A method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis

Abstract

This invention discloses a method for filling in voltage anomaly data in distribution networks based on fluctuation cross-correlation analysis, belonging to the field of power system voltage identification technology. The method includes: S1, acquiring historical data of the power grid; S2, inputting historical samples and dividing them into a training set and a test set; S3, constructing a similarity set WZ based on an attribute weighting strategy; S4, training a random forest using the similarity set WZ; S5, constructing a random forest using the N-Bootstrap sampling method; S6, training the random forest using the training set and testing the trained model using the test set; S7, if the test requirements are met, proceed to step 8; otherwise, proceed to step 6 to continue training; S8, predicting voltage anomalies using the trained random forest model, and using the prediction results as the filling results; S9, ending the filling operation. This invention can fully utilize historical voltage data to fill in missing voltage values, improving the accuracy of the filled voltage values.

Description

Technical Field

[0001] This invention relates to a method for filling abnormal distribution network voltage data based on fluctuation cross-correlation analysis (FSMDFCCA), belonging to the field of power system voltage identification technology. Background Technology

[0002] With the continuous development of the power grid, the accuracy and completeness of data are crucial for distribution network voltage control. However, with the exponential growth of collected data, voltage data gaps due to manual entry or data acquisition device malfunctions occur frequently. Therefore, it is necessary to identify or complete missing data. Traditional algorithms offer some solutions, but their completion results are not ideal because they rarely utilize similarity attributes as analytical basis. In recent years, with the introduction of fluctuation cross-correlation theory and the gradual maturation of traditional machine learning methods such as random forests, it has become possible to train random forests using similarity sets. Therefore, we propose a distribution network voltage anomaly data completion method based on fluctuation cross-correlation analysis to further improve the accuracy of missing distribution network voltage values ​​and meet the needs of power grid development. Summary of the Invention

[0003] This invention addresses the shortcomings of existing technologies by providing a method for filling in abnormal distribution network voltage data based on fluctuation cross-correlation analysis, thereby improving the accuracy of the filled data and providing support for distribution network regulation.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis includes the following steps:

[0006] S1. Obtain historical data of the power grid;

[0007] S2. Divide the historical data into a training set and a test set;

[0008] S3. Construct a similarity set WZ based on an attribute weighting strategy;

[0009] S4. Construct a random forest using the N-Bootstrap sampling method;

[0010] S5. The random forest is trained using a similarity set WZ, and the trained random forest model is tested using a test set.

[0011] S6. Determine whether the trained random forest model meets the test requirements. If the trained random forest model meets the test requirements, proceed to step S7; otherwise, proceed to step S5 to continue training.

[0012] S7. Use the trained random forest model to predict voltage anomalies, and use the prediction results as the filling results.

[0013] S8. End this filling process.

[0014] To optimize the above technical solution, the specific measures also include:

[0015] Furthermore, step S3 specifically includes:

[0016] S31. Obtain the voltage-related attribute set SX i , i = 1, 2, ..., U, where U represents the number of associated attributes;

[0017] S32. Use the fluctuation cross-correlation coefficient to analyze the cross-correlation between associated attributes and set a threshold. Calculate the fluctuation cross-correlation coefficient between associated attributes. When the fluctuation cross-correlation coefficient is greater than a threshold... When this happens, the associated attribute is stored in the attribute collection WG;

[0018] S33. Use the entropy weight method to obtain the weights of each associated attribute in the attribute set WG. W , , m This indicates the number of associated attributes in the attribute set WG;

[0019] S34. Obtain the comprehensive weighted value of the associated attributes based on the cross-correlation coefficient of fluctuations. , ;

[0020] S35. Select historical data sections for each typical time period from the training set and calculate the corresponding comprehensive weighted value of the associated attributes. And the associated attributes are combined and weighted. Sort from largest to smallest, given a correlation threshold. If the weighted average of the correlation attributes of historical data sections Greater than a given threshold Then the historical data cross-sections will be stored in the similarity set WZ.

