A hybrid time series-based power system anomaly identification method and related device

Abstract

The application discloses a kind of based on hybrid timing's power system anomaly identification method and related device, and collects the original multi-source sampling data of industrial field;Original multi-source sampling data is normalized, and normalized data is obtained;The normalized data is adaptively embedded with Gaussian kernel processing, and the hybrid timing with continuous timing and discrete timing is converted into the form of uniform continuous timing, and uniform continuous space data is obtained;Unified continuous space data is input into abnormal measuring point tracking algorithm, and the root abnormal measuring point that causes industrial system failure is obtained.The application can effectively identify the anomaly of power system even in the hybrid timing scene of multi-source physical quantity interference coupling.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis and reliability engineering, and specifically relates to a method and related device for identifying anomalies in a power system based on hybrid timing. Background Technology

[0002] In the context of industrial intelligent development, multi-source time series data has become a key basis for monitoring the operating status of complex power systems. However, current power system anomaly detection faces three major challenges: First, multi-source sensor data suffers from inconsistent sampling frequencies, inconsistent time bases, and severe data gaps, making it difficult to achieve standardized data fusion; second, the mixing and overlap of continuous time-series signals and discrete state signals makes it difficult to effectively extract and utilize the key fault information contained in the discrete data; finally, under the coupling effect of multi-physics fields, anomaly features often exhibit cross-domain propagation characteristics, and traditional single-point detection methods cannot accurately track the anomaly propagation path. Existing technologies often employ simple irregular data deletion operations to address these problems, which not only loses key feature information but also makes it difficult to establish effective cross-source correlation models, resulting in insufficient anomaly identification accuracy and severely restricting the accuracy and reliability of power system fault diagnosis. Summary of the Invention

[0003] The purpose of this invention is to provide a method and related apparatus for identifying anomalies in dynamic systems based on hybrid time series, so as to overcome the problem that it is difficult to track abnormal measurement points under the interference coupling of multiple physical quantities in the dynamic system process. Even in the hybrid time series scenario with interference coupling of multiple physical quantities, this invention can effectively identify anomalies in dynamic systems.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for anomaly identification in dynamical systems based on hybrid time series includes the following steps: Step 1: Collect raw multi-source sampling data from the industrial site; Step 2: Normalize the original multi-source sampling data to obtain normalized data; Step 3: Perform adaptive Gaussian kernel embedding on the normalized data to transform the mixed time series with continuous and discrete time series into a unified form with continuous time series, thus obtaining unified continuous spatial data; Step 4: Input the unified continuous spatial data into the abnormal measurement point tracking algorithm to obtain the root cause abnormal measurement points that cause industrial system failures.

[0005] Furthermore, the normalization process for the original multi-source sampling data specifically includes: The monitoring parameter screening method based on the proportion of effective parameters filters the sensor dimensions of the original multi-source sampling data; The backfilling method is used to align the timestamps of the remaining sensor channel data, with 0s as the unified start timestamp and the last timestamp of the shortest process as the unified end timestamp. Finally, the data is standardized after expanding according to the monitoring parameter dimensions.

[0006] Furthermore, the monitoring parameter screening method based on the proportion of effective parameters filters the sensor dimensions of the original multi-source sampling data, and the calculation formula is as follows:

[0007]

[0008]

[0009] In the formula: ——No. The percentage of each monitoring parameter in the entire process; —Total number of processes; ——No. The first process Sampling time series under each monitoring parameter; —A set of effective monitoring parameters; ——No. One monitoring parameter; —The set threshold for the percentage of effective parameters; , , J This represents the total number of sensors.

[0010] Furthermore, the backfilling method is used to align the timestamps of the remaining sensor channel data, with 0s as the unified start timestamp and the last timestamp of the shortest flow as the unified end timestamp. Specifically:

[0011] In the formula: —Resampled time series Timestamp of each sampling point; —Resampled time series Data values ​​of each sampling point; ——Original time series Timestamp of each sampling point; ——Original time series The data values ​​of each sampling point; among which, , , The number of sampling points in the original time series. This represents the number of sampling points in the resampled time series.

