Power system abnormal data recovery method and system considering time and precision priority

Abstract

The invention provides a power system abnormal data recovery method and system considering time and precision priority, and belongs to the technical field of power data security. The method includes acquiring power data and judging whether abnormal values exist or not; obtaining quantized values of each abnormal value in two dimensions based on a safety protection level index and a response recovery level index to form a row vector, and further establishing a decision index matrix; determining a weight matrix by using an analytic hierarchy process; calculating a comprehensive score of each abnormal value based on the decision index matrix and the weight matrix; determining a priority label according to the comprehensive score and a threshold value, and performing label processing on each abnormal value in the power data set to obtain a data set with the priority label; recovering each abnormal value in the data set by adopting a recovery algorithm corresponding to the priority label to obtain a normal data set; the recovery algorithm comprises a high-precision recovery algorithm and a rapid recovery algorithm; and sending the normal data set to a data control center to provide a data source for real-time decision analysis of a subsequent power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power data security, and in particular to a method and system for recovering abnormal data of a power system with a priority given to time and accuracy. Background Art

[0002] As an indispensable infrastructure of modern society, the power system supports the normal operation of the economy and society. With the continuous expansion of the scale and increase of complexity of the power system, the importance of power data in business forecasting, power generation dispatching and system operation has become increasingly prominent. The network architecture of the new power system can be divided into two categories according to the business type and security level: the network for production data transmission and the network for management data transmission.

[0003] Historical data is often used as the basis for analysis in power systems for real-time monitoring or future prediction. The time interval length unit of the historical data involved can be seconds, minutes, hours, and days. For some fine monitoring, even several time windows are extracted and analyzed for data within 1 second. Therefore, during the data transmission process, data delay or loss may cause data anomalies that may directly affect the safe and stable operation of the system, resulting in large errors in prediction results and inaccurate monitoring information. Therefore, recovering abnormal data is a common operation in power systems, and the requirements for data recovery operations are also very high. The prior art proposes to use linear interpolation to recover data. This method has a fast response and can ensure the continuity and security of the system, but the accuracy is not high. It is more suitable for businesses directly related to the real-time operation and control of the power system, such as dispatching automation, equipment monitoring and fault handling, etc., which require high data recovery capabilities for power production data. However, power management data is mainly used for long-term planning, data analysis and business management. These data play a vital role in power system planning, decision support and optimal scheduling. Its data accuracy directly affects the reliability of analysis results and the accuracy of decision-making. The prior art solution is difficult to support the high-precision requirements of power management data for data recovery. Summary of the invention

[0004] In view of this, the present invention provides a method and system for recovering abnormal data of an electric power system that takes time and accuracy into consideration, adopts corresponding recovery methods according to the requirements of different types of electric power data recovery, and at the same time meets the accuracy and response speed requirements in the process of recovering abnormal data of the electric power system, providing reliable data support for the stable operation and real-time decision-making of the electric power system.

[0005] The technical solution adopted by the embodiment of the present invention to solve the technical problem is:

[0006] A method for recovering abnormal data of a power system taking into account time and accuracy priority, comprising:

[0007] Step S1, obtaining a power data set and determining whether there are abnormal values; the abnormal values ​​include telecontrol information exchanged between the data center SCADA-EMS system and the plant station, data information exchanged between the data center SCADA and the EMS system, electric energy metering information, and other power market information, wherein the other power market information includes power generation cost, market clearing price, transaction volume, and user load;

[0008] Step S2: based on the safety protection level index and the response recovery level index, the quantized values ​​of each abnormal value in two dimensions are obtained to form a row vector A. ij , and further establish the decision indicator matrix A for all outliers, where A ij middle element a ij It represents the quantitative value of the i-th outlier under the j-th level indicator;

[0009] Step S3, using the analytic hierarchy process to determine the weight matrix W = [w1, w2] T , including the security protection level weight w1 and the response recovery level weight w2;

[0010] Step S4, calculating the comprehensive score Z of each outlier based on the decision indicator matrix A and the weight vector W i ;

[0011] Step S5, according to the comprehensive score Z i And the threshold Z′ determines the priority label Priority i , labeling each of the abnormal values ​​in the power data set to obtain a data set with priority labels; the priority labels correspond to a recovery algorithm for the abnormal values;

[0012] Step S6, using a recovery algorithm corresponding to the priority label to recover each of the abnormal values ​​in the data set with the priority label, and obtaining a normal data set after data recovery; the recovery algorithm includes a high-precision recovery algorithm and a fast recovery algorithm; the fast recovery algorithm uses a low-rank matrix decomposition algorithm for data recovery, and the high-precision recovery algorithm uses a spatiotemporal convolutional network STCN for data recovery, and the spatiotemporal convolutional network STCN is composed of a temporal convolutional network TCN, a graph convolutional network GCN and a fully connected layer;

[0013] Step S7, sending the normal data set to the data control center to provide a data source for subsequent power system analysis.

