A method and system for detecting and analyzing power data anomaly based on big data

Abstract

The present invention discloses a method and system for detecting and analyzing power data anomaly based on big data, which relates to the technical field of big data. pre , obtain the interaction influence relationship matrix I between devices inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter , predict the abnormal mode under the interaction between devices and obtain the abnormal model matrix A detect , and the generated comprehensive anomaly score S for the device abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , obtain the interactive operation power maintenance assessment plan of the equipment, solve the problem of cross-device anomalies, improve the fault location and emergency response capabilities of the power system, and significantly reduce equipment downtime and resource waste caused by unidentified anomalies, ensuring the stability of the power system.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a method and system for detecting and analyzing anomaly in power data based on big data. Background Art

[0002] In the field of smart grid, real-time monitoring and analysis of power data is one of the key tasks to ensure the stable operation of the power system. With the increase in the types of power equipment and the increase in system complexity, traditional single-device data monitoring can no longer meet the needs of modern power systems. Specifically, power data anomaly detection in the power system requires extracting valuable information from the collaborative data of multiple devices to ensure the stability and efficiency of the power grid.

[0003] In the invention of Chinese Patent Publication No. CN202311687634.4, a method and system for monitoring power data based on big data is disclosed. The method first obtains a time series sequence containing power data of power equipment of the same type, segments the power change curve fitted by the time series sequence to obtain curve segments, selects the operation stage and standby stage of the power equipment from all curve segments, and detects abnormal data of the power equipment from the operation stage and the standby stage respectively, corrects the corresponding abnormal data according to the obtained abnormal degree, obtains the corrected sequence of each power equipment, further predicts the future power data of the power equipment based on the corrected sequence, obtains the predicted data, and monitors the power data of the power equipment in real time according to the predicted data. The present invention can improve the accuracy of the prediction of future power data of power equipment, thereby improving the accuracy of real-time power data monitoring.

[0004] It can be seen that in power data anomaly detection, time series data of a single device is usually used for monitoring and analysis. However, the limitation of this method is that it cannot fully consider the mutual influence between devices in the power system, resulting in some cross-device anomalies that are difficult to accurately identify. For example, when a device is abnormal due to the operation or fluctuation of other devices, the existing method may misjudge it as a problem with the device itself, or fail to detect the potential source of the anomaly. Such shortcomings make the overall system fault location and response measures inefficient, and lack in-depth analysis of the collaborative relationship between multiple devices. In this case, once an anomaly occurs, the system may face unnecessary equipment downtime, resource scheduling errors or ineffective emergency response measures, further exacerbating the impact of the failure. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a method and system for detecting and analyzing anomaly in power data based on big data, which solves the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for detecting and analyzing power data anomalies based on big data, comprising the following steps:

[0007] S1. Sensors are arranged at each connection node of several connected power devices to collect power data P, power data W, temperature data T and vibration frequency data Z of the equipment in real time, and wavelet transform is used to remove noise to form the eigenvector matrix X of the equipment. pre ;

[0008] S2. Establish a device association interaction impact analysis model and transform the feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the interaction impact relationship matrix I between devices is obtained inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter ;

[0009] S3, through the obtained interaction influence relationship matrix I inter And the historical operation data of the equipment hist Perform multi-dimensional time series analysis to predict abnormal patterns under the interaction between devices and obtain the abnormal model matrix A detect ;

[0010] S4. Based on the acquired abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ;

[0011] S5. Comprehensive abnormality score of the equipment S abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , and based on the equipment prediction evaluation status result P pred Generate equipment interactive operation power maintenance assessment plan.

[0012] Preferably, step S1 also includes S11 and S12;

[0013] S11. By arranging sensors at each connection node of several connected power devices, the power data Pi, power data Wi, temperature data Ti and vibration frequency data Zi in the operation of the equipment are collected in real time. The sensors include power sensors, power sensors, temperature sensors and vibration sensors, wherein the power sensors include input and output data installed at the input and output ends of the power equipment to monitor the power, forming the collected power data Pi, including voltage Pi, V, current Pi, I and phase angle Pi, θ related parameters, and the power data Pi, power data Wi, temperature data Ti and vibration frequency data Zi signals obtained by different sensors are time-aligned to form a multi-dimensional data vector Di;

[0014] S12, perform wavelet transform on the collected multidimensional data vector Di to remove noise and extract relevant fluctuation characteristics, and generate the feature vector matrix X of device i pre ;