[0021] Furthermore, step S4 specifically includes:

[0022] S41. For each CART decision tree in the random forest, samples are drawn from the initial sample set S using N-Bootstrap sampling, and N subsets {S1, S2, ..., S} are constructed.N}, take half of the samples from the N subsets as the training sample set. S Ntr The other half of the samples were used as the test sample set. N Nte ;

[0023] S42, From the training sample set S Ntr Selected samples are used as root nodes for training, and a multi-index fusion split coefficient is constructed. RH ( S Ntr As a metric for whether a node has split:

[0024]

[0025] Require ,and They cannot be equal, where Gini() represents the Gini coefficient and Gain() represents the information gain; calculate the minimum fusion split coefficient value in the test set. The decision tree is formed after the node splits, which is used as the criterion for node splitting.

[0026] S43. Finally, construct a random forest from all the generated decision trees.

[0027] Further, in step S32, the calculation of the fluctuation cross-correlation coefficient between the associated attributes specifically involves:

[0028] S321. For two time series of equal length and ,in ;calculate , The sum of the differences from the mean:

[0029]

[0030] in: Indicates the sampling length. , They represent and At sampling length The sum of the differences between the lower and the average. and They represent and The average value;

[0031] S322, calculation respectively represent and Forward difference of autocorrelation:

[0032]

[0033] in: For each sampling time period All One difference, and They represent and Forward difference of autocorrelation;

[0034] S323, Calculation and Covariance:

[0035]

[0036]

[0037] in: express and covariance, express The average;

[0038] S324, Calculation and The cross-correlation coefficient of fluctuations, if and If there is a correlation, then Satisfies power-law distribution ;in, Let be the cross-correlation coefficient of the fluctuations, representing and Correlation degree; fluctuation cross-correlation coefficient obtained by fitting a power-law distribution. .

[0039] Furthermore, in step S31, the associated attributes include active power, reactive power, current, temperature, humidity, and proximity point voltage.

[0040] Furthermore, in step S6, the method used to determine whether the trained random forest model meets the test requirements is to calculate the root mean square error between the predicted value and the true value.

[0041] The beneficial effects of this invention are:

[0042] This invention proposes a method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis. This method can use historical data to train the voltage filling model, thereby improving the accuracy of voltage filling data. This invention introduces a fluctuation cross-correlation analysis algorithm to construct attributes related to missing attributes, and uses the entropy weight method to assign weights to strongly correlated attributes. Based on the comprehensive attribute ranking results of historical data, a similarity set is obtained, and the improved random forest model is trained using the similarity set to achieve accurate filling of missing values ​​in distribution network voltage. Attached Figure Description

[0043] Figure 1 This is a flowchart of the distribution network voltage anomaly data filling method based on fluctuation cross-correlation analysis proposed in this invention;

[0044] Figure 2 A flowchart for the comprehensive weighting of attributes based on fluctuation cross-correlation;

[0045] Figure 3 A comparison chart showing the root mean square error of results filled in by different algorithms;

[0046] Figure 4 Fill in the accuracy comparison chart of different algorithms;

[0047] Figure 5 A comparison chart of the results of filling in FSMDFCCA and the actual values. Detailed Implementation

[0048] The invention will now be described in further detail with reference to the accompanying drawings.

[0049] In one embodiment, the present invention proposes a method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis, the flowchart of which is as follows: Figure 1 As shown, it includes the following steps:

[0050] S1. Obtain historical data of the power grid;

[0051] S2. Divide the historical data into a training set and a test set;

[0052] S3. Construct a similarity set WZ based on an attribute weighting strategy;

[0053] S4. Construct a random forest using the N-Bootstrap sampling method;

[0054] S5. The random forest is trained using a similarity set WZ, and the trained random forest model is tested using a test set.

[0055] S6. Determine whether the trained random forest model meets the test requirements. If the trained random forest model meets the test requirements, proceed to step S7; otherwise, proceed to step S5 to continue training.

[0056] S7. Use the trained random forest model to predict voltage anomalies, and use the prediction results as the filling results.

[0057] S8. End this filling process.