[0012] Furthermore, the data is standardized after being expanded according to the monitoring parameter dimensions, specifically as follows:

[0013] In the formula, x ′ is the standardized value. μ and σ These are the mean and standard deviation of the monitoring parameter for all processes.

[0014] Furthermore, step 3 specifically includes: For discrete variables Each non-zero excitation time point With the non-zero excitation time point as the center, the bandwidth is superimposed. Gaussian kernels are used to generate continuous time series:

[0015] Integrating the Gaussian kernels at all non-zero excitation time points yields the initial continuous variables:

[0016] in, Given a set of discrete time points, where t is the time dimension variable, and designed to capture multi-scale time patterns, Gaussian kernels with different bandwidths, whose bandwidth parameters are generated according to an arithmetic sequence:

[0017] in, and The weight coefficients of each Gaussian kernel are learnable parameters. Normalization via Softmax:

[0018] in, The weighting coefficients of the Gaussian kernel. The weight for each timestamp is M, where M is the number of Gaussian kernels with different bandwidths. The initial continuous variables are represented as a weighted superposition result;

[0019] in, Gaussian curves at each timestamp, For time variables, This represents the variance of the Gaussian curve. This is random noise disturbance. This represents the variance of the random noise. Optimize Gaussian kernel parameters by reconstructing the original discrete-time series:

[0020] Where CE represents the cross-entropy loss function, For the original discrete data, The reconstructed discrete data is used to optimize the Gaussian kernel parameters in the original network by utilizing the cross-entropy loss function, ultimately indirectly obtaining unified continuous spatial data.

[0021] Furthermore, the step of inputting unified continuous spatial data into the anomaly tracking algorithm to obtain the root cause anomaly points causing industrial system failures specifically includes: A two-dimensional convolutional neural network with a kernel size of 1×n is used to extract independent temporal features from uniform, continuous spatial data by sliding the convolutional kernel in the time dimension.

[0022] in: — Convolution operation; — Convolution kernel parameter matrix; — Bias vector, — Sensor data input in the i-th dimension; the prediction objective function is:

[0023] The hidden layer variable matrix in the neural network is obtained through multiple iterations. , This is the parameter matrix of the neural network. For the actual value of the next timestamp, For the predicted value at the next timestamp, the prediction information of each variable is passed to other variables according to the corresponding prediction contribution rate. Accordingly, the following equation applies:

[0024] in, ——No. The scores of each variable, Given the causal information coefficient, under this mechanism, the following equation is established:

[0025] Equivalent to:

[0026] in, The score vector consists of the scores of all variables. It is a multi-level causal matrix; Quantification of outlier test point scores into a matrix The eigenvalue decomposition problem is solved by using an outlier score ranking algorithm to give a score to each variable. The variable with the highest score is identified as the root outlier.

[0027] A hybrid timing-based dynamic system anomaly identification device includes: Data acquisition module: used to collect raw, multi-source sampling data from the industrial site; The first processing module is used to normalize the raw multi-source sampling data to obtain normalized data. The second processing module is used to perform adaptive Gaussian kernel embedding processing on normalized data, transforming the mixed time series with continuous and discrete time series into a unified form with continuous time series, thus obtaining unified continuous spatial data. Identification module: Used to input unified continuous spatial data into the abnormal measurement point tracking algorithm to obtain the root cause abnormal measurement points that cause industrial system failures.

[0028] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for identifying anomalies in a dynamic system based on hybrid timing.

[0029] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for identifying anomalies in a dynamic system based on hybrid timing.

