SCADA data correction method fusing correlation constraints and multi-frequency modal calibration

Abstract

The application provides a SCADA data correction method combining association constraints and multi-frequency modal calibration, comprising: constructing an association rule set based on historical monitoring data of a SCADA system, and using association constraint relationships to define the feasible region boundaries of monitoring variables under various operating conditions; using a dynamically adjustable variational modal decomposition technology to decompose the historical monitoring data in continuous time into a low-frequency modal of long-term evolution of the system and a high-frequency modal affected by weather for a short time; using a gated recurrent unit to extract steady-state features from the low-frequency modal and transient-state features from the high-frequency modal, and constructing a deviation prediction model through feature weighted fusion; and correcting abnormal data based on the feasible region boundaries and confidence intervals. The application can improve the correction accuracy and governance reliability of abnormal data of the SCADA system by constructing a dual verification mechanism combining physical interpretation and data-driven, and then introducing a fuzzy confidence evaluation and hierarchical processing strategy.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, and specifically to a SCADA data correction method that integrates correlation constraints and multi-frequency modal calibration. Background Technology

[0002] SCADA systems are the core data hub of power generation-side industrial control systems. However, the trend towards openness, networking, and standardization in industrial control systems exposes them to increasingly severe cybersecurity threats. After a cyberattack or virus intrusion, attackers typically tamper with critical monitoring data in the SCADA system, such as artificially raising or lowering operating parameters like voltage, current, power, and frequency. This induces operators or automatic control devices to make incorrect judgments and actions, potentially leading to grid instability, equipment damage, or even widespread power outages. Current technologies, such as threshold checks, parity checks, and outlier detection based on simple statistical models, can only identify outliers that significantly exceed the hardware's measurement range. They are almost incapable of identifying and correcting carefully crafted, hidden tampered data that remains within the normal operating condition range. Furthermore, the data correction process is often a black-box operation. Without a quantitative evaluation mechanism for the reliability of the correction results, low-quality corrected data may be blindly adopted and written into historical databases, contaminating the original data assets and exacerbating errors in subsequent analysis and decision-making. Summary of the Invention

[0003] The technical problem to be solved by this invention is to overcome the shortcomings of existing technologies in correcting abnormal data in SCADA systems, and thus provide a SCADA data correction method that integrates correlation constraints and multi-frequency modal calibration. This invention improves the accuracy of correcting abnormal data in SCADA systems and the reliability of governance by constructing a dual verification mechanism that combines physical interpretability and data-driven approach, and by introducing fuzzy reliability assessment and hierarchical processing strategies.

[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a SCADA data correction method that integrates correlation constraints and multi-frequency modal calibration, comprising the following steps: constructing a set of correlation rules based on historical monitoring data of the SCADA system, and using correlation constraint relationships to define the feasible domain boundaries of monitoring variables under various operating conditions; Using dynamically adjustable variational mode decomposition technology, historical monitoring data over continuous time is decomposed into low-frequency modes of long-term system evolution and high-frequency modes affected by short-term weather conditions. Gated cyclic units are used to extract steady-state features from the low-frequency modes and transient features from the high-frequency modes. A bias prediction model is constructed through feature weighted fusion, and the confidence intervals of each monitoring variable under weather disturbances are estimated using time-series evolution patterns. Erroneous data are corrected based on feasible domain boundaries and confidence intervals, and the reliability of the data correction results is assessed using a fuzzy evaluation method combining joint information entropy and KL divergence. Based on the reliability level of the data correction, tiered measures are implemented for SCADA system erroneous data: automatic updating, manual review followed by updating, and rejection of updates with alarm triggering.

[0005] Optionally, an association rule set is constructed based on historical monitoring data from the SCADA system. The feasible domain boundaries of the monitoring variables under various operating conditions are defined using association constraints. This includes: treating each monitoring variable in the power generation-side SCADA system as an itemset, and discretizing the monitoring data value status using a sliding window; using a frequent pattern mining algorithm to extract frequent itemsets of each monitoring variable value under normal operating conditions based on a preset support threshold; selecting high-confidence combinations from the frequent itemsets to generate an association rule set with physical interpretability, and then redundancy merging of the association rule set; and determining the feasible domain boundaries of each monitoring variable under different operating conditions based on the association rule set.

