Online distributed security monitoring method for power system and storage medium

Abstract

The invention relates to an online distributed security monitoring method for a power system and a storage medium, which can independently execute multi-dimensional time sequence modeling, influence point selection, online streaming processing and model updating on each sensor node so as to realize real-time monitoring and anomaly detection of the system. Through distributed calculation and a neighbor diffusion strategy, nodes cooperatively generate a global model coefficient matrix, and in combination with consistency hypothesis testing, abnormal behaviors in the system are accurately detected. The system deployment and optimization steps further ensure the accuracy, real-time performance and stability of the method. According to the method, the calculation complexity is effectively reduced, the processing efficiency in a big data environment is remarkably improved, the method is suitable for wide power system application scenes, a reliable and efficient safety monitoring solution is provided, meanwhile, the method has high expandability and robustness, and the method can adapt to complex and changeable application requirements in the future.

Description

Technical Field

[0001] The present invention relates to data monitoring and analysis technology in power systems, and in particular to a method for distributed multidimensional time series modeling, big data influence point selection, online streaming processing and model updating. The method of the present invention is mainly used for real-time system monitoring and anomaly detection in large-scale data environments, and is particularly suitable for complex multi-node distributed networks such as power systems, intelligent transportation, industrial monitoring and smart cities. Background Art

[0002] With the rapid development of power systems, more and more sensors are deployed in various physical environments to collect and transmit large amounts of time series data in real time. These data are usually multi-dimensional, with huge amounts of data and complex processing, which poses great challenges to system monitoring and anomaly detection. Traditional centralized data processing methods are often unable to cope with such a large amount of data, and are prone to delays in data transmission and calculation, resulting in less than ideal system response speed.

[0003] In the big data environment, the real-time, accuracy and stability of data become particularly important. Therefore, how to effectively perform multidimensional time series modeling, big data influence point selection and online processing has become a problem that needs to be solved urgently. Distributed computing has gradually become an effective solution because it can decompose computing tasks to multiple nodes for execution, reducing the computing burden and delay of the system. However, the influence point selection algorithm, model update mechanism and system anomaly detection method in distributed data processing are not perfect enough, and it is difficult to maintain high accuracy and real-time performance in different application scenarios. Summary of the invention