[0030] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a solution for anomaly identification in hybrid time-series-based power systems by combining an adaptive Gaussian kernel embedding algorithm for continuous discrete-time data with an anomaly tracking algorithm. First, the invention performs effective parameter filtering and first / last timestamp alignment on unaligned data with missing channels, facilitating standardized data utilization. Second, the adaptive Gaussian kernel embedding algorithm converts discrete data into continuous time series, enabling subsequent algorithms to process data within a unified continuous value space. Finally, the invention performs convolution operations on multi-source time series data, helping to accurately capture the influence paths between sensors and quantify the importance of anomaly points using an anomaly ranking algorithm. This invention effectively improves the accuracy of anomaly identification in hybrid time-series industrial power systems under multi-source physical quantity interference, providing reliable technical support for predictive maintenance and possessing significant engineering application value. Attached Figure Description

[0031] The accompanying drawings are provided to further illustrate the invention and constitute a part of this invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0032] Figure 1 This is a schematic diagram for anomaly identification in the mixed timing of a power system. Figure 2 The original data for the parameters pitch angle x, pitch angle y, and pitch angle z; Figure 3 These are the normalized values ​​of the parameters pitch angle x, pitch angle y, and pitch angle z. Figure 4 This is a schematic diagram of the original discrete data; Figure 5 A schematic diagram of multiple Gaussian curves embedded through a Gaussian kernel; Figure 6 shows the continuous signal after Gaussian kernel weighted fusion; Figure 7 shows the pitch angle x single-pass timing data before alignment and downsampling; Figure 8 shows the pitch angle x single-pass timing data after alignment and downsampling; Figure 9 shows the degree of impact of abnormal measuring points on the fault. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] Example 1 An anomaly identification method for dynamic systems based on hybrid time series is proposed. First, data normalization is performed by removing irrelevant channels and aligning the first and last timestamps. Second, an adaptive Gaussian kernel embedding method is used to achieve continuous processing of discrete time series. Finally, an anomaly tracking model is constructed to highlight the correlation information between sensors in the fault propagation chain. An anomaly quantification algorithm is combined to comprehensively consider the degree of sensor contribution to the fault, such as... Figure 1 As shown, the specific steps include: Step 1: Collect raw multi-source sampling data from the industrial site; Step 2: Normalize the original multi-source sampling data to obtain normalized data; Step 3: Perform adaptive Gaussian kernel embedding on the normalized data to transform the mixed time series with continuous and discrete time series into a unified form with continuous time series, thus obtaining unified continuous spatial data; Step 4: Input the unified continuous spatial data into the abnormal measurement point tracking algorithm to obtain the root cause abnormal measurement points that cause industrial system failures.

[0036] Example 2 This invention unifies and continuousizes multi-source mixed time series containing both continuous and discrete time series, enabling subsequent anomaly tracking algorithms to operate within a unified continuous value space. It employs a multi-level convolutional neural network to highlight the correlation between sensors and utilizes an anomaly tracking algorithm to pinpoint the original location of the measurement points when anomalies occur in industrial processes. This method fully leverages the useful information contained in both continuous and discrete time series to track the location of measurement points at the time of anomalies, facilitating post-anomaly maintenance.

[0037] Specifically, the present invention provides a method for anomaly identification of a power system based on industrial hybrid timing, comprising the following steps: 1) Due to inconsistencies in data length caused by inconsistent sensor sampling frequencies, misaligned timestamps, and missing sensor channels in the multi-source data collected from industrial sites, the raw multi-source sampling data is first standardized. The specific process is as follows: First, the sensor dimensions of the original multi-source sampling data are filtered based on the monitoring parameter screening method according to the proportion of effective parameters. The calculation formula is as follows:

[0038]

[0039]

[0040] In the formula: ——No. The percentage of each monitoring parameter in the entire process; —Total number of processes; ——No. The next process Sampling time series under each monitoring parameter; —A set of effective monitoring parameters; ——No. One monitoring parameter; —The set effective parameter percentage threshold; wherein the above-mentioned , , J This represents the total number of sensors.

[0041] Subsequently, a backfilling method was used to align the timestamps of the remaining sensor channel data, with 0s as the unified start timestamp and the last timestamp of the shortest flow as the unified end timestamp.

[0042] In the formula: —Resampled time series Timestamp of each sampling point; —Resampled time series Data values ​​of each sampling point; ——Original time series Timestamp of each sampling point; ——Original time series The data values ​​of each sampling point; where the above-mentioned data are involved. , , The number of sampling points in the original time series. This represents the number of sampling points in the resampled time series.

