An industrial time series data processing method and device, a terminal device, and a medium

Abstract

The present application relates to a kind of industrial time series data processing method, device, terminal equipment and medium, method includes obtaining the original data generated by industrial equipment, and the original data is converted to obtain corresponding original time series data;Original time series data is detected in multidimension, and the abnormal data detected in the data structure of time series data point is retained and the state of abnormal data is marked;Based on the gap interval of original effective data is handled based on preset interpolation strategy, and interpolation candidate data is generated;Original effective data refers to the original time series data obtained from industrial equipment sampling and not judged as abnormal data;According to the unified time grid of pre-set requirement, and around each target time setting alignment pre-set time window;In pre-set time window, based on the dynamic decision target result point of asymmetric priority absorption rule, and the result data corresponding to target result point is output.The result data of the present application is high in reliability.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a method, apparatus, terminal equipment, and medium for processing industrial time-series data. Background Technology

[0002] In industrial time-series data processing scenarios, raw time-series data from sensors, field controllers, equipment terminals, and edge acquisition systems forms the core data foundation for equipment status analysis, operational condition playback, trend modeling, fault alarms, and statistical decision-making. Raw time-series data commonly suffers from issues such as out-of-bounds values, random noise, error codes, instantaneous jumps, and continuous abnormal segments. To meet the requirements of data playback, operational condition analysis, or modeling, it is necessary to clean and correct abnormal data, interpolate missing data segments, and simultaneously align the time series of multi-source heterogeneous data based on a unified time grid.

[0003] In existing industrial time-series data processing solutions, data cleaning, interpolation, and time-series alignment are generally performed as independent serial processing steps, with each step only focusing on the data correction effect of its own stage. In addition, most data cleaning strategies usually physically delete or completely remove abnormal data points after detecting abnormal data, erroneous data, or invalid sampling points. Although this processing method can ensure the continuity of output data in the short term, it will cause a series of deep-seated data reliability defects.

[0004] In summary, existing industrial time-series data processing technologies cannot simultaneously guarantee the integrity, authenticity, and auditability of time-series data, and cannot meet the actual needs of high-precision industrial analysis, accurate assessment of equipment status, and data compliance traceability. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, terminal equipment, and medium for processing industrial time-series data, so as to solve the problem of poor data reliability caused by abnormal processing of industrial time-series data in the prior art.

[0006] To address the aforementioned problems, the present invention relates to a method for processing industrial time-series data, comprising: Acquire raw data generated by industrial equipment and perform transformation processing on the raw data to obtain corresponding raw time-series data; Multi-dimensional anomaly detection is performed on the original time series data, and the detected abnormal data is retained in the data structure of the time series data points and the abnormal data is marked with a status. Based on a preset interpolation strategy, the gap interval of the original valid data is processed to generate interpolation candidate data; the original valid data refers to the original time-series data obtained from sampling of industrial equipment and not identified as abnormal data. Construct a uniform time grid according to preset requirements, and set a preset time window aligned around each target time; Within the preset time window, the target result point is dynamically decided based on the asymmetric priority absorption rule, and the result data corresponding to the target result point is output.

[0007] In some embodiments, the multi-dimensional anomaly detection of the original time-series data includes: Perform range verification on the original time series data; After the range verification, the slope verification is performed; After the slope verification, the median deviation verification is performed; After the median deviation check, a consistency check is performed.

[0008] In some embodiments, the data structure of time-series data points includes timestamps, numerical values, source types, quality markers, abnormal states, abnormal types, interpolation types, lineage identifiers, and upstream source reference relationships; wherein, the lineage identifier is used to link data lineages throughout the entire processing chain.

[0009] In some embodiments, the multi-dimensional anomaly detection of the original time series data further includes: setting an initial score for the confidence level of each original time series data, updating the confidence level of the original time series data according to the results of multi-dimensional anomaly detection, and writing the updated confidence level into the extended field of the pedigree metadata.

[0010] In some embodiments, the pedigree metadata includes a base field and extended fields; the base field includes a pedigree identifier; and the extended fields include the confidence level of the original value.