[0004] The purpose of the present invention is to propose a distributed online monitoring method for power systems, which realizes real-time monitoring and anomaly detection of the system through multi-dimensional time series modeling, big data influence point selection, online streaming processing and model updating. The method is deployed and optimized in the power system to ensure the real-time, accuracy and stability of the system. The method of the present invention is particularly suitable for monitoring needs in large-scale and complex network environments, and can effectively improve the response speed and monitoring accuracy of the system.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions: An online distributed safety monitoring method for a power system comprises the following steps: Step S1: the data initialization step: collecting and preprocessing multidimensional time series data from a sensor in the power system, performing preliminary statistical analysis, and calculating the sample covariance matrix through the initial sample to establish an initial reference standard for subsequent data analysis; Step S2: the multi-dimensional time series modeling step: constructing a vector autoregressive (VAR) model using the data initialized in step S1, estimating the model coefficient matrix by the least squares method, and optimizing the order of the model to accurately capture the temporal dependency between sensor nodes; Step S3: The big data key point selection step: Use Mahalanobis distance to evaluate the impact of the multidimensional time series data points in step S1, and set a threshold r Screen out key influencing points to ensure that the selected key influencing points can represent the global characteristics of the data and form a small-scale data set; Step S4: the online streaming data processing step: when the data stream is obtained, the sample covariance matrix and the coefficients of the autoregressive model are updated in real time, key points are dynamically adjusted and selected, and historical data are periodically integrated to ensure that the model can continuously reflect the latest data characteristics; Step S5: the distributed online monitoring step: repeating steps S1 to S4 at each sensor node until all sensors are traversed, and information is exchanged through a neighbor diffusion strategy to generate a global model coefficient matrix, and monitor the power system based on multidimensional stream data; Step S6: The consistency hypothesis test step: compare the global model coefficient matrix with the model coefficient matrix under the assumed normal state, and calculate p The system detects abnormal behavior of the value, integrates the inspection results of each node, and makes a consistent decision on whether there is a global anomaly and takes corresponding actions; Deploy and optimize the above method system: Deploy the online distributed security monitoring method, verify the effectiveness of the proposed method from multiple aspects such as accuracy, stability, and computational efficiency, continuously optimize the algorithm, and tune key parameters, such as the Mahalanobis distance threshold r and sliding window size, and verify the effectiveness of the proposed method, and continue to optimize the algorithm to improve its adaptability and stability; In the above technical solution, in step S1, the specific steps of data initialization are: Step S11: using the monitoring components in the power system to collect multi-dimensional time series data of each sensor node; Step S12: performing pre-processing operations such as cleaning, denoising and formatting on the data collected in step S11; Step S13: Performing a preliminary analysis on the data preprocessed in step S12 to determine the basic statistical characteristics and data distribution of the data; Step S14: arbitrarily select data in the previous time period of the time series as the initial sample to estimate the sample covariance matrix and provide a benchmark for the subsequent online monitoring algorithm; In the above technical solution, in step S2, the specific steps of multidimensional time series modeling are: Step S21: Based on the multidimensional time series data collected and processed in step S1, a VAR model is constructed to capture the temporal dependency between the data; Step S22: Determine the order of the VAR model p , based on the Bayesian Information Criterion (BIC), the order of the model is optimized to improve the fitting accuracy of the model; Step S23: using the least square method to estimate the coefficient matrix of the VAR model, ensuring that the coefficient matrix can accurately reflect the interdependence between the sensor nodes; Step S24: Use cross-validation to evaluate the performance of the model and improve the generalization ability of the model through parameter tuning; In the above technical solution, in step S3, the specific steps of selecting the key points of big data are: Step S31: Expand the VAR model obtained in step S2 to a nonlinear form to process more complex time series data, thereby improving the model's ability to capture nonlinear relationships; Step S32: Calculate the Mahalanobis distance of each time point in the time series data processed in step S1 to evaluate the impact of the data point on the model and select the key point set that contributes more to the model; Step S33: Based on the selected Mahalanobis distance threshold r , select the key point set SIPS , to reduce data size and improve computational efficiency; Step S34: By analyzing the key point set SIPS Re-evaluate the statistical properties of the data to ensure that it covers the main distribution interval of the data, otherwise repeat steps S31 to S33 and reselect the key point set to better represent the global characteristics; In the above technical solution, in step S4, the specific steps of online streaming data processing are: Step S41: Real-time update of the sample covariance matrix to reflect the latest data characteristics; Step S42: dynamically adjusting the calculation of the Mahalanobis distance according to the updated sample covariance matrix to ensure the accuracy of the influence point selection; Step S43: for each newly acquired data point, the model coefficient matrix is ​​updated using a single-pass algorithm to achieve online model update; Step S44: periodically integrating information from historical data to improve the robustness of the model and controlling the complexity of the model through a sliding window technique; In the above technical solution, in step S5, the specific steps of distributed online monitoring are: Step S51: Treat each sensor node as an independent computing unit and execute the key point selection algorithm on each node; Step S52: using the neighbor diffusion strategy, the local estimation result of each node is exchanged with its one-hop neighbor node to achieve global fusion of information; Step S53: recursively estimate the local model coefficients of each node using least squares, and gradually approach the global optimal solution through multiple iterations; Step S54: Summarize the local model coefficients of each node to generate a global model coefficient matrix as the overall monitoring result of the power system; In the above technical solution, in step S6, the specific steps of consistency hypothesis testing are: Step S61: Based on the global model coefficient matrix B t , calculates the hypothesis test statistic used to evaluate changes in system states H 0 : Current global model coefficient matrix B t Equal to the coefficient matrix under normal conditions B Normal ; Step S62: Calculate the statistic using the Wald test formula W , evaluate the current statistics W Is it significant? Step S63: Calculation p value, p The value represents the assumption H 0 When true, the current statistic is observed W or more extreme values; Step S64: If p If the value is significantly lower than the preset level, then reject H 0 , indicating that the system has detected significant abnormal behavior, triggering further monitoring and protection measures; Step S65: Collect and integrate the hypothesis test results of all nodes, process the asynchronous calculations between nodes through the distributed consensus algorithm, and generate a global p value; Step S66: Based on the global p The system makes consistent decisions based on the value and determines the overall system status. If an abnormality is detected in the entire system, the corresponding safety protection mechanism is activated to ensure the security and stability of the system. In the above technical solution, in step S7, the specific steps of system deployment and optimization are: Compared with the prior art, the present invention has the following beneficial effects: It has five significant benefits: real-time and high efficiency, adaptability and scalability, accuracy and robustness, reduced computational complexity, and consistent decision-making. These advantages make the present invention outstanding in monitoring and anomaly detection of power systems, especially in large-scale and complex network environments, and can achieve faster response speed, higher monitoring accuracy, and stronger system adaptability; BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram of the process of data initialization in the present invention; Figure 3 A schematic diagram of the process of multidimensional time series modeling in the present invention; Figure 4 A schematic diagram of the process of selecting key points of big data in the present invention; Figure 5 It is a schematic diagram of the online streaming data processing process in the present invention; Figure 6 This is a schematic diagram of the distributed online monitoring process in the present invention; Figure 7 It is a schematic diagram of the consistency hypothesis test process in the present invention; Figure 8 It is a schematic diagram of the system deployment and optimization process in the present invention. DETAILED DESCRIPTION