[0043] Finally, standardize the data by expanding the monitoring parameter dimensions:

[0044] In the formula x ′ is the standardized value. μ and σ These are the mean and standard deviation of this parameter for all processes, respectively.

[0045] 2) The input data is processed by adaptive Gaussian kernel embedding to transform the mixed time series with continuous and discrete time series into a unified continuous time series form. The specific steps are as follows: For discrete variables Each non-zero excitation time point With that point as the center, the superimposed bandwidth is Gaussian kernels are used to generate continuous time series:

[0046] Integrating the Gaussian kernels at all non-zero excitation time points yields the initial continuous variables:

[0047] in This is a set of discrete time points. To capture multi-scale time patterns, a design is implemented. Gaussian kernels with different bandwidths, whose bandwidth parameters are generated according to an arithmetic sequence:

[0048] in and These are learnable parameters. The weight coefficients of each Gaussian kernel. Normalization via Softmax:

[0049] in, The weighting coefficients of the Gaussian kernel. The weight for each timestamp is M, where M is the number of Gaussian kernels with different bandwidths. Finally, the continuous variables are represented as a weighted summation result:

[0050] in, Gaussian curves at each timestamp, For time variables, This represents the variance of the Gaussian curve. This is random noise disturbance. This represents the variance of the random noise. Optimize Gaussian kernel parameters by reconstructing the original discrete-time series:

[0051] In the formula, CE represents the cross-entropy loss function. For the original discrete data, The reconstructed discrete data is used to optimize the Gaussian kernel parameters in the original network by utilizing the cross-entropy loss function, ultimately indirectly obtaining unified continuous spatial data.

[0052] 3) The obtained unified continuous spatial data is fed into the anomaly tracking algorithm to obtain the root cause measurement points that cause industrial system failures. The specific steps are as follows: A two-dimensional convolutional neural network with a kernel size of (1×n) is used to extract independent temporal features from uniform continuous spatial data by sliding the convolutional kernel in the time dimension.

[0053] In the formula: — Convolution operation; — Convolution kernel parameter matrix; — Bias vector; — Sensor data input in the i-th dimension; the prediction objective function is:

[0054] The hidden layer variable matrix in the neural network is obtained through multiple iterations. , This is the parameter matrix of the neural network. For the actual value of the next timestamp, For the predicted value at the next timestamp, the predicted information of each variable is passed to other variables according to its corresponding predictive contribution rate. Accordingly, the following equation holds:

[0055] In the formula: ——No. The scores of each variable, Given the causal information coefficient, under this mechanism, the following equation is established:

[0056] This is equivalent to:

[0057] in, The score vector consists of the scores of all variables. Multi-level causal matrix Therefore, the quantization of outlier score points is transformed into a matrix. The problem involves eigenvalue decomposition. An outlier score ranking algorithm assigns a score to each variable, and the variable with the highest score is identified as the root cause outlier, thus providing a clear tracking result of the importance of each outlier.

[0058] Example 3 The data processing workflow first retrieves all batch number information for the industrial process from the edge data center server of the industrial power system, and then extracts a list of power system models based on the query results. For each power system model, the system retrieves all associated batch numbers, obtains the raw data by batch, and concatenates and integrates all batch data under the same power system model, storing them locally according to the power system model classification.

[0059] Following this, a parameter simplification phase is performed. The system filters channels based on parameter mask values ​​provided by the data center (0 indicates missing data, and 1 indicates complete data). This invention sets the parameter missingness threshold to 0.9, meaning that parameter channels with a missingness exceeding 90% in the entire batch are deleted. Common channels present in all power system models are then selected, and invalid parameters are removed for each power system model, ultimately retaining data from 11 valid channels.

[0060] To address data alignment issues caused by inconsistent sensor sampling frequencies, the system employs a backfill resampling method. Specific steps include: uniformly setting the start timestamp of all parameters to 0; using the end timestamp of the parameter with the shortest execution time as the termination benchmark; and resampling based on the highest sampling frequency to ensure all parameters have the same data length and time series consistency. The processed standardized data is stored categorized by power system model, providing a unified data foundation for subsequent analysis.