[0015] The eigenvector matrix X pre Specifically, the multidimensional data vector Di of device i is used to form the feature vector matrix X pre :

[0016]

[0017] Preferably, step S2 also includes S21 and S22;

[0018] S21. By using a graph convolutional network to establish a device association interaction impact analysis model, the relationship between devices is marked as a graph structure G(V, E), where V represents a device set, including n devices, and E represents the connection relationship between n devices. The feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the attention mechanism is introduced to adjust the relevant device relationships, and the interaction impact relationship matrix I between devices is obtained through the output of the device association interaction impact analysis model. inter ;

[0019] S22. Use the maximum likelihood estimation statistical method to optimize the parameters of the equipment association interaction impact analysis model and iterate the interaction impact relationship matrix I inter , iterate through the following calculation formula:

[0020]

[0021] In the formula, Represents the interaction influence relationship matrix I at the tth iteration inter , A represents the adjacency matrix, H represents the eigenvector matrix X pre .

[0022] Preferably, the interaction influence relationship matrix I inter Obtained through the following calculation formula:

[0023] I inter =σ(A·H (L-1) ·W (L) );

[0024] In the formula, σ represents the activation function, including sigmoid activation function, softmax activation function and ReLU activation function, which is specifically used for the input feature vector matrix X pre Data is transformed, A represents the adjacency matrix, which specifically represents the physical and logical connection relationships between devices, through the eigenvector matrix X pre To generate, H (L-1) Represents the feature vector matrix X of the L-1th layer pre , W (L) Represents the weight matrix of the output layer, which is specifically used to map the feature representation matrix of the previous layer to the output space;

[0025] The adjacency matrix A is an n*n matrix, where n represents the total number of devices, and is specifically expressed as follows:

[0026]

[0027] In the matrix, ai,j represents the connection weight relationship between device i and device j. When ai,j=1, it indicates that there is a physical connection relationship or a logical connection relationship between device i and device j. When ai,j=0, it indicates that there is no physical connection relationship or a logical connection relationship between device i and device j.

[0028] Preferably, step S3 also includes S31 and S32;

[0029] S31, the obtained interaction influence relationship matrix I inter And the historical operation data of the equipment hist Integrate historical operation data hist Including historical power data istP 、Historical power data H istW 、Historical temperature data H istT and historical vibration frequency data H istZ The integration includes data time axis alignment, multi-dimensional time series analysis and extraction of the integrated data, and obtaining the multi-dimensional time series prediction feature matrix T ext ;

[0030] S32, the obtained multi-dimensional time series prediction feature matrix T ext, detect abnormal patterns under the influence of interactions between devices, the detection includes using an autoregressive integrated moving average model to make predictions, and synchronously compare the predicted time series values ​​with the real-time acquired feature vector matrix X pre Match and obtain the deviation values ​​to form the deviation value matrix E anom , to construct the abnormal model matrix A detect .

[0031] Preferably, the multidimensional time series feature matrix T ext Obtained through the following calculation formula:

[0032]

[0033] In the formula, represents matrix concatenation, and PCA represents the principal component analysis method function.

[0034] Preferably, based on the acquired abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ;

[0035] The comprehensive abnormality score S abn Obtained through the following calculation formula:

[0036]

[0037] In the formula, K i represents the preset weight value of the i-th device, f(A detect,i ) represents the abnormality scoring function of the i-th device.

[0038] Preferably, step S5 also includes S51 and S52;

[0039] Among them, S51 is the comprehensive abnormality score of the equipment S abn Compare it with the preset device abnormality assessment threshold YC to obtain the predicted device's future interactive operation status and form the device prediction assessment status result P pred ;

[0040] The device predicts the evaluation status result P pred Obtained through the following comparison method:

[0041]

[0042] In the formula, Normal means that the state of the predicted device's future interactive operation is a normal result, and Anomalous means that the state of the predicted device's future interactive operation is an abnormal result.