[0007] In order to better illustrate the purpose, technical solutions and advantages of the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0008] The present invention proposes an online distributed safety monitoring method and process for a power system, which relates to a method for distributed multidimensional time series modeling, big data influence point selection, online streaming processing and model updating. The flowchart is shown in FIG. Figure 1 As shown, the following steps are included: Step S1: The data initialization step is to collect and pre-process the multi-dimensional time series data of a sensor in the power system, laying the foundation for subsequent distributed processing. The specific implementation includes the following steps: Step S11: using the monitoring components in the power system to collect multi-dimensional time series data of each sensor node; Step S12: performing pre-processing operations such as cleaning, denoising and formatting on the data collected in step S11; Step S13: Performing a preliminary analysis on the data preprocessed in step S12 to determine the basic statistical characteristics and data distribution of the data; Step S14: arbitrarily select data in the previous time period of the time series as the initial sample to estimate the sample covariance matrix and provide a benchmark for the subsequent online monitoring algorithm; Step S2: Distributed multidimensional time series modeling is one of the core steps of the present invention, which aims to perform distributed processing on sensor data and generate an initial global model coefficient matrix. The specific implementation steps are as follows: Step S21: Based on the multidimensional time series data collected and processed in step S1, a VAR model is constructed to capture the temporal dependency between the data; Step S22: Determine the order of the VAR model p , based on the Bayesian information criterion, the order of the model is optimized to improve the fitting accuracy of the model; Step S23: using the least square method to estimate the coefficient matrix of the VAR model, ensuring that the coefficient matrix can accurately reflect the interdependence between the sensor nodes; Step S24: Use cross-validation to evaluate the performance of the model and improve the generalization ability of the model through parameter tuning; Step S3: In the influence point selection step, the present invention uses a distributed algorithm to select key points in big data and re-evaluate to ensure that the selected data points can represent the overall data characteristics. The specific implementation steps are as follows: Step S31: Expand the VAR model obtained in step S2 to a nonlinear form to process more complex time series data, thereby improving the model's ability to capture nonlinear relationships; Step S32: Calculate the Mahalanobis distance of each time point in the time series data processed in step S1 to evaluate the impact of the data point on the model and select the key point set that contributes more to the model; Step S33: Based on the selected Mahalanobis distance threshold r , select the key point set SIPS , to reduce data size and improve computational efficiency; Step S34: By analyzing the key point set SIPS Re-evaluate the statistical properties of the data to ensure that it covers the main distribution interval of the data, otherwise repeat steps S31 to S33 and reselect the key point set to better represent the global characteristics; Step S4: When data continues to flow in, the system needs to update the model in real time to reflect the latest data. The online streaming processing and model update steps specifically include the following operations: Step S41: When the data stream is acquired, the sample covariance matrix is ​​updated in real time to reflect the latest data characteristics. Step S42: dynamically adjusting the calculation of the Mahalanobis distance according to the updated sample covariance matrix to ensure the accuracy of the influence point selection; Step S43: for each newly acquired data point, the model coefficient matrix is ​​updated using a single-pass algorithm to achieve online model update; Step S44: periodically integrating information from historical data to improve the robustness of the model and controlling the complexity of the model through a sliding window technique; Step S5: In the distributed online monitoring step, each sensor node independently executes the influence point selection algorithm and exchanges information through the neighbor diffusion strategy to generate a global model coefficient matrix to achieve system monitoring. The specific implementation method includes: Step S51: Treat each sensor node as an independent computing unit and execute the key point selection algorithm on each node; Step S52: using the neighbor diffusion strategy, the local estimation result of each node is exchanged with its one-hop neighbor node to achieve global fusion of information; Step S53: recursively estimate the local model coefficients of each node using least squares, and gradually approach the global optimal solution through multiple iterations; Step S54: Summarize the local model coefficients of each node to generate a global model coefficient matrix as the overall monitoring result of the power system; Step S6: The consistency hypothesis test step is used to detect abnormal behavior of the system and perform statistical tests through the global model coefficient matrix to ensure the security and stability of the system. The specific steps are as follows: Step S61: Based on the global model coefficient matrix B t , calculates hypothesis test statistics for evaluating changes in system states, assuming: the current model coefficient matrix B t Equal to the coefficient matrix under normal conditions B Normal ; Step S62: Calculate the statistic using the Wald test formula W , evaluate the current statistics W Is it significant? Step S63: Calculation p value, p The value represents the assumption H 0 When true, the current statistic is observed W or more extreme values; Step S64: If p If the value is significantly lower than the preset level, then reject H 0, indicating that the system has detected significant abnormal behavior, triggering further monitoring and protection measures; Step S65: Collect and integrate the hypothesis test results of all nodes, process the asynchronous calculations between nodes through the distributed consensus algorithm, and generate a global p value; Step S66: Based on the global p The system makes consistent decisions based on the value and determines the overall system status. If an abnormality is detected in the entire system, the corresponding safety protection mechanism is activated to ensure the security and stability of the system. Step S7: The method of the present invention needs to be deployed in the power system eventually, and its performance is ensured through experimental verification and tuning. The system deployment and optimization steps specifically include: Step S71: deploying the method in the power system to ensure the real-time performance and accuracy of the system; Step S72: verifying the effectiveness of the proposed method from multiple aspects such as accuracy, stability, and computational efficiency; Step S73: According to the indicators in step S72, continue to optimize the algorithm and adjust the Mahalanobis distance threshold r and sliding window size to improve the adaptability and robustness of the system in different scenarios.