[0006] Optionally, a dynamically adjustable variational mode decomposition technique is used to decompose continuous historical monitoring data into low-frequency modes of long-term system evolution and high-frequency modes affected by short-term weather conditions. This includes: adaptively decomposing the monitoring data using variational mode decomposition technology, decomposing the original sequence into a preset number of mode functions and residual terms, wherein the preset number of mode functions is dynamically adjusted according to the time-varying intensity of weather factors such as wind and sunlight; performing spectral analysis on each mode function to obtain the center frequency and bandwidth of different mode terms; and classifying low-frequency and high-frequency modes according to frequency distribution characteristics, wherein the low-frequency modes correspond to the long-term trend terms of the industrial control system itself, and the high-frequency modes correspond to short-term disturbance terms affected by weather conditions.

[0007] Optionally, a gated recurrent unit is used to extract steady-state features from low-frequency modes and transient features from high-frequency modes. A bias prediction model is constructed through feature weighted fusion, and the confidence intervals of each monitored variable under weather disturbances are estimated using the temporal evolution law. This includes: taking the decomposed mode sequence as input, constructing feature extraction branches based on gated recurrent units, where the high-frequency mode branch captures transient features and the low-frequency mode branch extracts steady-state features; introducing an attention mechanism to calculate the contribution weights of the features of each mode branch, fusing the weighted features to obtain a multi-scale temporal state vector; training the bias prediction model using the temporal state vector, and outputting the confidence intervals of the monitored variables under weather disturbances.

[0008] Optionally, erroneous data is corrected based on feasible region boundaries and confidence intervals. A fuzzy evaluation method combining joint information entropy and KL divergence is used to assess the reliability of the data correction results. This includes: defining the data value range based on feasible region boundaries; obtaining measurement deviation calibration based on confidence intervals; correcting abnormal data based on the value range and deviation calibration; calculating the information entropy between the corrected data and reference data; evaluating the entropy change index in the time dimension to measure the degree of information loss before and after data correction; calculating the KL divergence of global monitoring data to measure the degree of information distribution shift before and after data correction; and using a fuzzy comprehensive evaluation method to map multidimensional indicators to fuzzy evaluation levels of high reliability, medium reliability, and low reliability, outputting a reliability score for the correction of abnormal data.

[0009] Optionally, based on the confidence level of the data correction, tiered measures are implemented for erroneous data in the SCADA system, including automatic updating, manual review followed by updating, and rejection of updates with alarm triggering. Specifically: if the correction operation is of high confidence, the corrected data is directly used to update the abnormal data in the SCADA system, and the historical data records of the corresponding monitoring variables are updated synchronously; if the correction operation is of medium confidence, the corrected data is submitted to maintenance personnel for manual review, and the abnormal data update is performed in the SCADA system only after manual confirmation; if the correction operation is of low confidence, the corrected data is not adopted, and a data anomaly alarm is triggered in the SCADA system, prompting maintenance personnel to conduct troubleshooting and data analysis.

[0010] Secondly, embodiments of the present invention provide a SCADA data correction system that integrates correlation constraints and multi-frequency modal calibration, comprising: a correlation analysis module for constructing a set of correlation rules and defining the feasible domain boundary of the monitoring variables using correlation constraint relationships; a modal decomposition module for using variational modal decomposition technology to decompose the continuous-time monitoring data into low-frequency modes and high-frequency modes; a deviation prediction module for constructing a deviation prediction model and estimating the confidence interval of the monitoring variables using time-series evolution laws; a correction evaluation module for using fuzzy evaluation to evaluate the credibility of the data correction results after correcting the abnormal data; and a data correction module for implementing graded measures for abnormal data in the SCADA system according to the credibility level of the data correction.

[0011] Thirdly, embodiments of the present invention provide a computer device, including: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described in the first aspect, or any optional embodiment of the first aspect.

