Power grid operation inspection abnormal data processing method

Abstract

The invention relates to a data processing technology in the technical field of power grid operation inspection, in particular to a power grid operation inspection abnormal data processing method. The method comprises the following steps: dividing time allocation items, and performing spatial point clustering on multi-dimensional operation inspection abnormal data of different time allocation items; and obtaining clustering routes of various multi-dimensional operation inspection abnormal data of different time distribution items, and calculating a clustering route coincidence rate. In the clustering division process, along with the continuous increase of the data volume, each extreme value is divided into a far-end centralized cluster, the extreme values are automatically distinguished, the interference of the extreme values on each piece of multi-dimensional operation inspection abnormal data is avoided, the parameter range division accuracy is improved, and meanwhile, by dividing the range of the extreme values in the centralized cluster storing the extreme values, the accuracy of parameter range division is improved. And the extreme value change state of each multi-dimensional operation inspection abnormal data is determined, the parameter range is updated in time, and the parameter range division accuracy is further improved.

Description

Technical Field

[0001] The present invention relates to data processing technology in the field of power grid operation and inspection technology, and in particular to a method for processing abnormal data in power grid operation and inspection. Background Art

[0002] During the operation and inspection of power grids, the diversity and complexity of abnormal causes directly affect the stability and security of the power system. In the process of collecting data for power grid operation and inspection, it generally includes regular data and abnormal data. Since the range of regular data is limited, the storage and analysis of data are concise and clear. However, for abnormal data, due to its uncertain range, the values corresponding to abnormal data caused by different causes are quite different. Even abnormal data caused by the same cause will have numerical differences due to different degrees of impact.

[0003] The existing method for processing abnormal data of power grid operation and inspection generally includes the following steps:

[0004] The first step is to collect and process abnormal data of power grid operation and inspection, and classify and process the abnormal data according to different abnormal values;

[0005] The second step is to determine the range of each abnormal data according to the numerical distribution of abnormal data of power grid operation and inspection;

[0006] The third step is to update the scope of each abnormal data in real time according to the changes in the same type of data collected in real time.

[0007] Although the above method can count the changes in the range of various abnormal data, the range is too wide to make adaptive predictions. That is, the changing patterns of abnormal data are different in different time periods, and extreme values will appear in some extreme conditions, which will further expand the update range and greatly reduce the accuracy of the subsequent analysis of the causes of abnormalities based on abnormal values.

[0008] In order to solve the above problems, a method for processing abnormal data of power grid operation and inspection with time-sharing data update is urgently needed. Summary of the Invention

[0009] The purpose of the present invention is to provide a method for processing abnormal data of power grid operation and inspection to solve the problems raised in the above background technology.

[0010] To achieve the above object, a method for processing abnormal data of power grid operation and inspection is provided, comprising the following steps:

[0011] S1. Collect multi-dimensional operation and inspection abnormal data during the power grid operation and inspection process, and obtain multi-dimensional operation and inspection abnormal data matching items;

[0012] S2. Reorganize the time series of multi-dimensional operation and inspection abnormal data and integrate them into a unified sequence according to the time axis;

[0013] S3. Divide the time allocation projects and perform spatial point clustering on the multi-dimensional operation and inspection abnormal data of different time allocation projects;

[0014] S4. Obtain clustered routes of multi-dimensional inspection abnormal data for different time allocation projects and calculate the clustered route overlap rate;

[0015] S5. Obtain the cluster center point based on the cluster route overlap rate, and divide the parameter ranges of various multi-dimensional operation and inspection abnormal data in different time allocation projects.

[0016] As a further improvement of the present technical solution, the abnormal data matching items in S1 are the equipment model, the time point and the value of the abnormal item.

[0017] As a further improvement of the present technical solution, the method of integrating into a unified sequence according to the time axis in S2 includes the following steps:

[0018] S2.1. Define each timeline according to the industry standard for abnormal data collection;

[0019] S2.2. Obtain the device model and bind the values of the abnormal items collected in the device model through the time axis.

[0020] As a further improvement of the present technical solution, the time allocation items in S3 include the same day but different times and the same time but different days.