[0011] In some embodiments, if the time between two adjacent original valid data exceeds a preset step size, the points corresponding to the two adjacent original valid data are taken as the two endpoints of the gap interval; the interpolation candidate data generated in each gap interval includes source type, interpolation type and endpoint data information.

[0012] In some embodiments, the result data includes timestamps, numerical values, device models, device numbers, measurement point identifiers, source types, quality markers, abnormal states, abnormal types, interpolation types, lineage identifiers, and upstream source reference relationships.

[0013] To address the aforementioned problems, the present invention relates to an industrial time-series data processing device, comprising a data access module, a data cleaning module, an interpolation module, an alignment module, and a result output module. The data access module is used to acquire raw data generated by industrial equipment and to convert and process the raw data to obtain corresponding raw time-series data. The data cleaning module is used to perform multi-dimensional anomaly detection on the original time series data, and retain the detected abnormal data in the data structure of the time series data points and mark the abnormal data with a status. The interpolation module is used to process the gap interval of the original valid data based on a preset interpolation strategy to generate interpolation candidate data; the original valid data refers to the original time-series data obtained from sampling of industrial equipment and not identified as abnormal data; The alignment module is used to construct a unified time grid according to preset requirements and set a preset time window for alignment around each target time; The result output module is used to dynamically decide on the target result point based on the asymmetric priority absorption rule within the preset time window, and output the result data corresponding to the target result point.

[0014] To address the aforementioned problems, the present invention relates to a terminal device comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the aforementioned method for processing industrial time-series data.

[0015] To address the aforementioned problems, the present invention relates to a computer-readable storage medium storing a computer program, which, when executed on a processor, implements the aforementioned method for processing industrial time-series data.

[0016] The beneficial effects of this invention are as follows: The present invention provides a method for processing industrial time-series data, comprising: acquiring raw data generated by industrial equipment and transforming the raw data to obtain corresponding raw time-series data; performing multi-dimensional anomaly detection on the raw time-series data, retaining the detected anomaly data in the data structure of the time-series data points, and marking the anomaly data with a status; processing the gap interval of the raw valid data based on a preset interpolation strategy to generate interpolation candidate data; the raw valid data refers to the raw time-series data obtained from sampling by industrial equipment that has not been identified as anomaly data; constructing a unified time grid according to preset requirements, and setting a preset time window aligned around each target time; dynamically deciding on the target result point within the preset time window based on an asymmetric priority absorption rule, and outputting the result data corresponding to the target result point. By performing multi-dimensional anomaly detection on the raw time-series data, and when anomalies are detected, instead of performing physical deletion operations, the anomaly information is written into the status marker. This ensures the traceability, interpretability, and engineering usability of the result data without disrupting the continuity of the processing link, through status markers and priority absorption rules. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below: Figure 1 This is a flowchart illustrating the method for processing industrial time-series data in a specific embodiment of the present invention. Detailed Implementation

[0018] To make the technical objectives, technical solutions, and beneficial effects of the present invention clearer, the technical solutions of the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention; that is, the described embodiments are merely some embodiments of the present invention, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0020] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0021] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] In-depth analysis of the poor reliability of existing industrial data revealed the following: (i) In existing industrial time series data processing solutions, data cleaning, interpolation and time series alignment are generally performed as independent serial processing steps. Each step only focuses on the data correction effect of its own stage and lacks a unified data traceability and quality marking mechanism that runs through the entire chain.

[0023] (ii) The following considerations should be made regarding the potential data reliability defects caused by directly physically deleting or completely removing outliers during the cleaning phase: First, directly deleting abnormal data will cause a serious data lineage break problem, that is, it is impossible to effectively distinguish whether the current time series position is "a real empty state where the device has not generated sampling data" or "original sampling data existed but was cleaned and removed due to abnormal judgment", resulting in the failure of the entire chain data traceability, the complete loss of abnormal traces, and the inability to audit or review the data processing process.

