Abnormal detection and replacement method and system for off-line electric quantity data of power distribution network

Abstract

The invention relates to the technical field of novel power distribution network anomaly detection, and particularly discloses an anomaly detection and replacement method and system for off-line electric quantity data of a power distribution network, and the method comprises the steps: obtaining the off-line electric quantity data of the power distribution network in a historical time period, carrying out the preprocessing, inputting the preprocessed data into a trained data time sequence prediction model, obtaining off-line electric quantity prediction data of the power distribution network at each moment in the to-be-measured time period; obtaining off-line electric quantity real data of the power distribution network in a to-be-measured time period, calculating a residual term of the real data and the prediction data at each moment, clustering the obtained residual terms, and identifying abnormal data points; replacing the abnormal data points by using the electric quantity prediction data at the corresponding moment; according to the method, the input form of residual data is remodeled, and abnormal data can be effectively detected and replaced.

Description

Technical Field

[0001] The present invention relates to the technical field of novel distribution network anomaly detection, and in particular to a method and system for anomaly detection and replacement of offline power data of a distribution network. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the high proportion of new source-load equipment such as photovoltaics and electric vehicles connected, the new distribution network has shown a trend of load diversification and intelligence. A large number of power electronic equipment has increased the complexity and randomness of the flow of electric energy in the power grid. Obtaining reliable electricity data is an important part of ensuring the transparency of the distribution network.

[0004] The offline power data of the distribution network mainly includes three-phase voltage, three-phase current and power data. The quality of the current power data of the distribution network is threatened by many factors. First, some new energy equipment is affected by factors such as sunshine and weather, the data of equipment such as electric vehicles is affected by user habits, and some electrical loads are affected by seasons. These factors will increase the randomness of power data in the distribution network. In addition, a large amount of power data will also increase the pressure faced by data during collection and transmission, causing abnormal phenomena such as mutations, missing data, and long-term fluctuations. Therefore, the detection and replacement of abnormal power data is one of the important functions of state estimation and power dispatching in the power system, which can ensure the accurate identification of the operating status of the distribution network.

[0005] In power systems, measurement data mainly consists of measurement values ​​and measurement errors. Abnormal data refers to data that deviates from the normal measurement range, which may be caused by various reasons such as measurement equipment failure or data transmission problems.

[0006] Traditional anomaly detection methods based on mean, standard deviation, etc., are affected by data processing capabilities or efficiency, and often ignore the importance of redundant information and focus on single features or simple statistics. Therefore, traditional anomaly detection methods based on statistical analysis obviously do not make enough use of redundant information and are difficult to deal with problems such as residual pollution and residual drowning.

[0007] In terms of data prediction and replacement, most of the difference methods currently used are Lagrange interpolation replacement, spline interpolation, etc., but these methods are computationally intensive, have high requirements for memory, and have slow calculation speeds.

[0008] The prior art also discloses the use of machine learning methods (such as random forests and support vector machines, etc.) to predict electricity data and then replace abnormal data; however, the random forest method is not suitable for situations with more data, and the support vector machine method also has common problems such as difficulty in determining hyperparameters and prone to underfitting or overfitting, resulting in poor generalization ability of data and insufficient prediction accuracy. Summary of the invention

[0009] In order to solve the above problems, the present invention proposes a method and system for detecting and replacing abnormalities in offline power data of distribution networks. The time series prediction model of the improved extreme gradient boosting algorithm (XGboost) is used to predict the regional power consumption sequence data moment by moment. By improving the determination process of the decision tree, the step size factor is introduced to improve the calculation method of the objective function, avoiding the limitations of the traditional method in the process of formulating the decision tree, and avoiding overfitting. Finally, the residuals based on the predicted and true values ​​are clustered using the density clustering algorithm (DBSCAN), and the form of the residual data before being input into the clustering algorithm is reconstructed to detect and replace abnormal data.

[0010] In some embodiments, the following technical solutions are adopted:

[0011] A method for detecting and replacing abnormality of offline power data in a distribution network, comprising:

[0012] Obtain offline power data of the distribution network in the historical time period and preprocess it. The preprocessed data is input into the trained data time series prediction model to obtain offline power prediction data of the distribution network at each moment in the time period to be tested.

[0013] Obtain the real data of offline power consumption of the distribution network during the test period, calculate the residual term between the real data and the predicted data at each moment, cluster the obtained residual terms, and identify abnormal data points;

[0014] Replace abnormal data points with power forecast data at the corresponding time;

[0015] The data time series prediction model is constructed by using an improved XGboost algorithm, and the improved XGboost algorithm introduces a step factor between each layer of the decision tree.