[0012] One technical advantage of this invention is that: This invention improves the accuracy of correcting abnormal data in SCADA systems and enhances the reliability of governance by constructing a dual verification mechanism that combines physical interpretability and data-driven approaches, and by introducing fuzzy reliability assessment and hierarchical processing strategies. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating a SCADA data correction method that integrates correlation constraints and multi-frequency modal calibration, as provided in an embodiment of the present invention. Detailed Implementation

[0014] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present application.

[0015] The embodiments of this application will now be described in detail. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this 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.

[0016] This invention provides a SCADA data correction method that integrates correlation constraints and multi-frequency modal calibration, such as... Figure 1 As shown. The method specifically includes the following steps: Step S1: Construct a set of association rules based on historical monitoring data from the SCADA system, and use the association constraints to define the feasible domain boundaries of monitoring variables under various operating conditions.

[0017] In this embodiment, taking a new energy power generation system as an example, each monitoring variable in the SCADA system is regarded as an itemset. A sliding window is used to discretize the monitoring data values. The monitoring variables involve key variables such as wind turbine monitoring, photovoltaic monitoring, transformer substation monitoring, and booster station monitoring. The FP-Growth algorithm is used to extract frequent itemsets of each monitoring variable under normal operating conditions. During the frequent itemset mining process, the support of transaction item I in transaction set D needs to be evaluated sequentially, defined as... High-confidence combinations are selected from the frequent itemset to generate a set of association rules with physical interpretability. Then, the association rule set is redundantly merged, and a rule conflict detection mechanism is introduced during the merging process. If a new rule highly overlaps with existing rules in sample coverage, the rule with higher confidence is retained; if a new rule covers new samples not covered by existing rules, it is directly added to the rule set. Finally, the feasible region boundaries of each monitored variable under different operating conditions are determined based on the association rule set. The feasible region is defined as... ,in, and They represent the first The constraint boundaries of each variable in the associated constraint space.

[0018] Step S2: Using a dynamically adjustable variational mode decomposition technique, the historical monitoring data over continuous time is decomposed into low-frequency modes of long-term system evolution and high-frequency modes that are temporarily affected by weather.

[0019] In this embodiment, variational mode decomposition (VMD) is used to adaptively decompose the monitoring data, decomposing the original sequence into a preset number of mode functions and residual terms. Assuming the time series of a certain monitoring data is... The observed values ​​of this variable are affected by weather fluctuations: Variational mode decomposition technique is used to analyze... After adaptive decomposition, the following is obtained: ,in The intrinsic mode functions are defined for different frequency bands; spectral analysis is performed on each mode function to obtain the center frequency and bandwidth of different mode terms; low-frequency modes and high-frequency modes are divided according to frequency distribution characteristics, wherein the low-frequency modes correspond to the long-term trend terms of the industrial control system itself, and the high-frequency modes correspond to the short-term disturbance terms affected by weather.

[0020] Step S3: Use gated cyclic units to extract steady-state features from low-frequency modes and transient features from high-frequency modes. Construct a bias prediction model through feature weighted fusion and use the time-series evolution law to estimate the confidence intervals of each monitored variable under weather factor disturbances.

[0021] In this embodiment, the decomposed modal sequence is used as input to construct feature extraction branches based on gated recurrent units. The gated recurrent unit structure can capture long-term dependencies and nonlinear evolution patterns in the modal sequence. Simultaneously, environmental variables are used as conditional inputs to achieve unified modeling driven by both physical and data. The high-frequency modal branch captures transient features, while the low-frequency modal branch extracts steady-state features. An attention mechanism is introduced to calculate the contribution weights of the features in each modal branch. After fusing the weighted features, a multi-scale temporal state vector is obtained. For each modal term... Its state-space equation is defined as: The high-dimensional state vector of the system is obtained by weighted aggregation of the modal terms: The bias prediction model is trained using state vectors, and the confidence intervals of the monitored variables under weather disturbances are output. The prediction equation is as follows: Through observation values Compared with the predicted value To calculate the predicted residual .