[0024] Secondly, the interpolated data and the original sampled data lack a unified source identification and priority constraint mechanism. That is, the data generated by interpolation is highly consistent with the original sampled data in terms of numerical form. If the data source and generation method are not marked, the subsequent time alignment process is prone to mistakenly prioritizing the interpolated data and discarding the adjacent real original data, resulting in what appears to be continuous but deviates from the actual operating conditions of the equipment.

[0025] Furthermore, the current multi-stage processing flow lacks a unified lineage inheritance and status tracking mechanism. That is, when data goes through multiple processing steps such as cleaning, interpolation, and alignment, there is a lack of unified continuity rules for the processing status, correction logic, and data source information of each stage. Subsequent processing steps are prone to overwriting or losing the processing traces of previous stages, making it impossible to achieve a traceable and auditable governance effect throughout the entire life cycle.

[0026] Therefore, current industrial time-series data processing chains generally suffer from technical defects such as loss of anomaly traces, broken data lineage, chaotic prioritization of raw and interpolated data, indistinguishable data credibility, and lack of end-to-end traceability capabilities. This results in downstream alarms, trend analysis, statistical reports, and model training being unable to be used hierarchically based on credibility when the output results lack quality markers, cleaning status, and source information. Furthermore, it struggles to ensure the integrity, authenticity, and auditability of time-series data, failing to meet the practical needs of high-precision industrial analysis, accurate equipment status assessment, and compliant data traceability. To address this, this application, during the cleaning, interpolation, and alignment of industrial time-series data, ensures the traceability, interpretability, and engineering usability of the resulting data without disrupting the continuity of the processing chain through the preservation of anomaly data traces, status marking, source marking, and priority inclusion rules.

[0027] like Figure 1 As shown, the present invention relates to a method for processing industrial time-series data, comprising: S100: Acquire raw data generated by industrial equipment and perform conversion processing on the raw data to obtain the corresponding raw time series data.

[0028] Specifically, raw data from one or more industrial devices is acquired, and all raw data is uniformly converted into standardized time-series records. The standardized time-series records after conversion include at least uniform time format, uniform measurement point identification, and uniform device identification.

[0029] S200, perform multi-dimensional anomaly detection on the original time series data, retain the detected abnormal data in the data structure of the time series data points, and mark the abnormal data with a status.

[0030] In this embodiment of the invention, multi-dimensional anomaly detection includes, but is not limited to, range verification, slope verification, median deviation verification, and consistency verification, which are performed in a set order and the results are accumulated.

[0031] In one implementation, multi-dimensional anomaly detection of the original time-series data includes: (1) Perform range verification on the original time series data.

[0032] In this embodiment, range verification is a fundamental method for identifying out-of-bounds abnormal raw data in the industrial time-series data cleaning process. Specifically, based on the equipment's rated operating conditions, physical range, and process threshold preset upper and lower limits, a data quality verification method is used to determine whether each data point of the raw time-series data exceeds the legal range for each sampling point. This fundamentally filters out physically impossible values, i.e., screens out-of-bounds abnormalities in hardware range and process range.

[0033] (2) After range verification, slope verification is performed.

[0034] In this embodiment, slope verification is used as a second-order mutation verification method for the original time series data of industrial equipment. It calculates the slope of change per unit time based on the time interval and numerical difference between adjacent time series sampling points, and combines it with the maximum allowable rate of change threshold of the equipment to identify abrupt changes such as instantaneous jumps, pulse interference, and acquisition distortion at adjacent moments.

[0035] (3) After slope verification, median deviation verification is performed.

[0036] In this embodiment, median deviation verification belongs to the sliding window statistical anomaly detection category and is performed after range verification and slope verification. Using the median of the effective values ​​within a continuous time window of the current measurement point as a benchmark, the deviation between the current value and the median is calculated. Combined with a preset deviation threshold, long-term drift, small-amplitude continuous shifts, and isolated discrete anomalies are identified, compensating for the shortcomings of range verification and instantaneous slope verification in identifying slow-drift anomalies. In other words, median deviation verification identifies outlier data points within a local window.

[0037] (4) After the median deviation check, a consistency check is performed.