[0043] Preferably, S52, according to the device prediction evaluation status result P pred Generate equipment interactive operation power maintenance assessment plan;

[0044] The equipment interactive operation power maintenance assessment plan is obtained by the following generation method:

[0045] When the equipment predicts the evaluation status result P pred When there is no abnormal result, a qualified plan for equipment interactive operation power maintenance assessment is generated, and the operating status of the current power equipment and related equipment are not adjusted;

[0046] When the equipment predicts the evaluation status result P pred When the result is abnormal, the equipment interactive operation power maintenance evaluation unqualified plan is generated, the operation status of the current power equipment and the related equipment are adjusted, and the interactive impact relationship matrix I is simultaneously inter and the eigenvector matrix X pre Notify and push information to relevant maintenance personnel to inspect the operating status of related equipment and associated equipment.

[0047] A power data anomaly detection and analysis system based on big data, comprising a data acquisition module, an impact analysis module, a correlation analysis module, an anomaly analysis module and an anomaly assessment module;

[0048] The data acquisition module arranges sensors at each connection node of several connected power devices to collect power data P, power data W, temperature data T and vibration frequency data Z of the equipment in real time, and uses wavelet transform to remove noise to form the eigenvector matrix X of the equipment. pre ;

[0049] The impact analysis module establishes a device association interaction impact analysis model to convert the feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the interaction impact relationship matrix I between devices is obtained inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter ;

[0050] The association analysis module obtains the interaction influence relationship matrix I inter And the historical operation data of the equipment hist Perform multi-dimensional time series analysis to predict abnormal patterns under the interaction between devices and obtain the abnormal model matrix A detect ;

[0051] The abnormal analysis module obtains the abnormal model matrix A detectConstruct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ;

[0052] The abnormality assessment module obtains the comprehensive abnormality score S of the device. abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , and based on the equipment prediction evaluation status result P pred Generate an equipment interactive operation power maintenance assessment plan, and carry out specific execution and notification according to the content of the equipment interactive operation power maintenance assessment plan.

[0053] The present invention provides a method and system for detecting and analyzing power data anomaly based on big data, which has the following beneficial effects:

[0054] (1) Through S1 to S5, the feature vector matrix X of the real-time acquisition device is obtained. pre , obtain the interaction influence relationship matrix I between devices inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter , predict the abnormal mode under the interaction between devices and obtain the abnormal model matrix A detect , and the resulting composite anomaly score S of the generated device abn , generating an interactive power maintenance assessment plan for equipment operation, solving the problem of cross-equipment anomalies, improving the fault location and emergency response capabilities of the power system, and significantly reducing equipment downtime and resource waste due to unidentified anomalies, thereby ensuring the safety, stability and efficiency of the power system.

[0055] (2) Use wavelet transform to denoise and extract features from multidimensional data, and generate the eigenvector matrix X of the eigenvector matrix device. pre It provides an accurate representation of the device status. In the analysis of the interaction impact between devices, it introduces graph convolutional networks and attention mechanisms to effectively model the complex relationship between devices, and performs parameter optimization iterations through maximum likelihood estimation to generate a more accurate interaction impact relationship matrix I inter Ultimately, the system was able to accurately identify and analyze abnormal interaction patterns between devices, increasing the sensitivity of the power system to cross-device anomalies.

[0056] (3) The key data is refined through principal component analysis, thereby improving the analysis accuracy. detect The constructed multi-dimensional data fusion model further enhances the correlation analysis capability between multi-dimensional data, and generates a comprehensive abnormality score S for the equipment. abnThis provides a scientific basis for the quantitative evaluation of equipment health status. abn By comparing with the preset threshold, the system can accurately predict the future interactive operation status of the equipment and generate a maintenance assessment plan accordingly. It can respond effectively at the first time and notify relevant maintenance personnel to make necessary adjustments and inspections. This method not only solves the problem of insufficient analysis of multi-device collaborative relationships in traditional power data monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a schematic diagram of the steps of a method for detecting and analyzing power data anomalies based on big data according to the present invention;

[0058] Figure 2 The present invention is a schematic diagram of a block diagram of a power data anomaly detection and analysis system based on big data. DETAILED DESCRIPTION

[0059] 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, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0060] Example 1

[0061] The present invention provides a method for detecting and analyzing power data anomalies based on big data. Figure 1 , including the following steps:

[0062] S1. Sensors are arranged at each connection node of several connected power devices to collect power data P, power data W, temperature data T and vibration frequency data Z of the equipment in real time, and wavelet transform is used to remove noise to form the eigenvector matrix X of the equipment. pre ;

[0063] S2. Establish a device association interaction impact analysis model and transform the feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the interaction impact relationship matrix I between devices is obtained inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter ;

[0064] S3, through the obtained interaction influence relationship matrix I inter And the historical operation data of the equipment histPerform multi-dimensional time series analysis to predict abnormal patterns under the interaction between devices and obtain the abnormal model matrix A detect ;

[0065] S4. Based on the acquired abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ;

[0066] S5. Comprehensive abnormality score of the equipment S abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , and based on the equipment prediction evaluation status result P pred Generate equipment interactive operation power maintenance assessment plan.