[0021] As a further improvement of this technical solution, the method for performing spatial point clustering in S3 includes the following steps:

[0022] S3.1. Collect individual data of the same abnormal item at different times, count the individual data, and generate a data set;

[0023] S3.2. Randomly select individual data from the data set as the cluster center, calculate the distance between each data and the cluster center, and assign each data to the nearest cluster center to generate clusters;

[0024] S3.3. Calculate the average value of the individual data in the cluster to obtain a new cluster center, continue to calculate the distance between each individual data and the cluster center, and assign each individual data to the nearest cluster center;

[0025] S3.4. Continue to update the average value of each cluster and repeat the above steps until each cluster no longer changes.

[0026] As a further improvement of this technical solution, the optimization method for performing spatial point clustering in S3 includes the following steps:

[0027] S3.5. Select z points with the greatest possible batch distance and randomly select one as the initial cluster center.

[0028] S3.6. Select the point farthest from the z points as the initial cluster center, then select the point with the largest distance from the first two points as the third initial cluster center, and so on, until z initial cluster centers are selected.

[0029] As a further improvement of this technical solution, in S3.4, it is determined that each concentrated cluster no longer changes using a clustering algorithm, and the algorithm formula is as follows:

[0030]

[0031] Where F(iter) is the objective function value, is the abnormal value corresponding to different data, z j (iter) is the cluster center value of each cluster, i represents the data subscript, distinguishing different data, j represents the cluster center subscript, distinguishing different cluster centers. The specific steps are as follows:

[0032] The first step is to first give a data set of size n, let iter = 1, iter represents the number of iterations, and select k initial cluster centers z j (iter), where j = 1, 2, ..., k;

[0033] Step 2: Calculate each data x i , i = 1, 2, ..., n and the distance between the cluster center, x i Assigned to the nearest cluster center z j The cluster to which (iter) belongs, i.e. |x i -z j (iter)|≤|x i -z j !(iter)|, where j ！ is the subscript of another cluster center, i.e. j ！ ≠j,j ！ , j∈(1,2,…,k);

[0034] Step 3: Let iter = iter + 1, calculate the new cluster center, and calculate the objective function value F(iter);

[0035] Step 4. Judgment: If |F(iter+1)-F(iter)|<θ, θ is the convergence constant, or there is no category change in the concentrated cluster, the algorithm ends, otherwise return to step 2.

[0036] As a further improvement of the present technical solution, the method for calculating the cluster route overlap rate in S4 includes the following steps:

[0037] S4.1. Record the clusters and corresponding cluster centers of each abnormal data during the entire clustering process.

[0038] S4.2. Obtain the corresponding routes based on the order in which the clusters are divided and the corresponding cluster centers;

[0039] S4.3. Obtain the overlap rate of the routes through which each abnormal data item passes, and select the route with the highest overlap rate among the clustered routes as the comparison route;

[0040] S4.4. Define the overlap rate threshold of clustered routes;

[0041] When the coincidence rate of the comparison route passed by the abnormal data is greater than or equal to the coincidence rate threshold, the coincidence rate is marked as the coincidence rate matched by the abnormal data, and the corresponding clustering route is obtained;

[0042] When the coincidence rate of the comparison route passed by the entire abnormal data is less than the coincidence rate threshold, it indicates that the abnormal data is not related to time, and the abnormal value range of the abnormal data is obtained as the final parameter range.

[0043] As a further improvement of the present technical solution, the method of dividing the parameter ranges of each multi-dimensional operation and inspection abnormal data in different time allocation items in S5 includes the following steps:

[0044] S5.1. Obtain the clusters that the clustering route passes through and the corresponding cluster centers;

[0045] S5.2. Arrange them by size, select the minimum and maximum values, and combine them into the cluster center range.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] In the power grid operation and inspection abnormal data processing method, during the clustering process, as the amount of individual data continues to increase, the extreme values will be divided into remote centralized clusters, and the extreme values will be automatically distinguished to avoid the interference of extreme values on various multi-dimensional operation and inspection abnormal data, thereby improving the accuracy of parameter range division. At the same time, by dividing the extreme value range in the centralized cluster storing extreme values, the extreme value change status of various multi-dimensional operation and inspection abnormal data is determined, and the parameter range is updated in time, further improving the accuracy of parameter range division. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a diagram of the overall method steps of the present invention;