[0038] In this embodiment, consistency verification is a multi-point, cross-dimensional logical verification. It utilizes the inherent process linkage relationship of downhole equipment to verify the numerical matching between a group of physically coupled measurement points at the same time, thus compensating for the implicit data defect of single-point verification that cannot identify "normal values ​​at a single point but logical contradictions at multiple measurement points". Through consistency verification, it is determined whether the point conflicts with surrounding points and neighboring equipment.

[0039] In this invention, the original time-series data points are subjected to dimensional anomaly detection and multi-rule accumulation in sequence. Only the abnormal data is semantically marked and not physically deleted. That is, if a single original time-series data point meets a certain verification, it will be judged as abnormal data, and the abnormal data will be marked as abnormal, while the original value will be retained.

[0040] In this embodiment of the invention, the data structure of time-series data points includes fields such as timestamp, numerical value, source type, quality marker, anomaly status, anomaly type, interpolation type, lineage identifier, and upstream source reference relationship. The source type distinguishes between interpolated data, original data, compensation values, or other derived values; the anomaly status indicates whether the data point is determined to be an anomaly; the upstream source reference relationship expresses the association between the current data point and upstream points; and the lineage identifier is used to link the data lineage throughout the entire processing chain.

[0041] In this embodiment of the invention, marking abnormal data with a status includes at least: setting the abnormal status to abnormal and updating its verification rules in the abnormal type.

[0042] For example, if an anomaly is found after slope verification, the anomaly state is "anomaly," and the anomaly type is "slope verification anomaly." If anomalies are found after both range verification and consistency verification, the anomaly state is "anomaly," and the anomaly type is "range verification anomaly" and "consistency verification anomaly." Therefore, the anomaly type must include at least one. Anomaly data that has been status-marked should not be included as a candidate point in the final alignment output.

[0043] In this embodiment of the invention, the multi-dimensional anomaly detection process further includes: setting an initial score for the confidence level of each original time series data, updating the confidence level of the original time series data based on the results of multi-dimensional anomaly detection, and writing the updated confidence level into the extended field of the lineage metadata of the original time series data; deducting a first score when the range check is not met, deducting a second score when the slope check or median check is not met, and deducting a third score when the consistency check is not met; wherein, the first score is greater than the second score, and the second score is greater than the third score; the updated confidence level is written into the lineage metadata of the original time series data, which can be used in the subsequent alignment stage to compare the priority of low-confidence original time series data and interpolation points.

[0044] The pedigree metadata includes basic fields and extended fields. The basic fields include the pedigree identifier, and the extended fields include the confidence level of the original value. The pedigree identifier in the basic fields is a unique identifier that connects the entire processing chain and is located in the data structure. The extended fields are extended attributes that expand upon the pedigree identifier and are used to supplement the credibility and unreliability of the original data points. The extended fields are not added to the basic field list of individual data points, but are stored as bypass metadata or pedigree record content pointed to by the pedigree identifier. They can be written to disk separately or stored in a bypass index or business database as needed for the project. They do not forcibly occupy columns in the main data structure, thus preserving complete credibility information without significantly increasing the main storage overhead.

[0045] Specifically, the pedigree metadata for raw value confidence includes: raw_confidence_score is used to represent the current confidence score of the raw point; raw_confidence_level is used to represent high, medium, and low levels; raw_rule_hits is used to represent the exception rules that were hit; raw_rule_count is used to represent the number of rules that were hit; raw_neighbor_support is used to indicate whether a neighboring point or a nearby device provides support for that point; raw_source_status is used to indicate whether the source point is currently normal, abnormal, or degradeable. The decision_basis is used to indicate the reason for ultimately choosing or abandoning that point.

[0046] The above fields are used to record the credible and uncredible evidence for the origin point in its entirety, rather than just retaining an anomaly marker.