[0067] In this embodiment, through S1 to S5, the feature vector matrix X of the real-time acquisition device is obtained. pre , obtain the interaction influence relationship matrix I between devices inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter , predict the abnormal mode under the interaction between devices and obtain the abnormal model matrix A detect , and the resulting composite anomaly score S of the generated device abn , generating an interactive power maintenance assessment plan for equipment operation, solving the problem of cross-equipment anomalies, improving the fault location and emergency response capabilities of the power system, and significantly reducing equipment downtime and resource waste due to unidentified anomalies, thereby ensuring the safety, stability and efficiency of the power system.

[0068] Example 2

[0069] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: step S1 also includes S11 and S12;

[0070] S11. By arranging sensors at each connection node of several connected power devices, the power data Pi, power data Wi, temperature data Ti and vibration frequency data Zi in the operation of the equipment are collected in real time. The sensors include power sensors, power sensors, temperature sensors and vibration sensors, wherein the power sensors include input and output data installed at the input and output ends of the power equipment to monitor the power, forming the collected power data Pi, including voltage Pi, V, current Pi, I and phase angle Pi, θ related parameters, and the power data Pi, power data Wi, temperature data Ti and vibration frequency data Zi signals obtained by different sensors are time-aligned to form a multi-dimensional data vector Di;

[0071] S12, perform wavelet transform on the collected multidimensional data vector Di to remove noise and extract relevant fluctuation characteristics, and generate the feature vector matrix X of device i pre ;

[0072] The eigenvector matrix X pre Specifically, the multidimensional data vector Di of device i is used to form the feature vector matrix X pre :

[0073]

[0074] Step S2 also includes S21 and S22;

[0075] S21. By using a graph convolutional network to establish a device association interaction impact analysis model, the relationship between devices is marked as a graph structure G(V, E), where V represents a device set, including n devices, and E represents the connection relationship between n devices. The feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the attention mechanism is introduced to adjust the relevant device relationships, and the interaction impact relationship matrix I between devices is obtained through the output of the device association interaction impact analysis model. inter ;

[0076] S22. Use the maximum likelihood estimation statistical method to optimize the parameters of the equipment association interaction impact analysis model and iterate the interaction impact relationship matrix I inter , iterate through the following calculation formula:

[0077]

[0078] In the formula, Represents the interaction influence relationship matrix I at the tth iteration inter , A represents the adjacency matrix, H represents the eigenvector matrix X pre .

[0079] The interaction influence relationship matrix Iinter Obtained through the following calculation formula:

[0080] I inter =σ(A·H (L-1) ·W (L) );

[0081] In the formula, σ represents the activation function, including sigmoid activation function, softmax activation function and ReLU activation function, which is specifically used for the input feature vector matrix X pre Data is transformed, A represents the adjacency matrix, which specifically represents the physical and logical connection relationships between devices, through the eigenvector matrix X pre To generate, H (L-1) Represents the feature vector matrix X of the L-1th layer pre , W (L) Represents the weight matrix of the output layer, which is specifically used to map the feature representation matrix of the previous layer to the output space;

[0082] The adjacency matrix A is an n*n matrix, where n represents the total number of devices, and is specifically expressed as follows:

[0083]

[0084] In the matrix, ai,j represents the connection weight relationship between device i and device j. When ai,j=1, it indicates that there is a physical connection relationship or a logical connection relationship between device i and device j. When ai,j=0, it indicates that there is no physical connection relationship or a logical connection relationship between device i and device j.

[0085] In this embodiment, wavelet transform is used to perform denoising and feature extraction on multidimensional data, and the feature vector matrix X of the generated feature vector matrix device is pre It provides an accurate representation of the device status. In the analysis of the interaction impact between devices, it introduces graph convolutional networks and attention mechanisms to effectively model the complex relationship between devices, and performs parameter optimization iterations through maximum likelihood estimation to generate a more accurate interaction impact relationship matrix I inter Ultimately, the system was able to accurately identify and analyze abnormal interaction patterns between devices, increasing the sensitivity of the power system to cross-device anomalies.