Claims

1. An online distributed safety monitoring method for a power system, characterized in that It includes the following seven steps: data initialization, multidimensional time series modeling, big data key point selection, online streaming data processing, distributed online monitoring, and consistency hypothesis testing: S1, the data initialization step: collecting and preprocessing multidimensional time series data from a sensor in the power system, performing preliminary statistical analysis, and calculating the sample covariance matrix through the initial sample to establish an initial reference standard for subsequent data analysis; S2, the multidimensional time series modeling step: constructing a vector autoregression model using the data initialized in step S1, estimating the model coefficient matrix by the least squares method, and optimizing the order of the model to accurately capture the temporal dependency between sensor nodes; S3, the big data key point selection step: using Mahalanobis distance to evaluate the impact of the multidimensional time series data points in step S1, by setting the Mahalanobis distance threshold r Screen out key influencing points to ensure that the selected key influencing points can represent the global characteristics of the data and form a small-scale data set; S4, the online streaming data processing step: when the data stream is obtained, the sample covariance matrix and the coefficients of the autoregressive model are updated in real time, key points are dynamically adjusted and selected, and historical data are periodically integrated to ensure that the model can continuously reflect the latest data characteristics; S5, the distributed online monitoring step: repeating steps S1 to S4 at each sensor node until all sensors are traversed, and information is exchanged through a neighbor diffusion strategy to generate a global model coefficient matrix, and the power system is monitored based on multidimensional stream data; S6, the consistency hypothesis test step: compare the global model coefficient matrix with the model coefficient matrix under the assumed normal state, and calculate p The system detects abnormal behavior of the value system, integrates the inspection results of each node, and the system makes a consistent decision on whether there is a global abnormality and takes corresponding actions; According to the online distributed safety monitoring method for a power system as claimed in claim 1, it is characterized in that in step S1, the data initialization step includes the following operations: S11, using monitoring components in the power system to collect multi-dimensional time series data of each sensor node; S12, performing preprocessing operations of cleaning, denoising and formatting the data collected in step S11; S13, performing preliminary analysis on the data preprocessed in step S12 to determine basic statistical characteristics and data distribution of the data; S14. Randomly select data within a time period of the time series as the initial sample to estimate the sample covariance matrix and provide a benchmark for the subsequent online monitoring algorithm.