[0014] Preferably, the quantitative standard of the safety protection level index is: when the data type of the abnormal value belongs to the telecontrol information exchanged between the data center SCADA / EMS system and the plant station, the quantitative value is 4, when it belongs to the data information exchanged between the data center SCADA / EMS system, the quantitative value is 3, when it belongs to the electric energy metering information, the quantitative value is 2, and when it belongs to other information of the power market, the quantitative value is 1;

[0015] The quantification standard of the response recovery level indicator is: when the allowed response time for data recovery of the abnormal value is an immediate response, the quantization value is 5; when the allowed response time for data recovery is less than 1s, the quantization value is 4; when the allowed response time for data recovery is less than 1min, the quantization value is 3; when the allowed response time for data recovery is less than 5min and greater than 1min, the quantization value is 2; when the allowed response time for data recovery is more than 5min, the quantization value is 1.

[0016] Preferably, the step S2 comprises:

[0017] Step S21: derive the quantitative value a of the i-th abnormal value in the dimension of the security protection level index based on the security protection level index. i1 ; Based on the response recovery level index, the quantitative value a of the i-th abnormal value in the response recovery level index dimension is obtained i2 ;

[0018] Step S22, let row vector A ij =[a i1 ,a i2 ], i row vectors form the decision indicator matrix A:

[0019]

[0020] Where i∈[1,n];

[0021] Step S23, normalize the matrix A to obtain the decision indicator matrix A′:

[0022]

[0023] In the formula, A j represents the column vector in matrix A; min(A j ) and max(A j ) represent the minimum and maximum values ​​in column j, respectively.

[0024] Preferably, the step S3 comprises:

[0025] Construct the judgment matrix P:

[0026]

[0027] Among them, p 11 Indicates that the importance of the security protection index relative to itself is 1, p 12 Indicates the importance of security protection relative to response and recovery, p 12 ∈[1,9], let p 21 =1 / p 12 、p 22 =1, p 21 Indicates the importance of response recovery relative to security protection, p 22 Indicates that the importance of response recovery relative to itself is 1;

[0028] Perform column normalization on each element of the judgment matrix P to obtain a normalized judgment matrix P′:

[0029]

[0030] Among them, the element p′ ij The calculation method is:

[0031]

[0032] Perform average processing on each row of the judgment matrix P′:

[0033]

[0034] Get the security protection level weight w1 and response recovery level weight w2, as well as the weight matrix W = [w1, w2] T .

[0035] Preferably, the step S4 comprises:

[0036] By multiplying the decision indicator matrix A′ with the weight matrix W, the comprehensive score Z of each outlier is calculated. i :

[0037]

[0038] Among them, Z i Indicates i The composite score of the outliers.

[0039] Preferably, the priority tag Priority in step S5 i The generated rules are:

[0040]

[0041] Among them, Priority i A value of 1 indicates that the fast recovery algorithm is selected.i A value of 0 indicates that a high-precision recovery algorithm is selected.

[0042] Preferably, the step S6 comprises:

[0043] For the priority tag Priority i For outliers with a value of 1, a low-rank matrix decomposition algorithm is used for rapid recovery:

[0044] Let the data matrix Y represent the power data set, and establish a mask matrix W about the data matrix Y, where the element w in the mask matrix W is ij Represents the element y in the data matrix Y ij The effectiveness is expressed as:

[0045]

[0046] Establish low-rank matrices U and V, and use the optimization objective function to adjust the reconstruction matrix X=UV so that the reconstruction matrix X approaches the valid data in Y; the optimization objective function is:

[0047]

[0048] In the formula, w ij (Y ij -(UV) ij ) 2 Indicates that the error is calculated only at valid data locations; represents the regularization term; λ represents the regularization parameter;

[0049] Auxiliary variables Z and Lagrange multiplier matrix Λ are introduced, and the optimization objective function is decomposed into two sub-problems through the alternating direction multiplier method ADMM, optimizing X and Z respectively, and iteratively updating X, Z and Λ:

[0050] First, fix Z and Λ, optimize the value of X, and consider the current values ​​of Z and Λ. The iterative formula of X is:

[0051]

[0052] In the formula, k is the number of iterations, and ρ is the penalty parameter control;

[0053] Then optimize the auxiliary variable Z, the optimization formula of Z is:

[0054]

[0055] Update the Lagrange multiplier matrix Λ, the update formula of Λ is:

[0056] Λ (k+1) =Λ k +ρ(X (k+1) -Z(k+1) )

[0057] Calculate the error between the valid data locations in the reconstructed matrices X and Y ‖(X (k+1) -Y)⊙W‖ F , when ‖(X (k+1) -Y)⊙W‖ F <ε, where ε is the preset threshold.