[0047] The original time-series data of this invention undergoes a four-layer progressive quality check: first, range check identifies out-of-bounds values; then, slope check identifies instantaneous jump interference; subsequently, sliding window mean deviation check identifies sensor drift and slow shifts in operating conditions. After the first three layers of single-point time-series checks are completed, a multi-measurement point correlation consistency check is performed. Based on the process coupling constraints of multiple related measurement points under the same equipment and at the same time, it is determined whether there are logical conflicts in the data. During the cleaning stage, the original time-series data is not deleted throughout the four-layer check process. All anomaly judgments (e.g., anomaly rules, number of hit rules, etc.), calculation parameters (e.g., credit rating), and confidence decay results are uniformly written into the original value confidence lineage metadata, thus preventing data lineage breakage from the root. Data lineage refers to the source, state, and processing path information that a result point experiences from the original sampling to the final output. The data structure of this invention achieves lineage preservation, that is, it is not only a numerical carrier, but also carries source, processing process, and result quality information. By performing multi-dimensional anomaly detection on the raw time-series data, and upon detecting anomalies, instead of physically deleting them, the system writes the anomaly information into the data point's status tag, achieving an architectural shift from "physical discarding" to "semantic preservation." For anomalies such as out-of-bounds or jumps, the system expands the data point's metadata structure, transforming it into a "lineage node" with auditable value. This transformation ensures the complete reproducibility of the original working conditions in the industrial field and solves the problem of lineage breakage caused by missing data.

[0048] S300 processes the gaps in the original valid data based on a preset interpolation strategy to generate interpolation candidate data.

[0049] In this embodiment of the invention, the preset interpolation strategy includes linear interpolation and previous value padding. It should be noted that other point padding methods adapted to industrial time series are also applicable to this invention. For example, in linear interpolation, interpolation candidate data is calculated according to the time ratio of the preceding and following endpoints; in previous value padding, interpolation candidate data directly inherits from the previous valid point; when configured without padding, no interpolation candidate data corresponding to the candidate point is generated. The interpolation candidate data corresponding to each interpolation candidate point can be traced back to a specific upstream original point.

[0050] In this embodiment of the invention, original valid data refers to original time-series data obtained from sampling of industrial equipment that has not been identified as abnormal data.

[0051] In this embodiment of the invention, if the time interval between two adjacent original valid data points exceeds a preset step size, the points corresponding to these two adjacent original valid data points are taken as the two endpoints of the gap interval. The gap interval between the two endpoints needs to be interpolated to fill in time-series data points to ensure the integrity of the time-series data.

[0052] In some embodiments, the interpolation candidate data generated within each gap interval includes source type, interpolation type, and endpoint data information; the endpoint data information includes two original valid data points corresponding to the two endpoints of the gap interval, along with their timestamps, numerical values, measurement point identifiers, device identifiers, abnormal states, and lineage metadata (e.g., the current confidence score of the original point). In other embodiments, a quality category is also included to distinguish it from the original numerical value. For example, the source type is interpolation data; the interpolation type is linear interpolation.

[0053] The difference candidate data of this invention can trace back the source of the difference point (which two original valid data were used to calculate it), support subsequent recalculation and auditing, and provide a basis for judgment in the alignment stage (such as whether the endpoints at both ends are high-confidence original points, and whether the gap is too long).

[0054] Specifically, each gap interval is determined sequentially according to time sequence. When generating interpolation candidate data, candidate point metadata is written based on endpoint data information as the source information for interpolation. If the time difference between gap intervals is large, candidate points from multiple time points are continuously added along the time grid. For example, if the step size is 24 seconds and the difference between two valid points is 96 seconds, then candidate points need to be generated sequentially according to the 24-second grid. In this way, the alignment stage obtains a continuous segment of candidate data, rather than a single isolated filler point.

[0055] S400 constructs a uniform time grid according to preset requirements and aligns it around each target time to obtain a preset time window.

[0056] In this embodiment of the invention, a unified time grid is constructed according to the needs of playback, statistics or analysis, and an aligned time window is set around each target time point.

[0057] A unified time grid is essentially a set of target time points with fixed step sizes, such as one point every 24 seconds starting from midnight of the current day. Subsequent raw points and interpolated points are not directly compared to each other, but rather compete within a window around these target time points. In this way, data from different sources, different devices, and different timestamps can be placed on the same timeline, providing a unified benchmark for subsequent playback, comparison, statistics, and point selection.

