Abnormal power load data restoration method of adaptive improved dynamic PCC-FCM

Abstract

The invention discloses a self-adaptive improved dynamic PCC-FCM abnormal power load data restoration method, and relates to the technical field of smart power grids. Comprising the following steps: importing a power load data set with a time sequence, selecting a dynamic data set according to an abnormal power load data moment, and carrying out maximum value normalization processing; fCM clustering is improved based on the Pearson's correlation coefficient, and a PCC-FCM clustering algorithm is obtained; adopting a snake optimization algorithm to adaptively improve a PCC-FCM clustering algorithm, and clustering the dynamic data set to obtain each cluster; and performing feature extraction on each cluster to obtain a feature curve, and then repairing the abnormal power load data by adopting a curve replacement method. According to the method, the clustering algorithm is improved and optimized through the optimization algorithm, the abnormal load data of the power system can be effectively repaired, and the method has high abnormal data repairing precision.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and more specifically to an abnormal power load data repair method of an adaptively improved dynamic PCC-FCM. Background Art

[0002] With the continuous development of the intelligent power system, a large number of measurement and control terminals have been widely deployed, generating massive amounts of terminal data. However, since the power grid terminal meters or sensors may face force majeure factors such as insufficient configuration, measurement failures, and measurement errors, the data collected, including load data, usually have problems such as defaults and errors. Power load data is an important basis for the operation and dispatch of power systems. Abnormal load data will directly affect applications such as load forecasting, peak-shifting management, and abnormal power consumption detection in power systems. Therefore, studying the repair method of abnormal load data can improve the efficiency and accuracy of power supply and provide support for the safe and stable operation of the power grid.

[0003] At present, there are still relatively few studies on the repair of abnormal data in power systems. Commonly used methods mainly include horizontal and vertical comparison correction method, interpolation method, curve replacement method, statistical method and machine learning method. However, power load data has significant time series characteristics and is affected by multiple factors such as meteorological changes, seasonal factors and electricity price policies. It often shows basic properties such as periodicity, diversity and volatility, and exhibits strong curve characteristics. Therefore, the curve replacement method, also known as cluster analysis method, is more suitable for the repair of abnormal load data. This method extracts characteristic curves by clustering historical load curves and then repairs abnormal load data. Among them, the fuzzy mean clustering method solves the problems caused by the hard division of the traditional K-means clustering method by introducing the membership matrix, and can better express the structure and relationship between data. This method can effectively capture the dynamic changes of data and improve the accuracy of data repair, thereby ensuring the reliability of power system analysis and decision-making.

[0004] Therefore, it is an urgent problem for those skilled in the art to propose an abnormal power load data repair method of an adaptively improved dynamic PCC-FCM to solve the difficulties existing in the prior art. Summary of the invention

[0005] In view of this, the present invention provides a method for repairing abnormal power load data of adaptively improved dynamic PCC-FCM, by selecting a dynamic data set, using the snake optimization algorithm to adaptively improve the PCC-FCM clustering algorithm to extract characteristic curves, and then using the curve replacement method to repair the abnormal power load data. The present invention improves and optimizes the clustering algorithm through the optimization algorithm, which can effectively realize the repair of abnormal load data in the power system, and has a high accuracy of abnormal data repair.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] An abnormal power load data repair method of an adaptively improved dynamic PCC-FCM includes the following steps:

[0008] Import the power load data set with time series, select the dynamic data set according to the abnormal power load data moment, and perform maximum value normalization processing;

[0009] The FCM clustering was improved based on the Pearson correlation coefficient, and the PCC-FCM clustering algorithm was obtained;

[0010] The snake optimization algorithm is used to adaptively improve the PCC-FCM clustering algorithm, and the dynamic data set is clustered to obtain the clusters;

[0011] The characteristics of each cluster are extracted to obtain the characteristic curve, and then the curve replacement method is used to repair the abnormal power load data.

[0012] The above method optionally imports a power load data set with a time series, selects a dynamic data set according to the abnormal power load data moment, and performs maximum value normalization processing, including the following steps:

[0013] Selecting load curve data at a preset time of an abnormal load data curve to form a dynamic data set;

[0014] The maximum value normalization is performed on each power load curve data in the dynamic data set.

[0015] In the above method, optionally, the preset time is a period of time most recent to which the abnormal load data curve belongs.