[0022] Step S4: Correct the erroneous data based on the feasible region boundary and confidence interval, and use the fuzzy evaluation method of joint information entropy and KL divergence to evaluate the credibility of the data correction results.

[0023] In this embodiment, the data value range is defined according to the feasible region boundary, and the measurement deviation calibration is obtained according to the confidence interval. Abnormal data is corrected based on the value range and the deviation calibration. The information entropy between the corrected data Y and the reference data X is calculated. The smaller the information entropy value, the lower the joint uncertainty of the two variables, that is, the higher the dependence on the original information during the correction process. It is defined as: The entropy change index over time is evaluated to measure the degree of information loss before and after data correction; the KL divergence of the global monitoring data is calculated to measure the degree of information distribution shift before and after data correction. A larger KL divergence value indicates a more significant information distribution shift caused by the correction, defined as: The fuzzy comprehensive evaluation method is used to map multidimensional indicators into fuzzy evaluation levels of high credibility, medium credibility, and low credibility, and output the credibility score for correcting abnormal data.

[0024] Step S5: Based on the confidence level of the data correction, implement tiered measures for erroneous data in the SCADA system, including automatic updates, updates after manual review, and rejection of updates with alarm triggering.

[0025] In this embodiment, if the correction operation is of high confidence, the corrected data is directly used to update the abnormal data in the SCADA system, and the historical data records of the corresponding monitoring variables are updated synchronously. If the correction operation is of medium confidence, the corrected data is submitted to the operation and maintenance personnel for manual review. After manual confirmation, the abnormal data update is performed in the SCADA system. If the correction operation is of low confidence, the corrected data is not adopted, and the data anomaly alarm of the SCADA system is triggered. The operation and maintenance personnel then carry out fault investigation and data analysis.

[0026] This invention provides a SCADA data correction method that integrates correlation constraints and multi-frequency modal calibration. By constructing a dual verification mechanism that combines physical interpretability and data-driven approaches, and by introducing fuzzy reliability assessment and hierarchical processing strategies, the method can improve the accuracy of correction and the reliability of governance of abnormal data in SCADA systems.

[0027] This invention provides a SCADA data correction system that integrates correlation constraints and multi-frequency modal calibration. The system includes: a correlation analysis module for constructing a set of correlation rules and defining the feasible domain boundary of the monitoring variables using correlation constraint relationships; a modal decomposition module for using variational modal decomposition technology to decompose the continuous-time monitoring data into low-frequency and high-frequency modes; a deviation prediction module for constructing a deviation prediction model and estimating the confidence interval of the monitoring variables using time-series evolution laws; a correction evaluation module for using fuzzy evaluation to evaluate the reliability of the data correction results after correcting the abnormal data; and a data correction module for implementing graded measures for abnormal data in the SCADA system according to the reliability level of the data correction.

[0028] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A SCADA data correction method integrating correlation constraints and multi-frequency modal calibration, characterized in that, Includes the following steps: Based on historical monitoring data from the SCADA system, a set of association rules is constructed, and the feasible domain boundaries of monitoring variables under various operating conditions are defined using association constraints. Using dynamically adjustable variational mode decomposition technology, historical monitoring data over continuous time is decomposed into low-frequency modes of long-term system evolution and high-frequency modes that are temporarily affected by weather. By using gated cyclic units to extract steady-state features from low-frequency modes and transient features from high-frequency modes, a bias prediction model is constructed through feature weighted fusion. The confidence intervals of each monitored variable under weather disturbances are estimated by using the temporal evolution law. Abnormal data are corrected based on feasible region boundaries and confidence intervals, and the fuzzy evaluation method of joint information entropy and KL divergence is used to evaluate the credibility of the data correction results. Based on the credibility level of the data correction, tiered measures are implemented for abnormal data in the SCADA system, including automatic updates, updates after manual review, and rejection of updates with alarm triggering.