[0058] Using the final reconstructed matrix X = UV, a low-rank inference method is used to fill in missing data and repair abnormal data based on valid data in Y;

[0059] Further,

[0060] For the priority tag Priority i The outliers with a value of 0 are restored with high precision using the spatiotemporal convolutional network STCN:

[0061] Assume that the input time series of the power data in the power data set is Q = {q1, q2, q3, ..., q t}, where q t represents the power data collected at time t;

[0062] The convolution operation of TCN is defined as:

[0063]

[0064] In the formula, z t represents the temporal features extracted by TCN at time t; w i is the weight of the convolution kernel; k is the size of the convolution kernel, indicating the number of time points involved in a single convolution operation; d i is the expansion rate; b is the bias term; t>d k-1 ; After the dilated convolution operation of multiple layers of TCN, the output feature sequence Z is obtained t ={z t1 ,z t2 ,z t3 ,...,z tn};

[0065] Using Z t Further construct the initial feature matrix H (0) =[z t1 ,z t2 ,z t3 ,...,z tn ] as the input of the graph convolutional network GCN;

[0066] The spatial features of the data are extracted using the graph convolutional network GCN. The topological structure of the power system is abstractly represented as a graph G = (V, E), where V is a set of nodes, representing the measurement points in the power system; E is a set of edges, representing the physical connection between the measurement points; the adjacency matrix A describes the connection relationship between the nodes: if the nodes are connected, then A = 1; otherwise A = 0; the degree matrix D is used to represent the number of connections for each node; the graph convolution calculation formula for the l+1th layer is:

[0067]

[0068] In the formula, H (l+1) is the node feature matrix of the l+1th layer, initially H (0) is the time series feature matrix extracted by TCN; W (l) is the weight matrix of the lth layer; σ is the activation function; through the layer-by-layer operation of graph convolution, the node features are propagated and integrated in the topological structure of the power system to generate a node feature matrix H containing spatiotemporal information (L) As the final output matrix H of the graph convolutional network, the final output matrix H contains the temporal and spatial features of each node;

[0069] Finally, the final output matrix H is mapped to the final recovery result of the power data through the fully connected layer; the calculation formula of the fully connected layer is:

[0070] R=HW h +b h

[0071] Where R is the recovery matrix, which contains the complete estimation results of all nodes in the power system; W h is the weight matrix of the fully connected layer, which is used to project the feature matrix H into the output space of data recovery; b h It is a bias term that adjusts the output result.

[0072] It can be seen from the above technical scheme that the method and system for recovering abnormal data of a power system taking time and accuracy into consideration provided by the embodiment of the present invention first obtains power data to determine whether there are abnormal values; based on the safety protection level index and the response recovery level index, the quantitative values ​​of each abnormal value in two dimensions are obtained to form a row vector, and a decision indicator matrix is ​​further established; the weight matrix is ​​determined by the hierarchical analysis method; the comprehensive score of each abnormal value is calculated based on the decision indicator matrix and the weight matrix; the priority label is determined according to the comprehensive score and the threshold, and the label processing is performed for each abnormal value in the power data set to obtain a data set with a priority label; the recovery algorithm corresponding to the priority label is used to recover each abnormal value in the data set to obtain a normal data set; the recovery algorithm includes a high-precision recovery algorithm and a fast recovery algorithm; the normal data set is sent to the data control center to provide a data source for the subsequent real-time decision analysis of the power system. The present invention can realize effective data recovery in a short time by decomposing the data matrix and quickly reconstructing the abnormal data. The recovered data will be sent to the data control center to provide support for further analysis and decision-making of the power system. The present invention can effectively balance the accuracy and response speed in the process of abnormal data recovery of the power system, and provide reliable data support for the stable operation and real-time decision-making of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The present invention is a flow chart of a method for recovering abnormal data in an electric power system taking into account the priorities of time and accuracy.

[0074] Figure 2 This is a structural diagram of the power system abnormal data recovery system that takes time and accuracy priorities into consideration according to the present invention. DETAILED DESCRIPTION

[0075] The technical scheme and technical effects of the present invention are further elaborated in detail below in conjunction with the accompanying drawings of the present invention.

[0076] Power data usually includes production data and management data. The production data transmission network is mainly used for services directly related to the real-time operation and control of the power system, such as dispatching automation, equipment monitoring and fault handling. This type of network has high requirements for the recovery speed of data transmission, because data delay or loss may directly affect the safe and stable operation of the system. In order to ensure that the system can quickly return to normal when data anomalies or transmission interruptions occur, the data recovery of the production data transmission network needs to take rapid response as the core goal, give priority to reducing the recovery time, so as to rebuild a reliable data flow in the shortest time and ensure the continuity and security of the system. In contrast, the management data transmission network is mainly used for long-term planning, data analysis and business management. The data of this type of network mostly involves application scenarios such as load forecasting, energy efficiency evaluation and historical data analysis, and has lower requirements for data recovery speed. However, these data play a vital role in power system planning, decision support and optimal dispatching, and their data accuracy directly affects the reliability of analysis results and the accuracy of decision-making. Therefore, for the management data transmission network, data recovery needs to give priority to ensuring the integrity and accuracy of the data, that is, through refined processing and multiple verifications during the recovery process, ensure that the recovered data meets high-precision requirements and supports long-term data analysis and planning work. Therefore, there are obvious differences in the security protection level requirements for data in the two types of networks, and their respective characteristics also have different requirements for data recovery technology.