[0016] The above method, optionally, improves FCM clustering based on Pearson correlation coefficient to obtain PCC-FCM clustering algorithm, including the following steps:

[0017] The FCC distance measurement function between two data is obtained based on the Pearson correlation coefficient;

[0018] Based on the FCC distance metric function, the iterative formula of cluster center and membership matrix of FCM clustering is improved;

[0019] The Xie-Beni index value is used as the definition of clustering effectiveness, and the Xie-Beni index is improved based on the FCC distance metric function and applied to the evaluation of PCC-FCM clustering.

[0020] In the above method, optionally, the specific content of the FCC distance measurement function between two data obtained based on the Pearson correlation coefficient is:

[0021] The Pearson correlation coefficient is used to measure the distance between two load curve data, specifically:

[0022]

[0023] in, represents the Pearson correlation coefficient between curve X1 and curve X2, represents the standard deviation of curve X1; Indicates the standard deviation of curve X2; cov(X1,X2) indicates the covariance of curve X1 and curve X2; represents the mean value of curve X1; represents the mean value of curve X2;

[0024] The distance formula using FCC is:

[0025]

[0026] Among them, dist p ∈[0,2] represents the distance between curves X1 and X2.

[0027] The above method optionally uses the snake optimization algorithm to adaptively improve the PCC-FCM clustering algorithm, and clusters the dynamic data set to obtain clusters, including the following steps:

[0028] Initialize the population, set the initial position of the snake, the upper and lower bounds, the population number, the maximum number of iterations, and divide the population into two groups, male and female. Use the Xie-Beni indicator as the fitness function to solve the optimal number of clusters.

[0029] The exploration phase involves global search when food is scarce;

[0030] During the development phase, they conduct local searches when food is plentiful and forage when in hot temperatures and enter fighting or mating mode when in cold temperatures;

[0031] If any snake eggs hatch, new individuals are selected to replace the worst male and female individuals;

[0032] After reaching the maximum number of iterations or the set fitness threshold, the snake population individual with the best position is returned to obtain the optimal number of clusters;

[0033] The best number of clusters is used to perform PCC-FCM clustering, and the clustering results are output to obtain each cluster.

[0034] The above method, optionally, extracts features from each cluster to obtain a feature curve, and then uses a curve replacement method to repair the abnormal power load data, including the following steps:

[0035] Extract features from the load curve data of each cluster, and obtain the characteristic curve data of each cluster by mean processing;

[0036] Mark the abnormal load data values ​​in the load curve to be repaired, and remove them accordingly, so as to obtain a load curve data to be repaired without abnormal data, that is, clean load curve data to be repaired;

[0037] The distance between the clean load curve data to be repaired and the characteristic curve data of each cluster is measured based on the Pearson correlation coefficient, and the characteristic curve with the smallest distance is used as the replacement curve;

[0038] The geometric mean of the two normal data points closest to the abnormal load data point in the curve to be repaired and the corresponding data point of the replacement curve is solved, and the abnormal load data is repaired based on the geometric mean.

[0039] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for repairing abnormal power load data of an adaptively improved dynamic PCC-FCM, which has the following beneficial effects:

[0040] 1) Perform maximum normalization on the data set to ensure the clustering effect while improving the algorithm's computing speed;

[0041] 2) The Pearson correlation coefficient is used to improve the distance measurement function of FCM clustering, which can better measure the similarity of load curves and further improve the accuracy of clustering;

[0042] 3) The number of cluster centers of PCC-FCM clustering is optimized by snake optimization algorithm, which effectively solves the problem of the influence of human subjective determination of the number of clusters in PCC-FCM clustering algorithm;

[0043] 4) The present invention significantly improves the accuracy of clustering, thereby improving the accuracy of abnormal data repair, and provides an important basis for load forecasting, peak-shifting management, and optimized scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 A flowchart of an abnormal power load data repair method of an adaptively improved dynamic PCC-FCM provided by the present invention;

[0046] Figure 2The effectiveness index V based on PCC-FCM clustering provided by the embodiment of the present invention xb-pcc The change curve diagram of

[0047] Figure 3 A curve diagram showing the change in repair accuracy of the repair method under different clustering numbers provided by an embodiment of the present invention;

[0048] Figure 4 A flow chart of a snake optimization algorithm provided by an embodiment of the present invention;

[0049] Figure 5 Optimization process diagram of the SO algorithm provided by the embodiment of the present invention;