2. The SCADA data correction method integrating correlation constraints and multi-frequency modal calibration according to claim 1, characterized in that, A set of association rules is constructed based on historical monitoring data from the SCADA system. Association constraints are used to define the feasible domain boundaries of monitoring variables under various operating conditions, including: Each monitoring variable in the power generation-side SCADA system is treated as an itemset, and the monitoring data values ​​are discretized using a sliding window. A frequent pattern mining algorithm is used to extract frequent itemsets of the values ​​of each monitoring variable under normal operating conditions based on a preset support threshold. High-confidence combinations are selected from the frequent itemset to generate a set of association rules with physical interpretability, and then the association rule set is redundantly merged. The feasible domain boundaries of each monitored variable under different operating conditions are determined based on the association rule set.

3. The SCADA data correction method integrating correlation constraints and multi-frequency modal calibration according to claim 1, characterized in that, Using dynamically adjustable variational mode decomposition technology, continuous-time historical monitoring data is decomposed into low-frequency modes representing the long-term evolution of the system and high-frequency modes affected by short-term weather conditions, including: Variational mode decomposition technology is used to adaptively decompose the monitoring data, decomposing the original sequence into a preset number of mode functions and residual terms. The preset number of mode functions is dynamically adjusted according to the time-varying intensity of weather factors such as wind force and sunlight. Spectral analysis was performed on each modal function to obtain the center frequency and bandwidth of different modal terms; Based on the frequency distribution characteristics, low-frequency modes and high-frequency modes are divided. The low-frequency modes correspond to the long-term trend terms of the industrial control system itself, while the high-frequency modes correspond to the short-term disturbance terms affected by weather.

4. The SCADA data correction method integrating correlation constraints and multi-frequency modal calibration according to claim 1, characterized in that, Gated recurrent units are used to extract steady-state features from low-frequency modes and transient features from high-frequency modes. A bias prediction model is constructed through feature weighted fusion. The confidence intervals of each monitored variable under weather disturbances are estimated using time-series evolution patterns, including: The decomposed modal sequences are used as input to construct feature extraction branches based on gated recurrent units, where the high-frequency modal branch captures transient features and the low-frequency modal branch extracts steady-state features. An attention mechanism is introduced to calculate the contribution weights of features in each modality branch, and the weighted features are fused to obtain a multi-scale temporal state vector. A bias prediction model is trained using time-series state vectors, and the confidence intervals of the monitored variables under weather disturbances are output.

5. The SCADA data correction method integrating correlation constraints and multi-frequency modal calibration according to claim 1, characterized in that, Outlier data is corrected based on feasible region boundaries and confidence intervals. A fuzzy evaluation method using joint information entropy and KL divergence is employed to assess the reliability of the corrected data, including: The range of data values ​​is defined based on the feasible domain boundary, the measurement deviation calibration is obtained based on the confidence interval, and the abnormal data is corrected based on the range of values ​​and the deviation calibration. Calculate the information entropy between the corrected data and the reference data, evaluate the entropy change index in the time dimension, and measure the degree of information loss before and after data correction; Calculate the KL divergence of global monitoring data to measure the degree of information distribution shift before and after data correction; The fuzzy comprehensive evaluation method is used to map multidimensional indicators into fuzzy evaluation levels of high credibility, medium credibility, and low credibility, and output the credibility score for correcting abnormal data.

6. The SCADA data correction method integrating correlation constraints and multi-frequency modal calibration according to claim 1, characterized in that, Based on the reliability level of the data correction, tiered measures are implemented for abnormal data in the SCADA system, including automatic updates, updates after manual review, and rejection of updates with alarm triggering. If the correction operation is of high confidence, the correction data will be used directly to update the abnormal data in the SCADA system, and the historical data records of the corresponding monitoring variables will be updated synchronously. If the correction operation is considered medium trustworthy, the corrected data will be submitted to the operations and maintenance personnel for manual review. After manual confirmation, the abnormal data update will be performed in the SCADA system. If the correction operation is deemed low-reliability, the corrected data will not be adopted, and a data anomaly alarm will be triggered in the SCADA system, prompting maintenance personnel to conduct troubleshooting and data analysis.

7. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the SCADA data correction method of fusion correlation constraints and multi-frequency modal calibration as described in any one of claims 1-6.

Images