[0086] Example 3

[0087] This embodiment is explained in Example 2. Please refer to Figure 1 , specifically: step S3 also includes S31 and S32;

[0088] S31, the obtained interaction influence relationship matrix I inter And the historical operation data of the equipmenthist Integrate historical operation data hist Including historical power data istP 、Historical power data H istW 、Historical temperature data H istT and historical vibration frequency data H istZ The integration includes data time axis alignment, multi-dimensional time series analysis and extraction of the integrated data, and obtaining the multi-dimensional time series prediction feature matrix T ext ;

[0089] S32, the obtained multi-dimensional time series prediction feature matrix T ext , detect abnormal patterns under the influence of interactions between devices, the detection includes using an autoregressive integrated moving average model to make predictions, and synchronously compare the predicted time series values ​​with the real-time acquired feature vector matrix X pre Match and obtain the deviation values ​​to form the deviation value matrix E anom , to construct the abnormal model matrix A detect .

[0090] Example 4

[0091] This embodiment is explained in Example 3, please refer to Figure 1 , specifically: Among them, the multidimensional time series feature matrix T ext Obtained through the following calculation formula:

[0092]

[0093] In the formula, represents matrix concatenation, and PCA represents the principal component analysis method function.

[0094] Among them, based on the obtained abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ;

[0095] The comprehensive abnormality score S abn Obtained through the following calculation formula:

[0096]

[0097] In the formula, K i represents the preset weight value of the i-th device, f(A detect,i ) represents the abnormality scoring function of the i-th device.

[0098] Step S5 also includes S51 and S52;

[0099] Among them, S51 is the comprehensive abnormality score of the equipment S abn Compare it with the preset device abnormality assessment threshold YC to obtain the predicted device's future interactive operation status and form the device prediction assessment status result P pred ;

[0100] The device predicts the evaluation status result P pred Obtained through the following comparison method:

[0101]

[0102] In the formula, Normal means that the state of the predicted device's future interactive operation is a normal result, and Anomalous means that the state of the predicted device's future interactive operation is an abnormal result.

[0103] S52, based on the equipment prediction evaluation status result P pred Generate equipment interactive operation power maintenance assessment plan;

[0104] The equipment interactive operation power maintenance assessment plan is obtained by the following generation method:

[0105] When the equipment predicts the evaluation status result P pred When there is no abnormal result, a qualified plan for equipment interactive operation power maintenance assessment is generated, and the operating status of the current power equipment and related equipment are not adjusted;

[0106] When the equipment predicts the evaluation status result P pred When the result is abnormal, the equipment interactive operation power maintenance evaluation unqualified plan is generated, the operation status of the current power equipment and the related equipment are adjusted, and the interactive impact relationship matrix I is simultaneously inter and the eigenvector matrix X pre Notify and push information to relevant maintenance personnel to inspect the operating status of related equipment and associated equipment.

[0107] In this embodiment, the key data is refined by the principal component analysis method, thereby improving the analysis accuracy. detect The constructed multi-dimensional data fusion model further enhances the correlation analysis capability between multi-dimensional data, and generates a comprehensive abnormality score S for the equipment. abn This provides a scientific basis for the quantitative evaluation of equipment health status. abn By comparing with the preset threshold, the system can accurately predict the future interactive operation status of the equipment and generate a maintenance assessment plan accordingly. It can respond effectively at the first time and notify relevant maintenance personnel to make necessary adjustments and inspections. This method not only solves the problem of insufficient analysis of multi-device collaborative relationships in traditional power data monitoring.

[0108] Example 5

[0109] A power data anomaly detection and analysis system based on big data, please refer to Figure 2 ,Specifically: including data acquisition module, impact analysis module, correlation analysis module, anomaly analysis module and anomaly assessment module;

[0110] The data acquisition module arranges sensors at each connection node of several connected power devices to collect power data P, power data W, temperature data T and vibration frequency data Z of the equipment in real time, and uses wavelet transform to remove noise to form the eigenvector matrix X of the equipment. pre ;

[0111] The impact analysis module establishes a device association interaction impact analysis model to convert the feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the interaction impact relationship matrix I between devices is obtained inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter ;

[0112] The association analysis module obtains the interaction influence relationship matrix I inter And the historical operation data of the equipment hist Perform multi-dimensional time series analysis to predict abnormal patterns under the interaction between devices and obtain the abnormal model matrix A detect ;

[0113] The abnormal analysis module obtains the abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ;

[0114] The abnormality assessment module obtains the comprehensive abnormality score S of the device. abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , and based on the equipment prediction evaluation status result P pred Generate an equipment interactive operation power maintenance assessment plan, and carry out specific execution and notification according to the content of the equipment interactive operation power maintenance assessment plan.