[0050] Figure 6 A flowchart of the adaptively improved PCC-FCM clustering algorithm based on the snake optimization algorithm provided in an embodiment of the present invention;

[0051] Figure 7 A statistical diagram of the number of data in each cluster provided by the embodiment of the present invention;

[0052] Figure 8 Characteristic curve diagrams of each cluster provided by the embodiment of the present invention; wherein 8a is a characteristic curve diagram of category one, 8b is a characteristic curve diagram of category two, 8c is a characteristic curve diagram of category three, 8d is a characteristic curve diagram of category four, and 8e is a characteristic curve diagram of category five;

[0053] Fig. 9 A comparison chart of the repair accuracy of different repair methods under different abnormal rates provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 As shown, the present invention discloses an abnormal power load data repair method of an adaptively improved dynamic PCC-FCM, comprising the following steps:

[0056] Import the power load data set with time series, select the dynamic data set according to the abnormal power load data moment, and perform maximum value normalization processing;

[0057] The FCM clustering was improved based on the Pearson correlation coefficient, and the PCC-FCM clustering algorithm was obtained;

[0058] The snake optimization algorithm is used to adaptively improve the PCC-FCM clustering algorithm, and the dynamic data set is clustered to obtain the clusters;

[0059] The characteristics of each cluster are extracted to obtain the characteristic curve, and then the curve replacement method is used to repair the abnormal power load data.

[0060] Furthermore, a power load data set with a time series is imported, a dynamic data set is selected according to the abnormal power load data moment, and a maximum value normalization process is performed, including the following steps:

[0061] Selecting load curve data at a preset time of an abnormal load data curve to form a dynamic data set;

[0062] Specifically, the daily load curve represents the change of the power load value within a day. The load data collected by the power consumption information collection system can be divided into normal load data and abnormal load data after the system abnormal data detection. Then the load curve containing abnormal load data is called the abnormal daily load curve, and vice versa. By analyzing the fluctuation of the normal daily load curve, we can understand the daily power consumption behavior of different user types and provide an important basis for load data repair. At the same time, in a period of adjacent time, the load curve has a stronger similarity. Therefore, we assume that the detected abnormal daily load curve is the daily load curve X of the dth day. d ={x1,x2,...,x n}(t=24), and then select the daily load curves of the past 30 days to form a dynamic data set D d ={X d-1 ,X d-2 ,...,X d-30}.

[0063] The maximum value normalization is performed on each power load curve data in the dynamic data set.

[0064] Specifically, due to the large difference in load values ​​between different types of electricity users, in order to improve the operating efficiency and keep the curve characteristics of the data as much as possible, the dynamic data set is normalized to the maximum value and the data is mapped to the [0,1] interval:

[0065]

[0066] Among them, x′ i Indicates the normalized value of load data at a certain moment, x i ,x j It is the original data value of the load data at a certain moment.

[0067] Furthermore, the preset time is a recent period of time to which the abnormal load data curve belongs.

[0068] Furthermore, the FCM clustering is improved based on the Pearson correlation coefficient to obtain the PCC-FCM clustering algorithm, which includes the following steps:

[0069] The FCC distance measurement function between two data is obtained based on the Pearson correlation coefficient;

[0070] Specifically, since electricity users will gradually form a fixed electricity consumption behavior, more attention is paid to the similarity of load curves when analyzing load data. Compared with the commonly used Euclidean distance, the Pearson correlation coefficient pays more attention to the changes in the curve shape, which can improve the accuracy of data feature extraction, thereby better measuring the similarity of load curves.

[0071] In addition, FCC is used to describe the degree of correlation between two variables. The value range is [-1,1]. Negative values ​​indicate that as the value of one variable increases, the other decreases. Positive values ​​indicate that as the value of one variable increases, the other also increases. 0 indicates that the increase or decrease of one variable has no effect on the value of the other. Obviously, the larger the FCC, the higher the similarity of the two curves, that is, the smaller the distance.