[0058] The specific process of step S3 is as follows: Figure 2 As shown, it includes:

[0059] S31. Obtain the voltage-related attribute set SX i , i=1,2,...,U, where U represents the number of associated attributes; the associated attributes include active power, reactive power, current, temperature, humidity and proximity point voltage.

[0060] S32. Use the fluctuation cross-correlation coefficient to analyze the cross-correlation between associated attributes and set a threshold. Calculate the fluctuation cross-correlation coefficient between associated attributes. When the fluctuation cross-correlation coefficient is greater than a threshold... When this happens, the associated attribute is stored in the attribute collection WG;

[0061] S33. Use the entropy weight method to obtain the weights of each associated attribute in the attribute set WG. W , , m This indicates the number of associated attributes in the attribute set WG;

[0062] S34. Obtain the comprehensive weighted value of the associated attributes based on the cross-correlation coefficient of fluctuations. , ;

[0063] S35. Select historical data sections for each typical time period from the training set and calculate the corresponding comprehensive weighted value of the associated attributes. And the associated attributes are combined and weighted. Sort from largest to smallest, given a correlation threshold. If the weighted average of the correlation attributes of historical data sections Greater than a given threshold Then the historical data cross-sections will be stored in the similarity set WZ.

[0064] In step S32, the calculation of the fluctuation cross-correlation coefficient between the associated attributes specifically involves:

[0065] S321. For two time series of equal length and ,in ;calculate , The sum of the differences from the mean:

[0066]

[0067] in: Indicates the sampling length. , They represent and At sampling length The sum of the differences between the lower and the average. and They represent and The average value;

[0068] S322, calculation respectively represent and Forward difference of autocorrelation:

[0069]

[0070] in: For each sampling time period All One difference, and They represent and Forward difference of autocorrelation;

[0071] S323, Calculation and Covariance:

[0072]

[0073]

[0074] in: express and covariance, express The average;

[0075] S324, Calculation and The cross-correlation coefficient of fluctuations, if and If there is a correlation, then Satisfies power-law distribution ;in, Let be the cross-correlation coefficient of the fluctuations, representing and Correlation degree; fluctuation cross-correlation coefficient obtained by fitting a power-law distribution. ,when When, it means and Irrelevant; when When, it means and Positive correlation; when When, it means and Negative correlation; The larger the value, the more likely it is to represent and The higher the degree of correlation.

[0076] This invention is applied to voltage filling in a power distribution network. Two years of historical data are used as the sample set. In order to reflect the advantages of filling, the K-Nearest Neighbors (KNN) algorithm, the Random Forest (RF) algorithm, and the FSMDFCCA proposed in this paper are selected to analyze the filling results.

[0077] To verify the difference in filling effects, the root mean square error (RMSE) strategy is used to assess the similarity between the filled value and the true value. Indicates filling error, when The smaller the size, the better the filling effect. The calculation is as follows:

[0078]

[0079] In the formula: These are the true value and the imputed value, respectively. n It represents the number of missing values.

[0080] The filling accuracy is indicated by the following formula:

[0081]

[0082] This indicates the number of correct estimates. .

[0083] To analyze the effectiveness of the algorithm in imputation under different missing percentages, a random missing dataset with a missing percentage ranging from 1% to 30% was constructed. Experiments were conducted using FSMDFCCA, RF, and KNN to compare the root mean square error and imputation accuracy. Using the missing voltage value of a node in a real-world power grid as the imputation target, the imputation error of voltage missing data under different missing conditions was analyzed. Figure 3 As shown in the figure, the FSMDFCCA algorithm proposed in this paper achieves the best imputation effect. It can also be clearly seen from the figure that as the missing rate increases, the errors of RF and KNN increase significantly, while the FSMDFCCA algorithm remains relatively stable, especially when the missing rate is greater than 15%, the error change is not obvious. Figure 4The imputation accuracy of three algorithms under different missing value backgrounds is presented. The figures clearly show that FSMDFCCA has better imputation performance than RF and KNN. To further demonstrate the advantages of FSMDFCCA, this paper analyzes the case with a missing value rate of 15%. Figure 5 The comparative analysis of 30 consecutive sets of data under a certain missing condition is presented. The results in the figure clearly show that the curve drawn by the FSMDFCCA algorithm proposed in this paper has a good fit with the true value curve, the filling result is close to the true value, and the filling effect is obvious.