[0115] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting and analyzing power data anomalies based on big data, characterized in that: The following steps are involved: S1. Sensors are arranged at each connection node of several connected power devices to collect power data P, power data W, temperature data T and vibration frequency data Z of the equipment in real time, and wavelet transform is used to remove noise to form the eigenvector matrix X of the equipment. pre ; S2. Establish a device association interaction impact analysis model and transform the feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the interaction impact relationship matrix I between devices is obtained inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter ; S3, through the obtained interaction influence relationship matrix I inter And the historical operation data of the equipment hist Perform multi-dimensional time series analysis to predict abnormal patterns under the interaction between devices and obtain the abnormal model matrix A detect ; Step S3 also includes S31 and S32; S31, the obtained interaction influence relationship matrix I inter And the historical operation data of the equipment hist Integrate historical operation data hist Including historical power data istP 、Historical power data H istW 、Historical temperature data H istT and historical vibration frequency data H istZ The integration includes data time axis alignment, multi-dimensional time series analysis and extraction of the integrated data, and obtaining the multi-dimensional time series prediction feature matrix T ext ; S32, the obtained multi-dimensional time series prediction feature matrix T ext , detect abnormal patterns under the influence of interactions between devices, the detection includes using an autoregressive integrated moving average model to make predictions, and synchronously compare the predicted time series values ​​with the real-time acquired feature vector matrix X pre Match and obtain the deviation values ​​to form the deviation value matrix E anom , to construct the abnormal model matrix A detect ; Among them, the multidimensional time series feature matrix T ext Obtained through the following calculation formula: In the formula, represents matrix concatenation, PCA represents principal component analysis method function; S4. Based on the acquired abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ; S5. Comprehensive abnormality score of the equipment S abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , and based on the equipment prediction evaluation status result P pred Generate equipment interactive operation power maintenance assessment plan.

2. The method for detecting and analyzing power data anomalies based on big data according to claim 1, characterized in that: Step S1 also includes S11 and S12; S11. By arranging sensors at each connection node of several connected power devices, the power data Pi, power data Wi, temperature data Ti and vibration frequency data Zi in the operation of the equipment are collected in real time. The sensors include power sensors, power sensors, temperature sensors and vibration sensors, wherein the power sensors include input and output data installed at the input and output ends of the power equipment to monitor the power, forming the collected power data Pi, including voltage Pi, V, current Pi, I and phase angle Pi, θ related parameters, and the power data Pi, power data Wi, temperature data Ti and vibration frequency data Zi signals obtained by different sensors are time-aligned to form a multi-dimensional data vector Di; S12, perform wavelet transform on the collected multidimensional data vector Di to remove noise and extract relevant fluctuation characteristics, and generate the feature vector matrix X of device i pre ; The eigenvector matrix X pre Specifically, the multidimensional data vector Di of device i is used to form the feature vector matrix X pre :

3. The method for detecting and analyzing power data anomalies based on big data according to claim 1 is characterized in that: Step S2 also includes S21 and S22; S21. By using a graph convolutional network to establish a device association interaction impact analysis model, the relationship between devices is marked as a graph structure G(V, E), where V represents a device set, including n devices, and E represents the connection relationship between n devices. The feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the attention mechanism is introduced to adjust the relevant device relationships, and the interaction impact relationship matrix I between devices is obtained through the output of the device association interaction impact analysis model. inter ; S22. Use the maximum likelihood estimation statistical method to optimize the parameters of the equipment association interaction impact analysis model and iterate the interaction impact relationship matrix I inter , iterate through the following calculation formula: In the formula, Represents the interaction influence relationship matrix I at the tth iteration inter , A represents the adjacency matrix, H represents the eigenvector matrix X pre .