2. According to the method for online distributed safety monitoring of power system described in claim 1, it is characterized in that: In step S2, the multidimensional time series modeling step includes the following operations: S21, based on the multidimensional time series data collected and processed in step S1, construct a vector autoregression model to capture the temporal dependency between the data; S22. Determine the order of the vector autoregression model p , based on the Bayesian information criterion, the order of the model is optimized to improve the fitting accuracy of the model; S23, using the least squares method to estimate the coefficient matrix of the vector autoregression model, ensuring that the coefficient matrix can accurately reflect the interdependence between sensor nodes; S24. Use cross-validation to evaluate the performance of the model and improve the generalization ability of the model through parameter tuning.

3. According to the online distributed safety monitoring method for power system described in claim 1, it is characterized in that: In step S3, the big data key point selection step includes the following operations: S31, expanding the vector autoregression model obtained in step S2 to a nonlinear form for processing more complex time series data, thereby improving the model's ability to capture nonlinear relationships; S32, calculating the Mahalanobis distance of each time point in the time series data processed in step S1, so as to evaluate the influence of the data point on the model and screen out the key point set that contributes more to the model; S33, based on the selected Mahalanobis distance threshold r , select the key point set SIPS , to reduce data size and improve computational efficiency; S34. By analyzing the key point set SIPS The statistical properties of the key points are re-evaluated to ensure that they cover the main distribution interval of the data. Otherwise, steps S31 to S33 are repeated to reselect the key point set to better represent the global characteristics.

4. According to the online distributed safety monitoring method for power system described in claim 1, it is characterized in that: In step S4, the online streaming data processing step includes the following operations: S41, real-time update of the sample covariance matrix to reflect the latest data characteristics; S42, dynamically adjusting the calculation of the Mahalanobis distance according to the updated sample covariance matrix to ensure the accuracy of the influence point selection; S43, for each newly acquired data point, use a single-pass algorithm to update the model coefficient matrix to achieve online model update; S44. Periodically integrate historical data information to improve the robustness of the model and control the complexity of the model through sliding window technology.

5. According to the online distributed safety monitoring method for power system described in claim 1, it is characterized in that: In step S5, the distributed online monitoring step includes the following operations: S51, treating each sensor node as an independent computing unit, and executing a key point selection algorithm on each node; S52, using the neighbor diffusion strategy, exchanging the local estimation result of each node with its one-hop neighbor node to achieve global fusion of information; S53, performing recursive least square estimation on the local model coefficients of each node, and gradually approaching the global optimal solution through multiple iterations; S54, summarizing the local model coefficients of each node to generate a global model coefficient matrix as the overall monitoring result of the power system.

6. According to the method for online distributed safety monitoring of power system described in claim 1, it is characterized in that: In step S6, the consistency hypothesis testing step includes the following operations: S61, based on the global model coefficient matrix B t , calculates the hypothesis test statistic used to evaluate changes in system states H 0: Current global model coefficient matrix B t Equal to the coefficient matrix under normal conditions B Normal ; S62. Use the Wald test formula to calculate statistics W , evaluate the current statistics W Is it significant? S63, calculation p value, p The value represents the assumption H When 0 is true, the current statistic is observed W or more extreme values; S64, if p If the value is significantly lower than the preset level, then reject H 0, indicating that the system has detected significant abnormal behavior, triggering further monitoring and protection measures; S65, collect and integrate the hypothesis test results of all nodes, process the asynchronous calculations between nodes through the distributed consistency algorithm, and generate a global p value; S66, based on the global p The system makes consistent decisions based on the value and determines the overall system status. If an abnormality is detected in the entire system, the corresponding safety protection mechanism is activated to ensure the security and stability of the system.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the online distributed security monitoring method for a power system according to any one of claims 1 to 6.

Images