[0058] It should be noted that the unified time grid in this step and the unified time format in step S100 address different problems and serve different purposes. Step S100 addresses the question of whether records can enter the same processing chain, so its focus is on unifying the time format, measurement point identifiers, and device identifiers. This step addresses the question of whether these records can be compared on the same timeline, so its focus is on defining a unified target time point and window. The former is record standardization, while the latter is timeline standardization.

[0059] S500 dynamically decides on the target result point based on the asymmetric priority absorption rule within a preset time window and outputs the result data corresponding to the target result point.

[0060] In this embodiment of the invention, for each target time point, the target result point is dynamically decided based on the asymmetric priority absorption algorithm within a preset time window, including: selecting the target original point within the preset time window; if there is no target original point, dynamically selecting interpolation candidate data as the target interpolation point according to the confidence score of the interpolation algorithm; and using the target original point and the target interpolation point as the target result point.

[0061] In some embodiments, selecting a target origin point within the preset time window includes: determining whether there is any valid original data within the preset time window; if so, directly absorbing the origin point corresponding to the valid original data closest to the target time point as the target origin point; if not, then filtering out the origin points corresponding to the original time-series data with reference value from the abnormal data as the target origin points; wherein, having reference value means that the abnormal data is only slight noise and not an abnormal point corresponding to device offline. When abnormal data is used as a target origin point, it is absorbed with a lower weight, that is, when used as a target result point, the weight of abnormal data is less than the weight of valid original data.

[0062] It is also understandable that if there are original valid data points within the preset time window, they are given the highest weight and incorporated into the target result point; if there are only abnormal data and interpolation candidate data within the preset time window, they are downgraded and incorporated according to the confidence score or lineage priority to obtain the target result point.

[0063] Based on the data structure of raw time-series data and interpolation candidate data, the output result data corresponding to the target result point includes timestamps, numerical values, and lineage metadata. Specifically, after selecting a target result point, the absorption path of each target result point is written into the lineage field of the target result point. For example, the absorption path includes absorption strategy, number of competing points, weight, and other absorption path content. The absorption path is used to record how the result point is selected within a preset time window.

[0064] In addition, the pedigree metadata of the absorption path and the confidence of the original value belong to the same pedigree information system, and the two are linked by the pedigree identifier. The pedigree metadata focuses on recording whether the original point is trustworthy or not (generated in the anomaly detection stage), while the absorption path here focuses on recording the final selection decision process in the alignment stage (generated in the absorption stage); the two together constitute the complete pedigree link of this result point from the original sampling to the final output.

[0065] The final output data includes fields such as timestamp, value, device model, device number, measurement point identifier, source type, quality mark, abnormal status, abnormality type, interpolation type, lineage identifier, and upstream source reference relationship. Among these, the device model identifies the model category of the device to which the data point belongs; the device number uniquely identifies the specific device that generated the data; and the measurement point identifier identifies the specific measurement point (e.g., a sensor channel) corresponding to the value. The first five fields describe "which measurement point output what value at this moment," while the latter fields describe "where this value comes from, its quality, whether there is an abnormal history, and its relationship with upstream data." It should be noted that when the raw data is uniformly converted into standardized time-series records, the system simultaneously unifies the device identifier (model_name, device_no) and measurement point identifier (data_point), and these fields are passed hop-by-hop along with the lineage field in subsequent cleaning, interpolation, and alignment stages, ensuring that each result point can be traced back to the specific device and measurement point.

[0066] This invention prioritizes the original valid data for initial retrieval, then uses a lineage-based downgrading decision-making method to determine the target original point from the abnormal data. Finally, it selects interpolation points through a supplementary absorption decision-making method, injecting lineage identifiers into the output data. This ensures that each output point has a traceable "processing fingerprint," meeting industrial auditing requirements. It is understood that in the time alignment stage, this invention introduces an asymmetric priority absorption strategy to solve the "competition" problem between original sampled data and interpolated data within the time window. The priority order is: original valid points > original marked abnormal points > algorithm interpolation points. When executing alignment decisions, the system dynamically retrieves lineage identifiers from the candidate pool to ensure that physical sampled data has decision-making dominance, thereby greatly reducing the risk of data "drift" due to over-interpolation.