[0077] The present invention proposes a strategic data recovery method, which adopts a technical architecture for abnormal data recovery of power systems that considers the priority of time and accuracy. First, the priority judgment module is used to quantify the security and real-time requirements of the data according to the security protection level and response recovery level of the data, generate a comprehensive score and determine the priority label. The data recovery module selects a suitable recovery algorithm according to the priority. The high-precision recovery process uses the temporal convolutional network (TCN) and the graph convolutional network (GCN) to extract spatiotemporal features, while the fast response process uses low-rank matrix decomposition and ADMM optimization to achieve fast recovery and output complete results.

[0078] refer to Figure 1 and Figure 2 As shown, the abnormal data recovery method of the power system considering time and accuracy priority provided by the present invention is implemented by Figure 2 The specific implementation of the system and data recovery method includes:

[0079] Step S1, obtain the power data set and determine whether there are abnormal values; abnormal values ​​include telecontrol information exchanged between the data center SCADA-EMS system and the plant station, data information exchanged between the data center SCADA and EMS system, electric energy metering information, and other power market information. Other power market information includes power generation cost, market clearing price, transaction volume and user load; SCADA (Supervisory Control and Data Acquisition) and EMS (Energy Management System) are usually used in conjunction to jointly undertake power grid regulation and management.

[0080] Step S2: Based on the security protection level indicator U r and response recovery level index U p The quantized values ​​of each outlier in two dimensions form a row vector A ij , and further establish the decision indicator matrix A for all outliers, where A ij middle element a ij It represents the quantitative value of the i-th outlier under the j-th level indicator;

[0081] Step S3, using the analytic hierarchy process to determine the weight matrix W = [w1, w2] T , including the security protection level weight w1 and the response recovery level weight w2;

[0082] Step S4: Calculate the comprehensive score Z of each outlier based on the decision indicator matrix A and the weight vector W i ;

[0083] Step S5, based on the comprehensive score Z i And the threshold Z′ determines the priority label Priority i , labeling is performed for each outlier in the power data set to obtain a data set with priority labels; the priority labels correspond to the recovery algorithm of the outliers;

[0084] Step S6, using a recovery algorithm corresponding to the priority label to recover each outlier in the data set with the priority label, and obtaining a normal data set after data recovery; the recovery algorithm includes a high-precision recovery algorithm and a fast recovery algorithm; the fast recovery algorithm uses a low-rank matrix decomposition algorithm for data recovery, and the high-precision recovery algorithm uses a spatiotemporal convolutional network STCN for data recovery, and the spatiotemporal convolutional network STCN is composed of a temporal convolutional network TCN, a graph convolutional network GCN and a fully connected layer;

[0085] Step S7, sending the normal data set to the data control center to provide a data source for subsequent power system analysis.

[0086] The quantitative standard of the optimal safety protection level index is: when the data type of the abnormal value belongs to the telecontrol information exchanged between the data center SCADA / EMS system and the plant station, the quantitative value is 4; when it belongs to the data information exchanged between the data center SCADA / EMS system, the quantitative value is 3; when it belongs to the electric energy metering information, the quantitative value is 2; when it belongs to other information in the power market, the quantitative value is 1;

[0087] The quantitative standard of the response recovery level indicator is: when the allowable response time for data recovery of abnormal values ​​is immediate response, the quantitative value is 5; when the allowable response time for data recovery is less than 1s, the quantitative value is 4; when the allowable response time for data recovery is less than 1min, the quantitative value is 3; when the allowable response time for data recovery is less than 5min and greater than 1min, the quantitative value is 2; when the allowable response time for data recovery is greater than 5min, the quantitative value is 1.

[0088] Table 1 Safety protection level indicators

[0089]

[0090] Table 2 Recovery response level indicators

[0091]

[0092]

[0093] An optional implementation of step S2 is:

[0094] Step S21: derive the quantitative value a of the i-th abnormal value in the dimension of the security protection level index based on the security protection level index. i1 ; Based on the response recovery level index, the quantitative value a of the i-th abnormal value in the response recovery level index dimension is obtained i2 ;

[0095] Step S22, let row vector A ij =[a i1 ,a i2 ], the i row vectors form the decision indicator matrix A:

[0096]

[0097] Among them, i∈[1,n]; through this matrix, the score of each data on these two decision indicators can be centrally represented;

[0098] Step S23: In order to ensure the comparability of different indicators in calculation, the matrix A is standardized to obtain the decision indicator matrix A′:

[0099]

[0100] In the formula, A j represents the column vector in matrix A; min(A j ) and max(A j ) represent the minimum and maximum values ​​in column j, respectively. The purpose of normalization is to map the scores of different indicators to the same numerical range (usually 0 to 1) and eliminate the dimensional differences between different indicators. For example, the security protection score may be between 1 and 5, while the response recovery score may be between 1 and 10.