[0072] Based on the FCC distance metric function, the iterative formula of cluster center and membership matrix of FCM clustering is improved;

[0073] Specifically, the partitioning principle of the hard C-means clustering (HCM) algorithm is to divide the data set D containing n objects into d ={X d-1 ,X d-2 ,...,X d-30} is divided into c categories, and each object belongs to only one of them, where the membership degree u ij To measure the relationship between each sample object and the class, the principle of the fuzzy C-means clustering algorithm is almost the same as that of the hard C-means clustering. The biggest difference is that in the hard C-means clustering, the membership degree u ij It can only be 1 or 0, with a clear dividing line, but in fuzzy C-means clustering, the membership u ij The value of is between [0,1]. The fuzzy division allows us to more clearly understand the membership relationship between the object and each cluster. Finally, through a c×n-order membership matrix U={u ij} to represent the clustering results, as shown below:

[0074]

[0075] Among them, u ij represents the membership of the jth object to the ith category, and u ij ∈[0,1], Among them, i=1,2,…,c; j=1,2,…,n.

[0076] Fuzzy C-means clustering is an iterative calculation of membership u ij and cluster center v i Until they reach the optimal process, the goal is to minimize the objective function J w Among them, the distance dist is used to measure the similarity between data. The most widely used is the Euclidean distance, but the Pearson correlation coefficient has a better measurement effect on curve similarity. Therefore, the present invention adopts the FCC distance measurement function dist p :

[0077]

[0078] Among them, V i represents the cluster center of the i-th category, w is the fuzzy index, usually 2.

[0079] Based on the membership constraint, the Lagrange multiplier method and mechanism derivation method are used to obtain the update iterative formula of PCC-FCM cluster membership and cluster center as follows:

[0080]

[0081] The Xie-Beni index value is used as the definition of clustering effectiveness, and the Xie-Beni index is improved based on the FCC distance metric function and applied to the evaluation of PCC-FCM clustering.

[0082] Specifically, the Xie-Beni index is widely used in fuzzy clustering. It combines the compactness within the cluster and the separation between clusters. The compactness represents the sum of the distances from each sample data within the cluster to the cluster center, and the separation is measured by the minimum distance between each cluster center point. Since, in the PCC-FCM clustering algorithm, we have changed the metric distance function, the original Xie-Beni index is no longer applicable. Therefore, the present invention proposes an improvement based on the Xie-Beni index and proposes a validity index V suitable for PCC-FCM clustering. xb-pcc :

[0083]

[0084] The Xie-Beni index is defined as the ratio of compactness to separation. To achieve the best clustering effect, the compactness coefficient should be as small as possible and the separation coefficient should be as large as possible. Therefore, the smaller the Xie-Beni index, the better the clustering effect.

[0085] Furthermore, the specific content of the FCC distance measurement function between two data is obtained based on the Pearson correlation coefficient:

[0086] The Pearson correlation coefficient is used to measure the distance between two load curve data, specifically:

[0087]

[0088] in, represents the Pearson correlation coefficient between curve X1 and curve X2, represents the standard deviation of curve X1; Indicates the standard deviation of curve X2; cov(X1,X2) indicates the covariance of curve X1 and curve X2; represents the mean value of curve X1; represents the mean value of curve X2;

[0089] The distance formula using FCC is:

[0090]

[0091] Among them, dist p ∈[0,2] represents the distance between curves X1 and X2.

[0092] Specifically, when clustering sample data, the clustering effect is greatly affected by the number of clusters k. Figure 2 As shown, we can see that there is an obvious inflection point k = 3. When k∈[3,10], the effectiveness index value V xb-pcc The index value is less than 0.002, which indicates that the clustering effect is good. When k>10, the decrease of the index value has become flat. Figure 3 It can be seen that in the application of data repair, as the number of clusters k increases, the repair accuracy obtained by the three different methods all show an overall upward trend. This is because the larger the number of clusters k, that is, the finer the sample division, the more similar the repair curve can be. However, the continuous increase in the number of clusters k increases the complexity of the algorithm while improving the repair accuracy, resulting in lower operating efficiency. Therefore, it is crucial to select an optimal number of clusters that can achieve higher repair accuracy while minimizing the complexity of the algorithm. Considering that the snake optimization algorithm has the advantages of fewer parameters, stable search performance, and fast optimization speed, the present invention adopts the snake optimization algorithm to adaptively improve the PCC-FCM clustering algorithm. The algorithm flow is as follows Figure 4 shown.

[0093] Furthermore, the snake optimization algorithm is used to adaptively improve the PCC-FCM clustering algorithm, and the dynamic data set is clustered to obtain each cluster, including the following steps:

[0094] Initialize the population, set the initial position of the snake, the upper and lower bounds, the population number, the maximum number of iterations, and divide the population into two groups, male and female. Use the Xie-Beni indicator as the fitness function to solve the optimal number of clusters.