[0049] Figure 2A diagram of the method steps of the present invention integrated into a unified sequence according to the time axis;

[0050] Figure 3 This is one of the clustering flow charts of the present invention;

[0051] Figure 4 This is the second clustering flow chart of the present invention;

[0052] Figure 5 A diagram showing the steps of a method for performing spatial point clustering according to the present invention;

[0053] Figure 6 This is a step diagram of the optimization method for performing spatial point clustering of the present invention;

[0054] Figure 7 A diagram showing the steps of a method for calculating the cluster route overlap rate of the present invention;

[0055] Figure 8 This is a step diagram of the method for dividing the parameter ranges of various multi-dimensional inspection abnormal data into different time allocation items of the present invention;

[0056] Figure 9 This is an optimization flow chart for spatial point clustering of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0058] The following are technical terms used by those skilled in the art:

[0059] In the k-means algorithm, a data point refers to a single sample or observation in a data set, which is the basic unit of clustering processing.

[0060] See also Figure 1 As shown, a method for processing abnormal data of power grid operation and inspection is provided, comprising the following steps:

[0061] S1. Collect multi-dimensional operation and inspection abnormal data during the power grid operation and inspection process, and obtain multi-dimensional operation and inspection abnormal data matching items;

[0062] S2. Reorganize the time series of multi-dimensional operation and inspection abnormal data and integrate them into a unified sequence according to the time axis;

[0063] S3. Divide the time allocation projects and perform spatial point clustering on the multi-dimensional operation and inspection abnormal data of different time allocation projects;

[0064] S4. Obtain clustered routes of multi-dimensional inspection abnormal data for different time allocation projects and calculate the clustered route overlap rate;

[0065] S5. Obtain the cluster center point based on the cluster route overlap rate, and divide the parameter ranges of various multi-dimensional operation and inspection abnormal data in different time allocation projects.

[0066] In specific use, this solution provides a method for processing abnormal data of power grid operation and inspection with time-sharing data update. The specific contents are as follows:

[0067] First, during the data collection process, since the types of abnormal power grid operation and inspection data are diverse and will be affected by human and environmental factors, data division is required in advance. This solution divides the abnormal power grid operation and inspection data into multiple dimensions and matches them with the project. Figure 2 As shown, the multi-dimensional operation and inspection abnormal data matching items are equipment model, time point and abnormal item value. In order to distinguish the changes of various abnormal data in different time periods, the multi-dimensional operation and inspection abnormal data are reorganized into a unified sequence according to the time axis, that is, the data are bound by the time axis. The abnormal item values include different types of abnormal values that affect the operation of the power grid, such as temperature, voltage and output power. Values that exceed the normal range are divided according to different abnormal item values during the collection process. For example, in the process of predicting the temperature abnormal range, the target data is temperature, and the corresponding abnormal data matching items are equipment model-time point-temperature value. By collecting temperature values at different time points for analysis and processing, the temperature value change pattern at different times is obtained;

[0068] In order to distinguish the changes of abnormal values at different times, time allocation items are divided. The time allocation items include different times on the same day and different days. Different times on the same day means that abnormal values are collected at different time points on the same day. For example, for the collection of abnormal temperature inside power equipment, the collection time interval is 4 hours, the initial collection time is 8:00, and the corresponding time points are 8:00, 12:00, 16:00, 20:00 and 24:00. The abnormal temperature is collected respectively to obtain the corresponding temperature abnormality values;

[0069] The "simultaneous different days" refers to collecting abnormal values at the same time point on different days. For example, for the abnormal data collection of the output power of the power grid bus, the collection time is 8:00 every day, and the data is collected every two days.