[0101] After completing the standardization process, the analytic hierarchy process (AHP) is used to determine the weight of each indicator, namely the safety protection level weight w1 and the response recovery level weight w2. The analytic hierarchy process is a structured decision-making method that compares the importance of each indicator by constructing a judgment matrix to obtain a reasonable weight distribution. An optional implementation of step S3 is:

[0102] Construct the judgment matrix P based on expert opinions or experience:

[0103]

[0104] Among them, p 11 Indicates that the importance of the security protection index relative to itself is 1, p 12 Indicates the importance of security protection relative to response and recovery, p 12 ∈[1,9], the larger the value, the more important the security protection is. Let p 21 =1 / p 12 、p 22 =1, p 21 Indicates the importance of response recovery relative to security protection, p 22 Indicates that the importance of response recovery relative to itself is 1;

[0105] Each element of the judgment matrix P is normalized to calculate the relative weight of each indicator and obtain the normalized judgment matrix P′:

[0106]

[0107] Among them, the element p′ ij The calculation method is:

[0108]

[0109] Here, let n = 2;

[0110] Each row of the judgment matrix P′ is averaged to obtain the weight of each indicator:

[0111]

[0112] Get the security protection level weight w1 and response recovery level weight w2, as well as the weight matrix W = [w1, w2] T .

[0113] The specific implementation of step S4 includes: by multiplying the decision indicator matrix A′ and the weight matrix W, calculating the comprehensive score Z of each outlier i :

[0114]

[0115] Among them, Z i It represents the comprehensive score of the i-th outlier. The comprehensive score Z provides the basis for the subsequent recovery algorithm selection.

[0116] Priority tag in step S5 i The generated rules are:

[0117]

[0118] Among them, Priority i A value of 1 indicates that the fast recovery algorithm is selected. i A value of 0 indicates that a high-precision recovery algorithm is selected.

[0119] Finally, the abnormal data set with priority labels is exported to generate a large abnormal data set with clear priority identification. This data set provides clear decision support for the system in the subsequent abnormal data recovery process, thereby ensuring the real-time and accuracy of data recovery.

[0120] In data recovery, different algorithms will affect the accuracy and speed of recovery. Temporal Convolutional Network (TCN) is a deep learning method specifically designed for processing time series data, which can capture the long-term dependencies and complex temporal patterns of data. On this basis, Graph Convolutional Network (GCN) further models the spatial dependencies of data, which is particularly suitable for data with topological structures. GCN uses the connection relationship between nodes to capture the dependencies between nodes through graph convolution operations, further improving the accuracy of recovery. The combination of TCN and GCN can simultaneously enhance the accuracy and spatial consistency of time series data, but it also brings higher computational complexity and longer processing time. Therefore, although this combination can achieve extremely high recovery accuracy, its time overhead is relatively large.

[0121] In contrast, matrix decomposition is an optimization method based on linear algebra that approximates the recovery of missing or abnormal data by decomposing the data matrix into a low-rank form. Matrix decomposition has low computational complexity and can converge quickly through iterative optimization, so data recovery can be completed in a shorter time. Although low-rank matrix decomposition may be slightly inferior to deep learning models in terms of accuracy, its fast computing characteristics make it very suitable for scenarios with high requirements for recovery time.

[0122] The specific implementation of step S6 includes:

[0123] For the priority tag Priority i For outliers with a value of 1, a low-rank matrix decomposition algorithm is used for rapid recovery:

[0124] Let the data matrix Y represent the matrix of the original data of the power data set (including outliers and missing values), and establish a mask matrix W about the data matrix Y, where the element w in the mask matrix W is ij Represents the element y in the data matrix Y ij The effectiveness is expressed as:

[0125] This information may be abnormal or missing. In order to handle missing and abnormal data, a mask matrix W is introduced to mark the validity of the data:

[0126]

[0127] The mask matrix W ensures that the optimization objective function only focuses on valid data when decomposing, and ignores missing or abnormal data. Here, the method of determining whether the data is normal data refers to the existing technical solution, which will not be elaborated here.

[0128] Establish low-rank matrices U and V, and use the optimization objective function to adjust the reconstruction matrix X = UV so that the reconstruction matrix X is close to the valid data in Y; the optimization objective function is:

[0129]

[0130] In the formula, w ij (Y ij -(UV) ij ) 2 Indicates that only valid data positions (i.e. W ij =1) calculate the error, limit the size of U and V, and avoid overfitting; represents the regularization term; λ represents the regularization parameter, which is used to control the weight of the regularization term;

[0131] In order to simplify the solution process, auxiliary variables Z and Lagrange multiplier matrix Λ are introduced. The optimization objective function is decomposed into two sub-problems by alternating direction multiplier method ADMM. X and Z are optimized respectively, and X, Z and Λ are iteratively updated:

[0132] First, fix Z and Λ, optimize the value of X, and consider the current values ​​of Z and Λ. The iterative formula of X is:

[0133]

[0134] Where k is the number of iterations, and ρ is the penalty parameter control, which is used to control the consistency between X and Z;

[0135] Then optimize the auxiliary variable Z so that Z is more consistent with the current X. The optimization formula for Z is:

[0136]

[0137] g(Z) makes Z better meet the current optimization conditions;

[0138] Update the Lagrange multiplier matrix Λ to meet the asymptotic convergence of the constraints. The update formula of Λ is:

[0139] Λ (k+1) =Λ k +ρ(X (k+1) -Z (k+1) ) (17)

[0140] Through multiple iterations, X and Z will gradually converge, and the final reconstructed matrix X = UV will approach the valid data part of the original matrix Y, and fill in the missing data and repair the abnormal data. In the final reconstructed matrix X, the missing and abnormal data positions will be reasonably filled, and the filled values ​​are based on the low-rank inference of the valid data in Y.