[0095] Specifically, first randomly generate an initial snake population with uniform distribution:

[0096] K i =K min +r×(K max -K min )

[0097] Among them, K i is the position of the i-th snake, K max and K min are the upper and lower bounds of the search range, which represent the maximum and minimum number of clusters, and r is a random number in the range of [0, 1]. Then the population is divided into two parts, male and female. For female individuals, It is a male individual.

[0098] At the same time, in the subsequent exploration and development stages, the update of individual positions is mainly affected by temperature T and food quantity Q, which is defined as:

[0099] T=exp(-t / T)

[0100] Q = c1 × exp((tT) / T)

[0101] Among them, t represents the current iteration number, T represents the maximum number of iterations, and c1 represents the learning factor constant, which is 0.5 here.

[0102] The exploration phase involves global search when food is scarce;

[0103] Specifically, when Q < 0.25, it indicates a food shortage. SO randomly selects any position to explore food and updates the position through global search, which is the exploration phase:

[0104]

[0105]

[0106] in, Represent the positions of male and female individuals, represent the random positions of male and female individuals, c2 represents the learning factor constant, which is 0.05 here, and A f , A m Respectively represent the ability of male and female individuals to find food, and r represents a random number in the range of [0, 1]:

[0107]

[0108]

[0109] in, Respectively The fitness corresponding to the position, Respectively The fitness corresponding to the position.

[0110] During the development phase, they conduct local searches when food is plentiful and forage when in hot temperatures and enter fighting or mating mode when in cold temperatures;

[0111] Specifically, when Q>0.25, it indicates that food is sufficient and local search begins, which is the development stage.

[0112] In the development stage, the influence of temperature T can be further divided into two modes: foraging mode and fighting or mating mode. When T>0.6, it is in a hot temperature state, and the snake population individuals will only look for food, that is, foraging mode, and the position update formula is:

[0113] K i,j (t+1)=K food ±c3×T×r×(K food -K i,j (t))

[0114] Among them, K i,j represents the individual position of the entire snake population, K food represents the global optimal position, c3 represents the learning factor constant, which is 2 here;

[0115] When T is less than 0.6, in a cold temperature state, individuals in the snake population will enter fighting mode or mating mode.

[0116] In combat mode, the position update formula of individual snake populations is:

[0117]

[0118]

[0119] in, represent the best position in the female and male populations, respectively, and F f 、F m Respectively represent the fighting ability of female and male individuals:

[0120]

[0121]

[0122] in, Respectively The fitness corresponding to the position, Respectively The fitness corresponding to the position.

[0123] In the mating mode, the position update formula of the snake population individuals is:

[0124]

[0125]

[0126] Among them, M f 、M m Represent the mating ability of female and male individuals respectively:

[0127]

[0128]

[0129] If any snake eggs hatch, new individuals are selected to replace the worst male and female individuals;

[0130] Specifically, when snake eggs hatch, it means that new individuals are produced. At this time, the new individuals are used to replace the worst female and male individuals in the population.

[0131] After reaching the maximum number of iterations or the set fitness threshold, the snake population individual with the best position is returned to obtain the optimal number of clusters;

[0132] Specifically, from Figure 5 It can be seen that the snake optimization algorithm converges quickly and obtains the optimal fitness value, that is, the optimal number of clusters, in the first iteration.

[0133] The best number of clusters is used to perform PCC-FCM clustering, and the clustering results are output to obtain each cluster.

[0134] Specifically, the process of the dynamic PCC-FCM clustering algorithm based on the adaptive improvement of the snake optimization algorithm is as follows: Figure 6 As shown:

[0135] Step 1: Import dataset D d , the optimal number of clusters k is obtained using the SO algorithm;

[0136] Step 2: Determine the fuzzy index w, the iteration termination condition ε, randomly select the initial cluster center, and initialize the membership matrix;

[0137] Step 3: Update the membership matrix and cluster center according to the update iteration formula of the membership and cluster center of the PCC-FCM clustering algorithm;

[0138] Step 4: Calculate the objective function J w , judge the relationship between the two iterative errors of the objective function and the given threshold ε. If the iterative error is less than the threshold, clustering is terminated, otherwise jump to step 3;

[0139] Based on the above steps 1-4, k clusters D can be obtained by clustering d ={D1,D2,...,D k}, here the best number of clusters obtained in our specific example is 5, so a total of 5 clusters are obtained, and the number of each cluster is as follows Figure 7 shown.