[0061] Because the data volume is too large after resampling, the system uses the maximum triangular bucket downsampling algorithm for optimization, which reduces the data volume while retaining key outliers, local minor fluctuations, and overall trend features, such as... Figure 7 and Figure 8 As shown, data normalization is then performed, as follows: Figure 2 and Figure 3 As shown. These operations effectively improve the computational efficiency of subsequent algorithms and provide more lightweight data support for edge deployments.

[0062] Then, the discrete data is made continuous, i.e., an adaptive Gaussian kernel embedding operation is performed, such as... Figure 4 , Figure 5 and Figure 6 As shown, after obtaining unified multi-source continuous values, the fault data is fed into the anomaly measurement point tracking algorithm to obtain the contribution of each measurement point to the fault, such as... Figure 9 As shown, the flow measurement points are the final abnormal measurement points.

[0063] This invention belongs to the field of industrial power system fault diagnosis and reliability engineering technology, and proposes an anomaly identification method for industrial systems based on multi-source time-series data fusion. Addressing the technical challenge of accurately locating mixed time-series anomaly measurement points in industrial systems under multi-physics coupled conditions, this invention acquires a multi-source monitoring dataset of the target power system under fault conditions using an industrial data acquisition system and designs a standardized data processing flow. First, channel filtering is performed based on parameter mask values ​​to remove invalid parameters with excessively high missing rates. Second, a backfilling resampling method is used to achieve time alignment of the multi-source data. Then, the maximum triangular bucket downsampling algorithm (LTTB) is applied to optimize the resampled data, effectively reducing the data dimensionality while ensuring the integrity of data features, resulting in a regularized multi-source time-series dataset. For the complex situation of overlapping continuous and discrete time series in the multi-source time-series data, a Gaussian kernel embedding algorithm is used to map the discrete time series to a continuous value space, achieving a unified representation of the multi-source data. Finally, the processed multi-source continuous data is input into an anomaly measurement point tracking algorithm, and the precise location of anomaly measurement points is achieved by calculating the contribution of each measurement point to the system fault. This method solves the technical problem of tracking abnormal measurement points under multi-physics coupling conditions. The proposed data processing method can effectively cope with the complexity and incompleteness of industrial data. The algorithm is simple to implement and has high computational efficiency, which significantly improves the accuracy of fault diagnosis in industrial systems and provides reliable technical support for post-maintenance of power systems. It has important engineering application value.

[0064] Example 4 A hybrid timing-based dynamic system anomaly identification device includes: Data acquisition module: used to collect raw, multi-source sampling data from the industrial site; The first processing module is used to normalize the raw multi-source sampling data to obtain normalized data. The second processing module is used to perform adaptive Gaussian kernel embedding on normalized data, transforming the mixed time series with continuous and discrete time series into a unified form with continuous time series, thus obtaining unified continuous spatial data. Identification module: Used to input unified continuous spatial data into the abnormal measurement point tracking algorithm to obtain the root cause abnormal measurement points that cause industrial system failures.

[0065] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0067] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