[0067] To retain a more complete decision-making process, fields such as `raw_confidence_score`, `raw_confidence_level`, `raw_rule_hits`, `decision_basis`, and `decision_weight` can be further expanded. This way, the output will not only show the final value but also why it was selected, the initial confidence level of the original point, and the rules used during window contention. If the final disk format is not suitable for directly storing so many fields, this information can be stored separately in bypass metadata, index files, or the business database.

[0068] In response to the high-frequency concurrency characteristics of industrial big data, this invention optimizes the storage method of lineage information. By replacing long string records with predefined flag bits (i.e., using several binary bits to represent the status) and parent ref (i.e., a reference identifier pointing to the upstream source record), it ensures that while increasing lineage tracking capabilities, it does not significantly increase the system's I / O burden and storage costs.

[0069] In a preferred embodiment of the present invention, the method further includes lineage-driven intelligent alarms; specifically, if the result point is marked as "original," where "original" indicates that the result comes directly from the original sampling point, the alarm engine performs a standard threshold judgment. If the result point is marked as "interpolated," where "interpolated" indicates that the result comes from interpolation points, the alarm engine automatically increases the alarm trigger threshold and pushes a low-priority warning of "abnormal calculated data" to the administrator, thereby effectively filtering false alarms caused by interpolation errors.

[0070] This invention transforms traditional destructive data cleaning into a semantically preserved traceability mechanism. During the cleaning phase, abnormal data is marked with a status instead of physically deleted, preserving the integrity of industrial operating conditions. In the alignment phase, an asymmetric priority absorption rule is introduced, executing a dynamic decision-making logic of "prioritizing original valid points and downgrading interpolated candidate points" based on lineage identifiers within a time window. Simultaneously, it achieves atomic-level full-link inheritance of data lineage, driving differentiated business logic such as downstream alarm classification and statistical stratification. This invention can significantly improve the authenticity, interpretability, and business linkage capabilities of industrial time-series processing results, effectively solving the problems of lineage breakage and data drift.

[0071] The industrial time-series data processing method of the present invention has the following beneficial effects: Architectural innovation: By transforming data structures from destructive data cleaning to semantic preservation, the physical integrity of industrial data is guaranteed.

[0072] Scientific decision-making: The asymmetric priority absorption rule ensures the core position of real sampled data in the alignment process and effectively suppresses "data drift".

[0073] Business integration: Lineage information directly participates in downstream business logic, enabling intelligent alerts and hierarchical statistics based on data credibility.

[0074] Explainability: End-to-end lineage identification enables each output point to have a traceable "processing fingerprint," meeting the needs of industrial auditing.

[0075] High efficiency: The optimized bit-stamped storage structure ensures atomic-level lineage tracing without increasing system load.

[0076] The present invention also provides an industrial time-series data processing device, including a data access module, a data cleaning module, an interpolation module, an alignment module and a result output module; The data access module is used to acquire raw data generated by industrial equipment and to convert and process the raw data to obtain corresponding raw time-series data. The data cleaning module is used to perform multi-dimensional anomaly detection on the original time series data, and retain the detected abnormal data in the data structure of the time series data points and mark the abnormal data with a status. The interpolation module is used to process the gap interval of the original valid data based on a preset interpolation strategy to generate interpolation candidate data; the original valid data refers to the original time-series data obtained from sampling of industrial equipment and not identified as abnormal data; The alignment module is used to construct a unified time grid according to preset requirements and set a preset time window for alignment around each target time; The result output module is used to dynamically decide on the target result point based on the asymmetric priority absorption rule within the preset time window, and output the result data corresponding to the target result point.

[0077] It is understood that the apparatus in this embodiment corresponds to the industrial time-series data processing method in the above embodiments, and the options in the above embodiments are also applicable to this embodiment, so they will not be described again here.