[0084] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis, characterized in that, Includes the following steps: S1. Obtain historical data of the power grid; S2. Divide the historical data into a training set and a test set; S3. Construct a similarity set WZ based on an attribute-weighted strategy; Step S3 specifically includes: S31. Obtain the voltage-related attribute set SX i , i = 1, 2, ..., U, where U represents the number of associated attributes; S32. Use the fluctuation cross-correlation coefficient to analyze the cross-correlation between associated attributes and set a threshold. Calculate the fluctuation cross-correlation coefficient between associated attributes. When the fluctuation cross-correlation coefficient is greater than a threshold... When this happens, the associated attribute is stored in the attribute collection WG; S33. Use the entropy weight method to obtain the weights of each associated attribute in the attribute set WG. W , , m This indicates the number of associated attributes in the attribute set WG; S34. Obtain the comprehensive weighted value of the associated attributes based on the cross-correlation coefficient of fluctuations. , ; S35. Select historical data sections for each typical time period from the training set and calculate the corresponding comprehensive weighted value of the associated attributes. And the associated attributes are combined and weighted. Sort from largest to smallest, given a correlation threshold. If the weighted average of the correlation attributes of historical data sections Greater than a given threshold Then the historical data cross-sections will be stored in the similarity set WZ; S4. Construct a random forest using the N-Bootstrap sampling method; S5. The random forest is trained using a similarity set WZ, and the trained random forest model is tested using a test set. S6. Determine whether the trained random forest model meets the test requirements. If the trained random forest model meets the test requirements, proceed to step S7; otherwise, proceed to step S5 to continue training. S7. Use the trained random forest model to predict voltage anomalies, and use the prediction results as the filling results. S8. End of this filling process.

2. The method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis according to claim 1, characterized in that, Step S4 specifically includes: S41. For each CART decision tree in the random forest, samples are drawn from the initial sample set S using N-Bootstrap sampling, and N subsets {S1, S2, ..., S} are constructed. N }, take half of the samples from the N subsets as the training sample set. S Ntr The other half of the samples were used as the test sample set. N Nte ; S42, From the training sample set S Ntr Selected samples are used as root nodes for training, and a multi-index fusion split coefficient is constructed. RH ( S Ntr As a metric for whether a node has split: Require ,and They cannot be equal, where Gini() represents the Gini coefficient and Gain() represents the information gain; calculate the minimum fusion split coefficient value in the test set. The decision tree is formed after the node splits, which is used as the criterion for node splitting. S43. Finally, construct a random forest from all the generated decision trees.

3. The method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis according to claim 1, characterized in that, In step S32, the calculation of the fluctuation cross-correlation coefficient between the associated attributes specifically involves: S321. For two time series of equal length and ,in ;calculate , The sum of the differences from the mean: in: Indicates the sampling length. , They represent and At sampling length The sum of the differences between the lower and the average. and They represent and The average value; S322, calculation respectively represent and Forward difference of autocorrelation: in: For each sampling time period All One difference, and They represent and Forward difference of autocorrelation; S323, Calculation and Covariance: in: express and covariance, express The average; S324, Calculation and The cross-correlation coefficient of fluctuations, if and If there is a correlation, then Satisfies power-law distribution ;in, Let be the cross-correlation coefficient of the fluctuations, representing and Correlation degree; fluctuation cross-correlation coefficient obtained by fitting a power-law distribution. .

4. The method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis according to claim 1, characterized in that, In step S31, the associated attributes include active power, reactive power, current, temperature, humidity, and proximity point voltage.

5. The method for filling in distribution network voltage anomaly data based on fluctuation cross-correlation analysis according to claim 1, characterized in that, In step S6, the method used to determine whether the trained random forest model meets the test requirements is to calculate the root mean square error between the predicted value and the true value.

Images