[0070] It is worth noting that in order to improve the prediction accuracy of the variation range of each abnormal data, the same abnormal data is collected through two time allocation projects;

[0071] In the process of dividing the range of abnormal data, this solution uses spatial point clustering to cluster abnormal data, which is processed by the k-means clustering algorithm. Figure 5 The specific steps are as follows:

[0072] First, collect the individual data of the same abnormal item at different times, that is, the value size at different time points, and count the individual data to generate a data set;

[0073] Randomly select individual data from the data set as the cluster center, calculate the distance between each individual data and the cluster center, and assign each individual data to the nearest cluster center to generate each concentrated cluster, and calculate the average value of the individual data in the concentrated cluster to obtain a new cluster center, continue to calculate the distance between each individual data and the cluster center, and assign each individual data to the nearest cluster center;

[0074] Continue to update the average value of each cluster and repeat the above steps until each cluster no longer changes;

[0075] Depend on Figure 3 As shown in the figure, the temperature anomaly data of the power equipment is data. The anomaly values collected at different times on the same day are 1-19 respectively. The first clustering process is performed, where N1 is the cluster center of the first concentrated cluster and N2 is the cluster center of the second concentrated cluster. After the distance calculation, the difference between the temperature anomaly value and the cluster center value is calculated. The individual data are assigned to N1 and N2 according to the difference size to form concentrated clusters, as shown in the figure. Figure 2 As shown, individual data 1, 2, 6, 7, 8, 11, 12, 13, 14, 15, and 19 are allocated to the centralized cluster N1, and individual data 3, 4, 5, 9, 10, 16, 17, and 18 are allocated to the centralized cluster N2;

[0076] like Figure 4As shown, at this time, since each cluster can still be divided, multiple clustering processes are required (in this example, the process is completed by two clustering processes), where cluster center N3 is the average value of the individual data in cluster center N1, and cluster center N4 is the average value of the individual data in cluster center N2. At this time, the difference between individual data 1, 2, 6, 7, 8, 11, 12, 13, 14, 15 and 19 and cluster center N3 and cluster center N1 is calculated twice, and the difference between individual data 3, 4, 5, 9, 10, 16, 17 and 18 and cluster center N4 and cluster center N2 is calculated, and they are distributed according to the size of the difference. The difference comparison obtains the distance length. After the second clustering, 1, 2, 6, 13, 15 and 19 are still kept in the concentrated cluster formed by the cluster center N1, 7, 8, 11, 12 and 14 are divided into the concentrated cluster formed by the cluster center N3, and individual data 4, 9, 10 and 18 are still kept in the concentrated cluster formed by the cluster center N2. 3, 16 and 17 are divided into the concentrated cluster N4 formed by the cluster center N3. At this time, the concentrated cluster is continued to be updated and it is found that the data amount of each concentrated cluster does not change, which indicates that the clustering process is completed, forming the final concentrated clusters N1, N2, N3 and N4;

[0077] Due to the differences in random cluster centers, when minimizing the cost function, it is possible to stay at a local minimum, that is, the value of the initially selected cluster center is too small. When the cluster center is divided twice, it will stay in the concentrated cluster and cannot be clustered again, resulting in poor final clustering effect.

[0078] In order to solve the above problems, Figure 6 as well as Figure 9 As shown in the figure, in the process of selecting cluster centers, we first need to select z points with the farthest batch distance as possible (the number of cluster centers determined in advance by the management staff), and randomly select one as the initial cluster center. Then, we select the point farthest from the k points as the initial cluster center, and then select the point with the largest distance from the first two points as the third initial cluster center, and so on, until z initial cluster centers are selected, so as to avoid local concentrated clusters from staying.

[0079] The clustering algorithm is used to implement the above solution. The algorithm formula is as follows:

[0080]

[0081] Where F(iter) is the objective function value, is the abnormal value corresponding to different data, z j (iter) is the cluster center value of each cluster, i represents the data subscript, distinguishing different data, j represents the cluster center subscript, distinguishing different cluster centers;

[0082] The first step is to first give a data set of size n, let iter = 1, iter represents the number of iterations, that is, the number of clustering, and select k initial cluster centers z j (iter), where j = 1, 2, ..., k;

[0083] Step 2: Calculate each data x i , i = 1, 2, ..., n and the distance between the cluster center, x i Assigned to the nearest cluster center z j The cluster to which (iter) belongs, i.e. |x i -z j (iter)|≤|x i -z j !(iter)|, where j ！ is the subscript of another cluster center, i.e. j ！ ≠j,j ！ , j∈(1,2,…,k);

[0084] Step 3: Let iter = iter + 1, calculate the new cluster center (take the average value of the clustered values), and calculate the objective function value F(iter);

[0085] Step 4. Judgment: If |F(iter+1)-F(iter)|<θ (θ is the convergence constant) or there is no category change in the concentrated cluster, the algorithm ends, otherwise return to step 2.