[0078] This application also provides a terminal device, exemplary of which includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to enable the terminal device to perform the above-described method for processing industrial timing data.

[0079] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0080] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), and Electrically Erasable Programmable Read-Only Memory (EEPROM). The memory stores computer programs, and the processor, upon receiving execution instructions, can execute the computer programs accordingly.

[0081] This application also provides a computer-readable storage medium for storing the computer program used in the aforementioned terminal device. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0083] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0084] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0085] Finally, it should be noted that the above embodiments are only for illustration and not for limiting the technical solutions of the present invention. Any equivalent substitutions, modifications or partial substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for processing industrial time-series data, characterized in that, include: Acquire raw data generated by industrial equipment and perform transformation processing on the raw data to obtain corresponding raw time-series data; Multi-dimensional anomaly detection is performed on the original time series data, and the detected abnormal data is retained in the data structure of the time series data points and the abnormal data is marked with a status. Based on a preset interpolation strategy, the gap interval of the original valid data is processed to generate interpolation candidate data; the original valid data refers to the original time-series data obtained from sampling of industrial equipment and not identified as abnormal data. Construct a uniform time grid according to preset requirements, and set a preset time window aligned around each target time; Within the preset time window, the target result point is dynamically decided based on the asymmetric priority absorption rule, and the result data corresponding to the target result point is output.

2. The method for processing industrial time-series data according to claim 1, characterized in that, The multi-dimensional anomaly detection of the original time-series data includes: Perform range verification on the original time series data; After the range verification, the slope verification is performed; After the slope verification, the median deviation verification is performed; After the median deviation check, a consistency check is performed.

3. The method for processing industrial time-series data according to claim 2, characterized in that, The data structure of time-series data points includes timestamp, value, source type, quality marker, abnormal state, abnormal type, interpolation type, lineage identifier, and upstream source reference relationship; wherein, the lineage identifier is used to connect the data lineage in the entire processing chain.

4. The method for processing industrial time-series data according to claim 3, characterized in that, The multi-dimensional anomaly detection of the original time series data further includes: setting an initial score for the confidence level of each original time series data, updating the confidence level of the original time series data according to the results of multi-dimensional anomaly detection, and writing the updated confidence level into the extended field of the pedigree metadata.

5. The method for processing industrial time-series data according to claim 4, characterized in that, The pedigree metadata includes basic fields and extended fields; the basic fields include pedigree identifiers; and the extended fields include the confidence level of the original values.

6. The method for processing industrial time-series data according to claim 1, characterized in that, If the time between two adjacent valid original data exceeds a preset step size, the points corresponding to the two adjacent valid original data are taken as the two endpoints of the gap interval; the interpolation candidate data generated in each gap interval includes source type, interpolation type and endpoint data information.

7. The method for processing industrial time-series data according to claim 1, characterized in that, The results data include timestamps, numerical values, equipment models, equipment numbers, measurement point identifiers, source types, quality markers, abnormal states, abnormal types, interpolation types, lineage identifiers, and upstream source reference relationships.

8. An industrial time-series data processing device, characterized in that, It includes a data access module, a data cleaning module, an interpolation module, an alignment module, and a result output module; The data access module is used to acquire raw data generated by industrial equipment and to convert and process the raw data to obtain corresponding raw time-series data. The data cleaning module is used to perform multi-dimensional anomaly detection on the original time series data, and retain the detected abnormal data in the data structure of the time series data points and mark the abnormal data with a status. The interpolation module is used to process the gap interval of the original valid data based on a preset interpolation strategy to generate interpolation candidate data; the original valid data refers to the original time-series data obtained from sampling of industrial equipment and not identified as abnormal data; The alignment module is used to construct a unified time grid according to preset requirements and set a preset time window for alignment around each target time; The result output module is used to dynamically decide on the target result point based on the asymmetric priority absorption rule within the preset time window, and output the result data corresponding to the target result point.

9. A terminal device, characterized in that, The terminal device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the industrial time-series data processing method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed on a processor, implements a method for processing industrial time-series data according to any one of claims 1-7.

Images