initial abnormal phase position is between the control command occurrence position and the equipment response position, it is determined to be a control response abnormality. If the initial abnormal phase position is between the equipment response position and the steady state formation position, it is determined to be an operational response abnormality. If the initial abnormal phase position is after the steady state formation position and the continuity relationship of the target operational state changes, it is determined to be a stable operational abnormality. The industrial equipment abnormality identification result is obtained.

10. An industrial equipment anomaly identification system based on an industrial cloud platform, based on the industrial equipment anomaly identification method based on an industrial cloud platform as described in any one of claims 1 to 9, characterized in that: include, The phase division module defines the state transition range according to the control command, determines the device response position and the stable state formation position through local distribution merging and minimum description length filtering, unifies the same type of state transition phases, and generates a state transition phase sequence. The feature encoding module uses seasonal trend decomposition to distinguish between isolated residuals and continuous residuals, corrects sampling interference and encodes the direction of adjacent phase changes to obtain a multidimensional feature sequence of state transition. The nearest neighbor retrieval module performs phase pairing between the state transition multidimensional feature sequence and the same type of normal historical feature sequence, pre-sorts the distance of the running parameter dimension and retrieves the nearest non-overlapping subsequence dimension by dimension to form the state transition multidimensional matrix outline. The anomaly detection module determines continuous deviation segments based on the state transition multidimensional matrix profile, verifies the deviations by combining residual continuity and slope direction differences, and determines the anomaly stage based on the equipment response position and the stable state formation position, thus obtaining the industrial equipment anomaly identification result.