4. The method for detecting and analyzing power data anomalies based on big data according to claim 3 is characterized in that: The interaction influence relationship matrix I inter Obtained through the following calculation formula: I inter =σ(A·H (L-1) ·W (L) ); In the formula, σ represents the activation function, including sigmoid activation function, softmax activation function and ReLU activation function, which is specifically used for the input feature vector matrix X pre Data is transformed, A represents the adjacency matrix, which specifically represents the physical and logical connection relationships between devices, through the eigenvector matrix X pre To generate, H (L-1) Represents the feature vector matrix X of the L-1th layer pre , W (L) Represents the weight matrix of the output layer, which is specifically used to map the feature representation matrix of the previous layer to the output space; The adjacency matrix A is an n*n matrix, where n represents the total number of devices, and is specifically expressed as follows: In the matrix, ai,j represents the connection weight relationship between device i and device j. When ai,j=1, it indicates that there is a physical connection relationship or a logical connection relationship between device i and device j. When ai,j=0, it indicates that there is no physical connection relationship or a logical connection relationship between device i and device j.

5. The method for detecting and analyzing power data anomalies based on big data according to claim 1 is characterized in that: in, Based on the obtained abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ; The comprehensive abnormality score S abn Obtained through the following calculation formula: In the formula, K i represents the preset weight value of the i-th device, f(A detect,i ) represents the abnormality scoring function of the i-th device.

6. The method for detecting and analyzing power data anomalies based on big data according to claim 1 is characterized in that: Step S5 also includes S51 and S52; Among them, S51 is the comprehensive abnormality score of the equipment S abn Compare it with the preset device abnormality assessment threshold YC to obtain the predicted device's future interactive operation status and form the device prediction assessment status result P pred ; The device predicts the evaluation status result P pred Obtained through the following comparison method: In the formula, Normal means that the state of the predicted device's future interactive operation is a normal result, and Anomalous means that the state of the predicted device's future interactive operation is an abnormal result.

7. The method for detecting and analyzing power data anomalies based on big data according to claim 6 is characterized in that: S52, based on the equipment prediction evaluation status result P pred Generate equipment interactive operation power maintenance assessment plan; The equipment interactive operation power maintenance assessment plan is obtained by the following generation method: When the equipment predicts the evaluation status result P pred When there is no abnormal result, a qualified plan for equipment interactive operation power maintenance assessment is generated, and the operating status of the current power equipment and related equipment are not adjusted; When the equipment predicts the evaluation status result P pred When the result is abnormal, the equipment interactive operation power maintenance evaluation unqualified plan is generated, the operation status of the current power equipment and the related equipment are adjusted, and the interactive impact relationship matrix I is simultaneously inter and the eigenvector matrix X pre Notify and push information to relevant maintenance personnel to inspect the operating status of related equipment and associated equipment.

8. A power data anomaly detection and analysis system based on big data, comprising a power data anomaly detection and analysis method based on big data as claimed in any one of claims 1 to 7, characterized in that: It includes data collection module, impact analysis module, correlation analysis module, anomaly analysis module and anomaly assessment module; The data acquisition module arranges sensors at each connection node of several connected power devices to collect power data P, power data W, temperature data T and vibration frequency data Z of the equipment in real time, and uses wavelet transform to remove noise to form the eigenvector matrix X of the equipment. pre ; The impact analysis module establishes a device association interaction impact analysis model to convert the feature vector matrix X pre As the input training data of the device association interaction impact analysis model, the interaction impact relationship matrix I between devices is obtained inter , and use maximum likelihood estimation to optimize the parameters of the equipment association interaction impact analysis model to iterate the interaction impact relationship matrix I inter ; The association analysis module obtains the interaction influence relationship matrix I inter And the historical operation data of the equipment hist Perform multi-dimensional time series analysis to predict abnormal patterns under the interaction between devices and obtain the abnormal model matrix A detect ; The abnormal analysis module obtains the abnormal model matrix A detect Construct a multi-dimensional data fusion model to evaluate the correlation between multi-dimensional data and the interaction between devices, and generate a comprehensive abnormality score S for the device. abn ; The abnormality assessment module obtains the comprehensive abnormality score S of the device. abn Predict the future interactive operation status of the device and obtain the device prediction evaluation status result P pred , and based on the equipment prediction evaluation status result P pred Generate an equipment interactive operation power maintenance assessment plan, and carry out specific execution and notification according to the content of the equipment interactive operation power maintenance assessment plan.

Images