[0140] Furthermore, feature extraction is performed on each cluster to obtain a feature curve, and then the abnormal power load data is repaired using a curve replacement method, including the following steps:

[0141] Extract features from the load curve data of each cluster, and obtain the characteristic curve data of each cluster by mean processing;

[0142] Specifically, by calculating the average of the load curve data of each cluster, the characteristic curve set d = {d1, d2, ..., d k}:

[0143]

[0144] Among them, d i represents the characteristic curve data of the i-th cluster, and m represents the number of samples contained in the i-th cluster. Figure 8 As shown, we can obtain 5 characteristic curves, among which 8a is the characteristic curve diagram of category one, 8b is the characteristic curve diagram of category two, 8c is the characteristic curve diagram of category three, 8d is the characteristic curve diagram of category four, and 8e is the characteristic curve diagram of category five.

[0145] Mark the abnormal load data values ​​in the load curve to be repaired, and remove them accordingly, so as to obtain a load curve data to be repaired without abnormal data, that is, clean load curve data to be repaired;

[0146] Specifically, assume that the load curve detected to contain abnormal load data is X d-dirty ={x1,x2,...,x n}, where x6, x7, x8 are abnormal data points, then the data points x6, x7, x8 marked as abnormal are removed from the load curve X d-dirty ={x1,x2,...,x n}, and then obtain a clean curve X to be repaired d-clean ={x1,...,x5,x9,...,x n}, thereby eliminating the influence of abnormal load data on the selection of characteristic curves.

[0147] The distance between the clean load curve data to be repaired and the characteristic curve data of each cluster is measured based on the Pearson correlation coefficient, and the characteristic curve with the smallest distance is used as the replacement curve;

[0148] Specifically, based on the proposed FCC metric distance formula, by calculating the curve to be repaired X d-clean ={x1,...,x5,x9,...,x n} and each cluster characteristic curve d={d1,d2,...,d k}, select the characteristic curve with the smallest distance as the replacement curve of the curve to be repaired:

[0149]

[0150] The geometric mean of the two normal data points closest to the abnormal load data point in the curve to be repaired and the corresponding data point of the replacement curve is solved, and the abnormal load data is repaired based on the geometric mean.

[0151] Specifically, based on the normal data points at the two moments closest to the abnormal data point, the geometric mean method is used to calculate the ratio of the data points corresponding to the characteristic curve, so that the abnormal data point is replaced with the corresponding data point on the characteristic curve based on the ratio to achieve abnormal data repair.

[0152] For abnormal data repair, assume that the abnormal point is X d (x i ), then the repaired load curve X′ d (x i )for:

[0153]

[0154] Among them, X′ d (x i ) represents the load curve after repair, X d (x i ) represents the load curve before repair, X replace (x i ) represents the characteristic curve with the highest similarity to the curve to be repaired, x i-close1 、x i-close2 They represent the normal data points at the two moments closest to the abnormal data point.

[0155] Here, we set the data anomaly rate and randomly select abnormal data points for repair experiments, such as Fig. 9As shown in the figure, we can see that the repair accuracy of the method proposed in the present invention is higher than the commonly used ratio method and difference method at different data anomaly rates, and when the anomaly rate is 50%, the repair accuracy can still reach nearly 88%, which is significantly higher than the other two methods, further verifying the effectiveness and superiority of the method proposed in the present invention. At the same time, the repair accuracy of abnormal data shows a downward trend with the increase of the anomaly rate, but there is a fluctuation. This is because the abnormal data points randomly selected by the system may be continuous or discontinuous. Continuous abnormal data points will further reduce the available information when selecting the characteristic curve, and thus the repair accuracy is low. Among them, when the data anomaly rate is 25%, the abnormal data points of the daily load curve are 6, and the randomly selected discontinuous abnormal points are [1, 3, 7, 16, 18, 22], and the continuous abnormal points are [3, 4, 5, 6, 7, 8]. After repair, the repair accuracy of discontinuous abnormal data is 97%, and the repair accuracy of continuous abnormal data is 94.9%. It can be seen that the repair accuracy of continuous abnormal data is significantly lower than that of discontinuous abnormal data.