[0086] Since the abnormal values corresponding to each abnormal data are often related to time changes, Figure 7 As shown in , in order to analyze the changing rules of abnormal data corresponding to different time points, this solution obtains the clustering routes of each abnormal data in the clustering process. First, it records the clusters and corresponding cluster centers that each abnormal data passes through during the entire clustering process. Then, according to the order in which it is divided into each cluster, the corresponding routes are obtained in combination with the corresponding cluster centers, as shown in Figure 3-Figure 4 As shown, the clustering route for the abnormal temperature value 7 is N1-N3, and the clustering route for the abnormal temperature value 9 is N1-N3;

[0087] In order to obtain the regularity between abnormal data and time points, the overlap rate threshold of clustering routes is defined. For example, the clustering routes of abnormal temperature value 7 and abnormal temperature value 9 are both N1-N3, that is, the overlap rate between the two is 100%, which exceeds the overlap rate threshold. At the same time, abnormal temperature value 7 and abnormal temperature value 9 are abnormal temperature values collected at the same time on different days. Figure 8As shown, at this time, the concentrated clusters that the clustering route passes through are obtained, the cluster centers corresponding to the concentrated clusters are obtained, and they are arranged according to size. The minimum and maximum values are selected and combined into the cluster center range. The cluster center range is used as the parameter range of the abnormal data at different times on the same day;

[0088] It is worth noting that 10 sets of temperature anomaly data were collected at different times on the same day, namely M1, M2, M3, M4, M5, M6, M7, M8, M9 and M 10 , among which M1, M2, M3, M4 and M7 have all passed through the clusters m1-m6, with a corresponding overlap rate of 50%, among which M5, M6 and M 10 Both routes pass through clusters m1-m7, with an overlap rate of 30%. M8 and M9 also pass through clusters m1-m8, with a corresponding overlap rate of 20%. If the overlap rate threshold is 40%, only cluster route m1-m6 meets the criteria, and the final cluster route is m1-m6.

[0089] At the same time, in the clustering process, as the amount of individual data continues to increase, the extreme values will be divided into remote centralized clusters, and the extreme values will be automatically distinguished to avoid the interference of extreme values on various multi-dimensional operation and inspection abnormal data, thereby improving the accuracy of parameter range division. At the same time, by dividing the extreme value range in the centralized cluster that stores extreme values, the extreme value change status of various multi-dimensional operation and inspection abnormal data is determined, and the parameter range is updated in time to further improve the accuracy of parameter range division.

[0090] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing abnormal data of power grid operation and inspection, characterized in that: The steps include: S1. Collect multi-dimensional operation and inspection abnormal data during the power grid operation and inspection process, and obtain multi-dimensional operation and inspection abnormal data matching items; S2. Reorganize the time series of multi-dimensional operation and inspection abnormal data and integrate them into a unified sequence according to the time axis; S3. Divide the time allocation projects and perform spatial point clustering on the multi-dimensional operation and inspection abnormal data of different time allocation projects; S4. Obtain clustered routes of multi-dimensional inspection abnormal data for different time allocation projects and calculate the clustered route overlap rate; S5. Obtain the cluster center point based on the cluster route overlap rate, and divide the parameter ranges of various multi-dimensional operation and inspection abnormal data in different time allocation projects.

2. The method for processing abnormal data of power grid operation and inspection according to claim 1, characterized in that: The abnormal data matching items in S1 are the equipment model, time point, and the value of the abnormal item.