[0016] Furthermore, the offline power data of the distribution network is voltage, current or active power.

[0017] Furthermore, the objective function of each decision tree in the improved XGboost algorithm is the sum of the loss function and the regularization term, and regularization is performed in each iteration.

[0018] Furthermore, the objective function of each decision tree in the improved XGboost algorithm is based on the leaf node value w j is a quadratic function of the variable, for w j After taking the partial derivative, we can find the extreme value of the objective function and obtain the simplest objective function and the corresponding optimal weight. Form the weight vector w.

[0019] Furthermore, the simplified objective function obj of each decision tree (t)* and the optimal weight Specifically:

[0020]

[0021] Among them, G j and H j are all fixed values, T represents the number of leaf nodes of the tree after t iterations, and λ represents the penalty parameter of the leaf node value.

[0022] Furthermore, taking every three layers as an optimization unit, the benefits of each layer are weighted summed to determine the split benefits of the decision tree in the improved XGboost algorithm;

[0023] Starting from the tree depth of 0, the leaf nodes are continuously divided. When the K layer is reached, if the splitting benefit meets the set requirements, the decision tree at this time is taken as the optimal decision tree, and the number of leaf nodes T and the value of the leaf nodes are obtained.

[0024] Furthermore, the obtained residual items are clustered, specifically:

[0025] The DBSCAN clustering algorithm is used to perform cluster analysis on the residual term R, dividing the data into several clusters and outliers. These clusters are aggregated from normal residual term data, and the outliers are the abnormal data identified by the DBSCAN clustering algorithm.

[0026] In other embodiments, the following technical solutions are adopted:

[0027] A distribution network offline power data anomaly detection and replacement system, comprising:

[0028] The power data prediction module is used to obtain the offline power data of the distribution network in the historical time period and preprocess it. The preprocessed data is input into the trained data time series prediction model to obtain the offline power prediction data of the distribution network at each moment in the time period to be tested;

[0029] The abnormal data identification module is used to obtain the real data of offline power of the distribution network during the test period, calculate the residual term between the real data and the predicted data at each moment, cluster the obtained residual terms, and identify abnormal data points;

[0030] An abnormal data replacement module is used to replace abnormal data points with power forecast data at the corresponding moment;

[0031] The data time series prediction model is constructed by using an improved XGboost algorithm, and the improved XGboost algorithm introduces a step factor between each layer of the decision tree.

[0032] In other embodiments, the following technical solutions are adopted:

[0033] A terminal device comprises a processor and a memory, wherein the processor is used to implement instructions; the memory is used to store a plurality of instructions, wherein the instructions are suitable for being loaded by the processor and executing the above-mentioned method for detecting and replacing abnormalities in offline power data of a distribution network.

[0034] In other embodiments, the following technical solutions are adopted:

[0035] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the above-mentioned method for detecting and replacing abnormality of offline power data of a distribution network.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) The present invention predicts the time series data three months later based on the prediction model obtained by training in the first three months to form a prediction sequence. The residual term at each moment is obtained by subtracting the true value at each moment from the predicted value at the corresponding moment. The residual terms of the true value and the predicted value are clustered and analyzed using the improved DBSCAN clustering density algorithm to detect abnormal values, and then the abnormal values ​​are replaced by the corresponding predicted values; the method of the present invention reshapes the input form of the residual data and can effectively detect and replace abnormal data.

[0038] (2) The present invention constructs a data time series prediction model based on the improved XGboost algorithm. The improved XGboost algorithm introduces a step factor between each layer of the decision tree, avoiding the limitations of the traditional method in the process of formulating the decision tree and avoiding the overfitting phenomenon. The loss function is expanded by a second-order Taylor, and the decision tree function f is defined as a function of the leaf node value w. This can simplify the objective function, reduce the amount of calculation, and can be applied to different error functions to improve the calculation efficiency.

[0039] (3) The present invention improves the calculation method of splitting benefits. It takes three layers as a unit, performs weighted summation of benefits of different layers, and determines splitting benefits, thereby avoiding excessive local benefits caused by chance.

[0040] Other features and advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a method for detecting and replacing abnormality of offline power data in a distribution network according to an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of a DBSCAN clustering algorithm in an embodiment of the present invention;

[0043] FIG. 3( a ) and FIG. 3( b ) are schematic diagrams of the residual R before and after coordinate adjustment, respectively, in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0046] Embodiment 1

[0047] In one or more embodiments, a method for detecting and replacing abnormalities in offline power data of a distribution network is disclosed, combining Figure 1 , specifically including the following process:

[0048] S101: Obtain offline power data of the distribution network in a historical time period and perform preprocessing, and input the preprocessed data into a trained data time series prediction model to obtain offline power prediction data of the distribution network at each moment in the time period to be tested.