[0141] Calculate the error between the valid data positions in the reconstructed matrices X and Y ∥ (X (k+1) -Y)⊙W∥ F , when ∥(X (k +1) -Y)⊙W∥ F <ε, where ε is the preset threshold; ⊙ represents element-by-element multiplication, W is the mask matrix, and only valid data positions (w ij =1) will calculate the error;

[0142] Using the final reconstructed matrix X = UV, a low-rank inference method is used to fill in missing data and repair abnormal data based on valid data in Y;

[0143] Due to its particularity, power system data has significant temporal and spatial characteristics. It not only changes regularly over time, but is also affected by the mutual influence between nodes in different geographical locations and physical connections in the system. This characteristic requires the use of a deep learning structure that can simultaneously capture the characteristics of data in both time and space dimensions to improve the accuracy of recovery of abnormal data. To this end, the present invention introduces a temporal convolutional network (TCN) to capture the time-dependent characteristics of data, and combines a graph convolutional network (GCN) to extract the structured characteristics of data in the spatial dimension.

[0144] For the priority tag Priority i Outliers with a value of 0 are restored with high accuracy using the spatiotemporal convolutional network (STCN). First, the time series data is processed using TCN to capture its long-term dependencies and extract key time series features. TCN enlarges the receptive field through dilated convolution operations to ensure that the long-term patterns of the data are modeled.

[0145] Assume that the input time series of power data in the power data set is Q = {q1, q2, q3, ..., q t}, where q t represents the power data collected at time t;

[0146] The convolution operation of TCN is defined as:

[0147]

[0148] In the formula, z t represents the temporal features extracted by TCN at time t, which is used to capture the temporal dependency of the input sequence; w i is the weight of the convolution kernel, which controls the contribution of each input point in the convolution; k is the size of the convolution kernel, which indicates the number of time points involved in a single convolution operation; d i is the expansion rate, which is used to control the step size of the convolution kernel in the time series, so that the convolution kernel can skip several time points to achieve dependency capture in a longer time range; b is the bias term, which adjusts the output of the convolution; the above formula is the difference between the input data corresponding to the current time t and the previous time td i Convolution operation is performed on the data; t>d k-1 ; After the dilated convolution operation of multiple layers of TCN, the output feature sequence Z is obtained t ={z t1 ,z t2 ,z t3 ,...,z tn} can fully characterize the temporal characteristics of the input time series; let the set Represents a set composed of elements selected from set Q Taken from the collection

[0149] Using Z t Further construct the initial feature matrix H (0) =[z t1 ,z t2 ,z t3 ,...,z tn ] as the input of the graph convolutional network GCN;

[0150] The graph convolutional network (GCN) is used to extract the spatial features of the data to reflect the physical connection relationship between nodes in the power system. The topological structure of the power system is abstractly represented as a graph G = (V, E), where V is a set of nodes, representing measurement points in the power system (such as substations, power stations, etc.), and each node corresponds to an observation location; E is a set of edges, representing the physical connection between measurement points; the adjacency matrix A describes the connection relationship between nodes: if the nodes are connected, then A = 1; otherwise A = 0; the degree matrix D is used to represent the number of connections for each node; the graph convolution operation of GCN realizes the extraction of spatial features by fusing the information of adjacent nodes layer by layer. The graph convolution calculation formula of the l+1 layer is:

[0151]

[0152] In the formula, H (l+1) is the node feature matrix of the l+1th layer, initially H (0) is the time series feature matrix extracted by TCN; W (l) is the weight matrix of the lth layer, which is used to learn the characteristic relationship between nodes; σ is the activation function, which introduces nonlinear transformation so that the model can more effectively express complex spatial structural relationships; through the layer-by-layer operation of graph convolution, the node features are propagated and integrated in the topological structure of the power system to generate a node feature matrix H containing spatiotemporal information (L) As the final output matrix H of the graph convolutional network, the final output matrix H contains the temporal and spatial features of each node; through the spatial feature extraction of GCN, the temporal and spatial dependency of the data is enhanced, thereby improving the recovery accuracy;

[0153] Finally, the final output matrix H is mapped to the final recovery result of the power data through the fully connected layer; the calculation formula of the fully connected layer is:

[0154] R=HW h +b h (20)

[0155] Where R is the recovery matrix, which contains the complete estimation results of all nodes in the power system; Wh is the weight matrix of the fully connected layer, which is used to project the feature matrix H into the output space of data recovery; b h is a bias term that adjusts the output result. The output of the fully connected layer is the result of high-precision recovery.