[0156] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0157] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An adaptive improved dynamic PCC-FCM abnormal power load data repair method, characterized in that: The following steps are involved: Import the power load data set with time series, select the dynamic data set according to the abnormal power load data moment, and perform maximum value normalization processing; The FCM clustering was improved based on the Pearson correlation coefficient, and the PCC-FCM clustering algorithm was obtained; The snake optimization algorithm is used to adaptively improve the PCC-FCM clustering algorithm, and the dynamic data set is clustered to obtain the clusters; The characteristics of each cluster are extracted to obtain the characteristic curve, and then the curve replacement method is used to repair the abnormal power load data.

2. According to claim 1, the method for repairing abnormal power load data of an adaptively improved dynamic PCC-FCM is characterized in that: Import the power load data set with time series, select the dynamic data set according to the abnormal power load data moment, and perform maximum value normalization processing, including the following steps: Selecting load curve data at a preset time of an abnormal load data curve to form a dynamic data set; The maximum value normalization is performed on each power load curve data in the dynamic data set.

3. The abnormal power load data repair method of the adaptive improved dynamic PCC-FCM according to claim 2 is characterized in that: The preset time is the most recent period of time to which the abnormal load data curve belongs.

4. According to claim 1, the method for repairing abnormal power load data of an adaptively improved dynamic PCC-FCM is characterized in that: Based on the Pearson correlation coefficient, the FCM clustering is improved to obtain the PCC-FCM clustering algorithm, which includes the following steps: The FCC distance measurement function between two data is obtained based on the Pearson correlation coefficient; Based on the FCC distance metric function, the iterative formula of cluster center and membership matrix of FCM clustering is improved; The Xie-Beni index value is used as the definition of clustering effectiveness, and the Xie-Beni index is improved based on the FCC distance metric function and applied to the evaluation of PCC-FCM clustering.

5. The method for repairing abnormal power load data of an adaptively improved dynamic PCC-FCM according to claim 4 is characterized in that: The specific content of the FCC distance measurement function between two data based on the Pearson correlation coefficient is: The Pearson correlation coefficient is used to measure the distance between two load curve data, specifically: in, represents the Pearson correlation coefficient between curve X1 and curve X2, represents the standard deviation of curve X1; Indicates the standard deviation of curve X2; cov(X1,X2) indicates the covariance of curve X1 and curve X2; represents the mean value of curve X1; represents the mean value of curve X2; The distance formula using FCC is: Among them, dist p ∈[0,2] represents the distance between curves X1 and X2.

6. The method for repairing abnormal power load data of an adaptively improved dynamic PCC-FCM according to claim 1 is characterized in that: The snake optimization algorithm is used to adaptively improve the PCC-FCM clustering algorithm, and the dynamic data set is clustered to obtain each cluster cluster, including the following steps: Initialize the population, set the initial position of the snake, the upper and lower bounds, the population number, the maximum number of iterations, and divide the population into two groups, male and female. Use the Xie-Beni indicator as the fitness function to solve the optimal number of clusters. The exploration phase involves global search when food is scarce; During the development phase, they conduct local searches when food is plentiful and forage when in hot temperatures and enter fighting or mating mode when in cold temperatures; If any snake eggs hatch, new individuals are selected to replace the worst male and female individuals; After reaching the maximum number of iterations or the set fitness threshold, the snake population individual with the best position is returned to obtain the optimal number of clusters; The best number of clusters is used to perform PCC-FCM clustering, and the clustering results are output to obtain each cluster.

7. The method for repairing abnormal power load data of an adaptively improved dynamic PCC-FCM according to claim 1 is characterized in that: The characteristic curve is obtained by extracting the features of each cluster, and then the abnormal power load data is repaired by using the curve replacement method, which includes the following steps: Extract features from the load curve data of each cluster, and obtain the characteristic curve data of each cluster by mean processing; Mark the abnormal load data values ​​in the load curve to be repaired, and remove them accordingly, so as to obtain a load curve data to be repaired without abnormal data, that is, clean load curve data to be repaired; The distance between the clean load curve data to be repaired and the characteristic curve data of each cluster is measured based on the Pearson correlation coefficient, and the characteristic curve with the smallest distance is used as the replacement curve; The geometric mean of the two normal data points closest to the abnormal load data point in the curve to be repaired and the corresponding data point of the replacement curve is solved, and the abnormal load data is repaired based on the geometric mean.

Images