[0049] Specifically, this embodiment obtains time series power data, such as voltage, current or power, from a distribution network area to form an overall data sample. For a certain type of data within a certain time period, this embodiment takes the active power data within three months as an example, and first establishes a data time series prediction model based on the improved XGboost algorithm.

[0050] This embodiment uses the active power data within three months as the training model, and the data does not contain abnormal values ​​and missing values. The data set is Z = {z1, z2, ..., z m}, representing a set of active power time series data of size 1*m.

[0051] Next, the dataset Z is reconstructed into the input dataset X and output dataset Y of the data time series prediction model.

[0052]

[0053] Among them, n represents the number of samples in the prediction model, τ represents the data window length used for training each sample, that is, the number of features, and the data in the range of (t-τ, t-1) can be used to predict the data at time t. And n+τ=m is satisfied.

[0054] This patent uses an improved XGBoost algorithm to build a data time series prediction model to achieve accurate prediction, so as to extract and replace outliers. After obtaining X and Y, the traditional XGBoost algorithm builds an additive model consisting of t decision trees, namely:

[0055]

[0056] In the formula, f k (x i ) represents the model prediction result of the kth decision tree, x i represents the i-th eigenvector in X, x i =(z i ,z i+1 ,…,z i+τ-1 ), is the output vector, i.e. the prediction result of the tth tree, and F represents the function space composed of all trees.

[0057] This embodiment improves the XGBoost algorithm as follows:

[0058] A step factor η is introduced between each layer of the decision tree, as follows:

[0059]

[0060] This step factor η can avoid overfitting and is set to 0.1 in this embodiment. Subsequent iterations of the XGboost algorithm rely on the calculation of the objective function for each tree. The objective function of this embodiment consists of the loss function of the model and the regularization term. The loss function can measure the accuracy of the model by the gap between the predicted value and the true value. The regularization term can be used to control the complexity of the model and prevent overfitting in the calculation. The regularization term Ω is directly added to the objective function and regularized in each iteration:

[0061]

[0062] Among them, obj (t) represents the objective function after the tth iteration, T represents the number of leaf nodes of the tree after t iterations, and w j Represents the value of the jth leaf node, y i is the true output value of the i-th sample. γ and λ represent the penalty parameters for leaf nodes and leaf node values, respectively, which are used to adjust the complexity of the decision tree. l is the error function, which is used to calculate the error between the true value and the predicted value.

[0063] Since the error calculation function l is selected differently for different situations, a calculation method that can adapt to all functions is needed. This embodiment performs a second-order Taylor expansion on the loss function and defines the decision tree function f as a function of the leaf node value w: f t (x) = w q(x) The objective function approximately changes to:

[0064]

[0065] Among them, I j ={i|q(x i )=j} is the set of samples divided into the jth leaf region, q(x i ) can represent the input vector x i The number of the leaf node assigned, that is, the corresponding j. All g i and h i are all constants. Therefore, the objective function is a function with node values ​​w j is a quadratic function of the variable, for w j After taking the partial derivative, we can find the extreme value of the objective function and obtain the simplified objective function and the corresponding optimal weight. Form the weight vector w.

[0066]

[0067] G j and H j Respectively is also a constant. Finally, the optimal weight value and objective function need to be optimized as follows:

[0068]

[0069] This embodiment is based on an improved greedy algorithm and takes every three layers as an optimization unit to determine the structure of the decision tree.

[0070] In the traditional Xboost algorithm, when looking for the optimal split point of the leaf node, a greedy algorithm is often used. For each feature, the training samples belonging to the node are sorted in ascending order according to the feature value, and the best split point of the feature is determined by linear scanning, and the split benefit of the feature is recorded:

[0071] gain (t) =obj (t-1)* -obj (t)* (8)

[0072] Among them, gain (t) It represents the loss reduction between the t-1 layer and the t layer of the decision tree, that is, the splitting benefit. In the traditional algorithm, the maximum gain is taken as the goal, and the best splitting method is selected for each leaf node of the previous layer, and finally the best tree structure is obtained step by step. However, the traditional method has limitations. It takes each layer as the optimization unit and can only focus on the benefits between two layers. It has limitations and cannot take into account the benefits generated by subsequent divisions.

[0073] Therefore, this patent improves the calculation method of splitting benefits, weights the benefits of different layers, and determines the structure of the tree with three layers as a unit.