[0156] The present invention optimizes resource allocation efficiency by executing priority judgment and reasonably dividing the selection priority of data recovery according to the security and real-time requirements of the data. For data with high real-time requirements, matrix decomposition is used to achieve rapid recovery, shortening the response time of data recovery and ensuring the continuity and stability of the system. For data with high precision requirements, a spatiotemporal convolutional network (STCN) is used to combine time and space features for high-precision recovery, ensuring the accuracy and integrity of the data, thereby improving the quality of data recovery.

[0157] The multiple, multi-layer, and multiple times mentioned in this solution refer to a number greater than 1, and the specific number is given in the solution based on actual needs.

[0158] What is disclosed above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A method for recovering abnormal data of a power system taking into account time and accuracy priority, characterized in that: include: Step S1, obtaining a power data set and determining whether there are abnormal values; The abnormal values ​​include telecontrol information exchanged between the data center SCADA-EMS system and the plant station, data information exchanged between the data center SCADA and EMS system, electric energy metering information, and other information in the power market, wherein the other information in the power market includes power generation cost, market clearing price, transaction volume, and user load; Step S2: based on the safety protection level index and the response recovery level index, the quantized values ​​of each abnormal value in two dimensions are obtained to form a row vector A. ij , and further establish the decision indicator matrix A for all outliers, where A ij middle element a ij It represents the quantitative value of the i-th outlier under the j-th level indicator; Step S3, using the analytic hierarchy process to determine the weight matrix W = [w1, w2] T , including the security protection level weight w1 and the response recovery level weight w2; Step S4, calculating the comprehensive score Z of each outlier based on the decision indicator matrix A and the weight vector W i ; Step S5, according to the comprehensive score Z i And the threshold Z′ determines the priority label Priority i , labeling each of the abnormal values ​​in the power data set to obtain a data set with priority labels; the priority labels correspond to a recovery algorithm for the abnormal values; Step S6, using a recovery algorithm corresponding to the priority label to recover each of the abnormal values ​​in the data set with the priority label, and obtaining a normal data set after data recovery; the recovery algorithm includes a high-precision recovery algorithm and a fast recovery algorithm; the fast recovery algorithm uses a low-rank matrix decomposition algorithm for data recovery, and the high-precision recovery algorithm uses a spatiotemporal convolutional network STCN for data recovery, and the spatiotemporal convolutional network STCN is composed of a temporal convolutional network TCN, a graph convolutional network GCN and a fully connected layer; Step S7, sending the normal data set to the data control center to provide a data source for subsequent power system analysis.

2. The method for recovering abnormal data of a power system taking into account time and accuracy priority as claimed in claim 1, characterized in that: The quantitative standard of the safety protection level indicator is: when the data type of the abnormal value belongs to the telecontrol information exchanged between the data center SCADA / EMS system and the plant station, the quantitative value is 4; when it belongs to the data information exchanged between the data center SCADA / EMS system, the quantitative value is 3; when it belongs to the electric energy metering information, the quantitative value is 2; when it belongs to other information of the power market, the quantitative value is 1; The quantification standard of the response recovery level indicator is: when the allowed response time for data recovery of the abnormal value is an immediate response, the quantization value is 5; when the allowed response time for data recovery is less than 1s, the quantization value is 4; when the allowed response time for data recovery is less than 1min, the quantization value is 3; when the allowed response time for data recovery is less than 5min and greater than 1min, the quantization value is 2; when the allowed response time for data recovery is more than 5min, the quantization value is 1.

3. The method for recovering abnormal data of a power system taking into account time and accuracy priority as claimed in claim 2, characterized in that: The step S2 comprises: Step S21: derive the quantitative value a of the i-th abnormal value in the dimension of the security protection level index based on the security protection level index. i1 ; Based on the response recovery level index, the quantitative value a of the i-th abnormal value in the response recovery level index dimension is obtained i2 ; Step S22, let row vector A ij =[a i1 ,a i2 ], i row vectors form the decision indicator matrix A: Where i∈[1,n]; Step S23, normalize the matrix A to obtain the decision indicator matrix A′: In the formula, A j represents the column vector in matrix A; min(A j ) and max(A j ) represent the minimum and maximum values ​​in column j, respectively.

4. The method for recovering abnormal data of a power system taking into account time and accuracy priority as claimed in claim 3, characterized in that: The step S3 comprises: Construct the judgment matrix P: Among them, p 11 Indicates that the importance of the security protection index relative to itself is 1, p 12 Indicates the importance of security protection relative to response and recovery, p 12 ∈[1,9], let p 21 =1 / p 12 、p 22 =1, p 21 Indicates the importance of response recovery relative to security protection, p 22 Indicates that the importance of response recovery relative to itself is 1; Perform column normalization on each element of the judgment matrix P to obtain a normalized judgment matrix P′: Among them, the element p′ ij The calculation method is: Perform average processing on each row of the judgment matrix P′: Get the security protection level weight w1 and response recovery level weight w2, as well as the weight matrix W = [w1, w2] T .