3. The method for processing abnormal data of power grid operation and inspection according to claim 1, characterized in that: The method of integrating into a unified sequence according to the time axis in S2 includes the following steps: S2.

1. Define each timeline according to the industry standard for abnormal data collection; S2.

2. Obtain the device model and bind the values of the abnormal items collected in the device model through the time axis.

4. The method for processing abnormal data of power grid operation and inspection according to claim 1, characterized in that: The time allocation items in S3 include the same day but different times and the same time but different days.

5. The method for processing abnormal data of power grid operation and inspection according to claim 4, characterized in that: The method for performing spatial point clustering in S3 comprises the following steps: S3.

1. Collect individual data of the same abnormal item at different times, count the individual data, and generate a data set; S3.

2. Randomly select individual data from the data set as the cluster center, calculate the distance between each data and the cluster center, and assign each data to the nearest cluster center to generate clusters; S3.

3. Calculate the average value of the individual data in the cluster to obtain a new cluster center, continue to calculate the distance between each individual data and the cluster center, and assign each individual data to the nearest cluster center; S3.

4. Continue to update the average value of each cluster and repeat the above steps until each cluster no longer changes.

6. The method for processing abnormal data of power grid operation and inspection according to claim 5, characterized in that: The optimization method for performing spatial point clustering in S3 includes the following steps: S3.

5. Select z points with the greatest possible distance from each other in the batch and randomly select one as the initial cluster center. S3.

6. Select the point farthest from the z points as the initial cluster center, and then select the point with the largest distance from the first two points as the third initial cluster center, until z initial cluster centers are selected.

7. The method for processing abnormal data of power grid operation and inspection according to claim 5, characterized in that: In S3.4, it is determined that each concentrated cluster no longer changes using a clustering algorithm, and the algorithm formula is as follows: Where F(iter) is the objective function value, is the abnormal value corresponding to different data, z j (iter) is the cluster center value of each cluster, i represents the data subscript, distinguishing different data, j represents the cluster center subscript, distinguishing different cluster centers. The specific steps are as follows: The first step is to first give a data set of size n, let iter = 1, iter represents the number of iterations, and select k initial cluster centers z j (iter), where j = 1, 2, ..., k; Step 2: Calculate each data x i , i = 1, 2, ..., n and the distance between the cluster center, x i Assigned to the nearest cluster center z j The cluster to which (iter) belongs, i.e. where j ！ is the subscript of another cluster center, i.e. j ！ ≠j,j ！ , j∈(1,2,…,k); Step 3: Let iter = iter + 1, calculate the new cluster center, and calculate the objective function value F(iter); Step 4. Judgment: If |F(iter+1)-F(iter)|<θ, θ is the convergence constant, or there is no category change in the concentrated cluster, the algorithm ends, otherwise return to step 2.

8. The method for processing abnormal data of power grid operation and inspection according to claim 1, characterized in that: The method for calculating the cluster route overlap rate in S4 comprises the following steps: S4.

1. Record the clusters and corresponding cluster centers of each abnormal data during the entire clustering process. S4.

2. Obtain the corresponding routes based on the order in which the clusters are divided and the corresponding cluster centers; S4.

3. Obtain the overlap rate of the routes through which each abnormal data item passes, and select the route with the highest overlap rate among the clustered routes as the comparison route; S4.

4. Define the overlap rate threshold of clustered routes; When the coincidence rate of the comparison route passed by the abnormal data is greater than or equal to the coincidence rate threshold, the coincidence rate is marked as the coincidence rate matched by the abnormal data, and the corresponding clustering route is obtained; When the coincidence rate of the comparison route passed by the entire abnormal data is less than the coincidence rate threshold, it indicates that the abnormal data is not related to time, and the abnormal value range of the abnormal data is obtained as the final parameter range.

9. The method for processing abnormal data of power grid operation and inspection according to claim 1, characterized in that: The method of dividing the parameter ranges of each multi-dimensional operation inspection abnormal data in different time allocation items in S5 includes the following steps: S5.

1. Obtain the clusters that the clustering route passes through and the corresponding cluster centers; S5.

2. Arrange them by size, select the minimum and maximum values, and combine them into the cluster center range.

Images