[0074] gain * =a1×gain (t) +a2×gain (t-1) +a3×gain (t-2) (9)

[0075] Among them, a1, a2 and a3 represent weights respectively, and their sum is 1. Since the lower layers have a higher proportion, the weights are set to a1=0.5, a2=0.3, and a3=0.2 in this embodiment.

[0076] Starting from the tree depth of 0, the leaf nodes are continuously divided. When the K layer is reached, if gain * <10 -5 , which means that continuing to divide the leaf nodes has limited effect on reducing the prediction loss. The decision tree at this time can be regarded as the best decision tree. The number of leaf nodes T and the value of the leaf nodes are calculated, and finally according to the formula Calculate the predicted value.

[0077] Based on the active power time series data of the previous three months, the trained data time series prediction model can be used to predict the time series data three months later to form a prediction sequence; based on this, the distribution network power data of any time period to be tested can be predicted.

[0078] S102: Acquire the real data of offline power consumption of the distribution network during the time period to be tested, calculate the residual term between the real data and the predicted data at each moment, cluster the obtained residual terms, and identify abnormal data points.

[0079] The active power data within three months is used as the outlier screening object, and the data set to be processed is P = {p1, p2, ..., p D}, represents a set of active power time series data of size 1*D. P and Z in step S101 come from different time periods of the same measurement device, Z is used to train the model, and P is the data set to be processed.

[0080] Obtain the predicted active power value within three months, subtract it from the actual active power value, and calculate the residual term at the corresponding time.

[0081]

[0082] Among them, res d is the residual term at the dth moment, and p d are the predicted value and true value at the dth moment respectively, and all the residuals constitute the residual term vector R.

[0083] The DBSCAN clustering algorithm is used to perform cluster analysis on the residual term R, dividing the data into several clusters and outliers. These clusters are aggregated from normal residual term data, and the outliers are the abnormal data identified by the DBSCAN clustering algorithm.

[0084] Combination Figure 2 The DBSCAN clustering algorithm of this embodiment is as follows:

[0085] S1021: Select a sample point R in R that has not been visited i , check its set neighborhood range E ps All sample points in the neighborhood are core points if they contain at least Minpts sample points (Minpts is a set value), and they form a cluster with all points in the neighborhood.

[0086] S1022: Repeat S1021 for all sample points to determine whether each point meets the core point condition. If so, take each core point as the center and add all sample points in its neighborhood to the cluster of the center point.

[0087] S1023: If the sample point does not meet the core point condition, but is located in the neighborhood of another cluster, it is treated as a boundary point and added to the cluster. If the sample point does not meet the core point condition and is not in the neighborhood of a cluster, it is marked as an outlier.

[0088] The residual R in this embodiment is also time series data. A two-dimensional graph is drawn for the residual R data, with the time sequence number as the horizontal axis and the residual value as the vertical axis.

[0089] If the ordinate interval is not selected properly, it will lead to the inability to distinguish the outliers between the data, and the DBSCAN algorithm will fail. Therefore, this embodiment reconstructs the data set R for the residual between the predicted value and the true value of the power data, and draws a two-dimensional graph as input. In this embodiment, R is a 1*D row vector, and a ordinate Y is set for each data. d .

[0090]

[0091] Among them, d is the horizontal axis number in R, and the outlier value r is screened out in R o After that, the corresponding data p in P o Also considered as an outlier; R max and R min are the maximum and minimum values ​​in the data set R respectively; D represents the value of the horizontal axis.

[0092] Figure 3(a) and Figure 3(b) respectively show the original data before and after the adjustment of the ordinate. By adjusting the range of the ordinate, the failure of the density-based clustering algorithm caused by small data fluctuations and concentrated density is avoided.

[0093] S103: Using the power forecast data at the corresponding time Replace outlier data points.

[0094] This embodiment uses an improved XGboost time series prediction model to predict the regional electricity consumption sequence data moment by moment. By improving the determination process of the decision tree and introducing a step factor to improve the calculation method of the objective function, the limitations of the traditional method in the process of formulating the decision tree are avoided, and the overfitting phenomenon is avoided. Finally, DBSCAN clustering is performed based on the residuals of the prediction and the true value, and the form of the residual data before inputting into the clustering algorithm is reconstructed to detect and replace abnormal data.