5. The method for recovering abnormal data of a power system taking into account the priority of time and accuracy as claimed in claim 4, characterized in that: The step S4 comprises: By multiplying the decision indicator matrix A′ with the weight matrix W, the comprehensive score Z of each outlier is calculated. i : Among them, Z i represents the comprehensive score of the i-th outlier.

6. The method for recovering abnormal data of a power system taking into account the priority of time and accuracy as claimed in claim 5, characterized in that: The priority label Priority in step S5 i The generated rules are: Among them, Priority i A value of 1 indicates that the fast recovery algorithm is selected. i A value of 0 indicates that a high-precision recovery algorithm is selected.

7. The method for recovering abnormal data of a power system taking into account the priority of time and accuracy as claimed in claim 6, characterized in that: The step S6 comprises: For the priority tag Priority i For outliers with a value of 1, a low-rank matrix decomposition algorithm is used for rapid recovery: Let the data matrix Y represent the power data set, and establish a mask matrix W about the data matrix Y, where the element w in the mask matrix W is ij Represents the element y in the data matrix Y ij The effectiveness is expressed as: Establish low-rank matrices U and V, and use the optimization objective function to adjust the reconstruction matrix X=UV so that the reconstruction matrix X approaches the valid data in Y; the optimization objective function is: In the formula, w ij (Y ij -(UV) ij ) 2 Indicates that the error is calculated only at valid data locations; represents the regularization term; λ represents the regularization parameter; Auxiliary variables Z and Lagrange multiplier matrix Λ are introduced, and the optimization objective function is decomposed into two sub-problems through the alternating direction multiplier method ADMM, optimizing X and Z respectively, and iteratively updating X, Z and Λ: First, fix Z and Λ, optimize the value of X, and consider the current values ​​of Z and Λ. The iterative formula of X is: In the formula, k is the number of iterations, and ρ is the penalty parameter control; Then optimize the auxiliary variable Z, the optimization formula of Z is: Update the Lagrange multiplier matrix Λ, the update formula of Λ is: L (k+1) =L k +ρ(X (k+1) -Z (k+1) ) Calculate the error between the valid data locations in the reconstructed matrices X and Y ‖(X (k+1) -Y)⊙W‖ F , when ‖(X (k+1) -Y)⊙W‖ F <ε, where ε is the preset threshold. Using the final reconstructed matrix X = UV, a low-rank inference method is used to fill in missing data and repair abnormal data based on valid data in Y; Further, For the priority tag Priority i The outliers with a value of 0 are restored with high precision using the spatiotemporal convolutional network STCN: Assume that the input time series of the power data in the power data set is Q = {q1, q2, q3, ..., q t }, where q t represents the power data collected at time t; The convolution operation of TCN is defined as: In the formula, z t represents the temporal features extracted by TCN at time t; w i is the weight of the convolution kernel; k is the size of the convolution kernel, indicating the number of time points involved in a single convolution operation; d i is the expansion rate; b is the bias term; t>d k-1 ; After the dilated convolution operation of multiple layers of TCN, the output feature sequence Z is obtained t ={z t1 ,z t2 ,z t3 ,...,z tn }; Using Z t Further construct the initial feature matrix H (0) =[z t1 ,z t2 ,z t3 ,...,z tn ] as the input of the graph convolutional network GCN; The spatial features of the data are extracted using the graph convolutional network GCN. The topological structure of the power system is abstractly represented as a graph G = (V, E), where V is a set of nodes, representing the measurement points in the power system; E is a set of edges, representing the physical connection between the measurement points; the adjacency matrix A describes the connection relationship between the nodes: if the nodes are connected, then A = 1; otherwise A = 0; the degree matrix D is used to represent the number of connections for each node; the graph convolution calculation formula for the l+1th layer is: In the formula, H (l+1) is the node feature matrix of the l+1th layer, initially H (0) is the time series feature matrix extracted by TCN; W (l) is the weight matrix of the lth layer; σ is the activation function; through the layer-by-layer operation of graph convolution, the node features are propagated and integrated in the topological structure of the power system to generate a node feature matrix H containing spatiotemporal information (L) As the final output matrix H of the graph convolutional network, the final output matrix H contains the temporal and spatial features of each node; Finally, the final output matrix H is mapped to the final recovery result of the power data through the fully connected layer; the calculation formula of the fully connected layer is: R=HW h +b h Where R is the recovery matrix, which contains the complete estimation results of all nodes in the power system; W h is the weight matrix of the fully connected layer, which is used to project the feature matrix H into the output space of data recovery; b h It is a bias term that adjusts the output result.

8. A power system abnormal data recovery system that takes time and accuracy into consideration, characterized in that: Used to implement the method according to any one of claims 1 to 7.

9. An electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images