Claims

1. A method for anomaly identification in a dynamic system based on hybrid time series, characterized in that, Includes the following steps: Step 1: Collect raw multi-source sampling data from the industrial site; Step 2: Normalize the original multi-source sampling data to obtain normalized data, specifically as follows: The monitoring parameter screening method based on the proportion of effective parameters filters the sensor dimensions of the original multi-source sampling data; The backfilling method is used to align the timestamps of the remaining sensor channel data, with 0s as the unified start timestamp and the last timestamp of the shortest process as the unified end timestamp. Finally, the data is standardized after being expanded according to the monitoring parameter dimensions; Step 3: Perform adaptive Gaussian kernel embedding on the normalized data to transform the mixed time series with continuous and discrete time series into a unified continuous time series form, resulting in unified continuous spatial data; specifically including: For discrete variables Each non-zero excitation time point With the non-zero excitation time point as the center, the bandwidth is superimposed. Gaussian kernels are used to generate continuous time series: Integrating the Gaussian kernels at all non-zero excitation time points yields the initial continuous variables: in, Given a set of discrete time points, where t is the time dimension variable, and designed to capture multi-scale time patterns, Gaussian kernels with different bandwidths, whose bandwidth parameters are generated according to an arithmetic sequence: in, and The weight coefficients of each Gaussian kernel are learnable parameters. Normalization via Softmax: in, The weighting coefficients of the Gaussian kernel. The weight for each timestamp is M, where M is the number of Gaussian kernels with different bandwidths. The initial continuous variables are represented as a weighted superposition result; in, For random noise disturbance, This represents the variance of the random noise. Optimize Gaussian kernel parameters by reconstructing the original discrete-time series: Where CE represents the cross-entropy loss function, For the original discrete data, The reconstructed discrete data is used to optimize the Gaussian kernel parameters in the original network by utilizing the cross-entropy loss function, ultimately indirectly obtaining unified continuous spatial data. Step 4: Input the unified continuous spatial data into the anomaly tracking algorithm to obtain the root cause anomaly measurement points that cause industrial system failures, specifically including: A two-dimensional convolutional neural network with a kernel size of 1×n is used to extract independent temporal features from uniform, continuous spatial data by sliding the convolutional kernel in the time dimension. in: — Convolution operation; — Convolution kernel parameter matrix; — Bias vector, — Sensor data input in the i-th dimension; the prediction objective function is: The hidden layer variable matrix in the neural network is obtained through multiple iterations. , This is the parameter matrix of the neural network. For the actual value of the next timestamp, For the predicted value at the next timestamp, the prediction information of each variable is passed to other variables according to the corresponding prediction contribution rate. Accordingly, the following equation applies: in, ——No. The scores of each variable, For causal information coefficient, , J Given the total number of sensors, the following equation can be established under this mechanism: Equivalent to: in, The score vector consists of the scores of all variables. It is a multi-level causal matrix; Quantification of outlier test point scores into a matrix The eigenvalue decomposition problem is solved by using an outlier score ranking algorithm to give a score to each variable. The variable with the highest score is identified as the root outlier.

2. The method for anomaly identification of a dynamic system based on hybrid time series as described in claim 1, characterized in that, The monitoring parameter screening method based on the proportion of effective parameters filters the sensor dimensions of the original multi-source sampling data. The calculation formula is as follows: In the formula: ——No. The percentage of each monitoring parameter in the entire process; —Total number of processes; ——No. The first process Sampling time series under each monitoring parameter; —A set of effective monitoring parameters; ——No. One monitoring parameter; —The set threshold for the percentage of effective parameters; , , J This represents the total number of sensors.

3. The method for anomaly identification of a dynamic system based on hybrid time series as described in claim 1, characterized in that, The backfilling method is used to align the timestamps of the remaining sensor channel data, with 0s as the unified start timestamp and the last timestamp of the shortest flow as the unified end timestamp. Specifically: In the formula: —Resampled time series Timestamp of each sampling point; —Resampled time series Data values ​​of each sampling point; ——Original time series Timestamp of each sampling point; ——Original time series The data values ​​of each sampling point; among which, , , The number of sampling points in the original time series. This represents the number of sampling points in the resampled time series.

4. A dynamic system anomaly identification device based on hybrid timing, used to implement the dynamic system anomaly identification method based on hybrid timing as described in claim 1, characterized in that, include: Data acquisition module: used to collect raw, multi-source sampling data from the industrial site; The first processing module is used to normalize the raw multi-source sampling data to obtain normalized data. The second processing module is used to perform adaptive Gaussian kernel embedding on normalized data, transforming the mixed time series with continuous and discrete time series into a unified form with continuous time series, thus obtaining unified continuous spatial data. Identification module: Used to input unified continuous spatial data into the abnormal measurement point tracking algorithm to obtain the root cause abnormal measurement points that cause industrial system failures.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for identifying anomalies in a dynamic system based on hybrid timing as described in any one of claims 1 to 3.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the dynamic system anomaly identification method based on hybrid timing as described in any one of claims 1 to 3.

Images