[0095] Embodiment 2

[0096] In one or more embodiments, a system for detecting and replacing abnormality of offline power data in a distribution network is disclosed, comprising:

[0097] The power data prediction module is used to obtain the offline power data of the distribution network in the historical time period and preprocess it. The preprocessed data is input into the trained data time series prediction model to obtain the offline power prediction data of the distribution network at each moment in the time period to be tested;

[0098] The abnormal data identification module is used to obtain the real data of offline power of the distribution network during the test period, calculate the residual term between the real data and the predicted data at each moment, cluster the obtained residual terms, and identify abnormal data points;

[0099] An abnormal data replacement module is used to replace abnormal data points with power forecast data at the corresponding moment;

[0100] The data time series prediction model is constructed by using an improved XGboost algorithm, and the improved XGboost algorithm introduces a step factor between each layer of the decision tree.

[0101] The specific implementation of the above modules is the same as that in Example 1 and will not be described in detail.

[0102] Embodiment 3

[0103] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method for detecting and replacing abnormalities in offline power data of a distribution network in Embodiment 1 when executing the program. For the sake of brevity, it will not be described in detail here.

[0104] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0105] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0106] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0107] Embodiment 4

[0108] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the method for detecting and replacing anomaly of offline power data of a distribution network described in Embodiment 1.

[0109] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for detecting and replacing abnormalities in offline power data of a distribution network, characterized in that: include: Obtain offline power data of the distribution network in the historical time period and preprocess it. The preprocessed data is input into the trained data time series prediction model to obtain offline power prediction data of the distribution network at each moment in the time period to be tested. Obtain the real data of offline power consumption of the distribution network during the time period to be tested, calculate the residual term between the real data and the predicted data at each moment, cluster the obtained residual terms, and identify abnormal data points; Replace abnormal data points with power forecast data at the corresponding time; The data time series prediction model is constructed by using an improved XGboost algorithm, and the improved XGboost algorithm introduces a step factor between each layer of the decision tree.

2. A method for detecting and replacing abnormalities in offline power data of a distribution network according to claim 1, characterized in that: The offline power data of the distribution network is voltage, current or active power.

3. The method for detecting and replacing abnormality of offline power data in a distribution network according to claim 1, characterized in that: The objective function of each decision tree in the improved XGboost algorithm is the sum of the loss function and the regularization term, and regularization is performed in each iteration.

4. The method for detecting and replacing abnormality of offline power data in a distribution network according to claim 1, characterized in that: The objective function of each decision tree in the improved XGboost algorithm is based on the leaf node value w j is a quadratic function of the variable, for w j After taking the partial derivative, we can find the extreme value of the objective function and obtain the simplest objective function and the corresponding optimal weight. Form the weight vector w.

5. A method for detecting and replacing abnormalities in offline power data of a distribution network as claimed in claim 4, characterized in that: The simplified objective function obj of each decision tree (t)* and the optimal weight Specifically: Among them, G j and H j are all fixed values, T represents the number of leaf nodes of the tree after t iterations, and λ represents the penalty parameter of the leaf node value.

6. The method for detecting and replacing abnormality of offline power data in a distribution network according to claim 1, characterized in that: Taking every three layers as an optimization unit, the benefits of each layer are weighted summed to determine the split benefits of the decision tree in the improved XGboost algorithm; Starting from the tree depth of 0, the leaf nodes are continuously divided. When the K layer is reached, if the splitting benefit meets the set requirements, the decision tree at this time is taken as the optimal decision tree, and the number of leaf nodes T and the value of the leaf nodes are obtained.

7. A method for detecting and replacing abnormalities in offline power data of a distribution network as claimed in claim 1, characterized in that: The obtained residual items are clustered, specifically: The DBSCAN clustering algorithm is used to perform cluster analysis on the residual term R, dividing the data into several clusters and outliers. These clusters are aggregated from normal residual term data, and the outliers are the abnormal data identified by the DBSCAN clustering algorithm.

8. A distribution network offline power data anomaly detection and replacement system, characterized in that: include: The power data prediction module is used to obtain the offline power data of the distribution network in the historical time period and preprocess it. The preprocessed data is input into the trained data time series prediction model to obtain the offline power prediction data of the distribution network at each moment in the time period to be tested; The abnormal data identification module is used to obtain the real data of offline power of the distribution network during the test period, calculate the residual term between the real data and the predicted data at each moment, cluster the obtained residual terms, and identify abnormal data points; An abnormal data replacement module is used to replace abnormal data points with power forecast data at the corresponding moment; The data time series prediction model is constructed by using an improved XGboost algorithm, and the improved XGboost algorithm introduces a step factor between each layer of the decision tree.

9. A terminal device, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the method for detecting and replacing abnormality of offline power data in a distribution network as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instruction is suitable for being loaded by a processor of a terminal device and executing the method for detecting and replacing abnormality of offline power data of a distribution network as described in any one of claims 1-7.

Images