CN118570334A - A method for reconstructing power load curve - Google Patents

Abstract

The present invention discloses a method for reconstructing a power load curve, and belongs to the field of data processing. The method comprises the following steps: decomposing the original power load curve into multiple intrinsic mode component subsequences by variational mode decomposition; spatially reconstructing all intrinsic mode component subsequences to obtain a corresponding information table, wherein each row corresponds to a row vector composed of data points; for each information table, clustering the row vectors therein respectively, and then performing cross clustering to obtain the final clustering result; dividing each type of cluster into different sets according to the number of elements in each type of cluster in the final clustering result, and interpolating the corresponding rows in different ways respectively; obtaining the component reconstruction curve according to the interpolated information table and superimposing and merging them to obtain the final reconstruction curve. The present invention can better adapt to the complex changes of the curve, greatly improve the accuracy of reconstruction, and does not require training data, and also saves more computing resources.

Description

A method for reconstructing power load curve

Technical Field

The invention belongs to the field of data processing, and in particular relates to a method for reconstructing a power load curve.

Background Art

High-resolution load curves are crucial for in-depth exploration of the laws of load fluctuations. They can not only accurately predict the fluctuations of renewable energy power generation (such as solar and wind power), but also have a significant impact on the consumption of renewable energy and spot electricity trading.

Although high-frequency data collection and uploading can be achieved through hardware, the existing hardware equipment and communication networks need to be modified to support high-frequency data collection tasks and high-speed data transmission. The cost is relatively high and it is mainly suitable for gateway metering data collection. It is difficult to promote for non-gateway metering scenarios.

Given the limitations of hardware transformation, software reconstruction has become the current mainstream solution. Currently, there are two main ways to reconstruct the power load curve:

1. Reconstruction through a single interpolation algorithm: For low-resolution data, interpolation functions (such as Lagrange interpolation, Newton interpolation, etc.) are used to interpolate between data points to obtain a high-resolution load curve. Although this method can maintain the original rules, it will cause large errors when filling in the full amount, affecting the accuracy of load forecasting, and thus adversely affecting the regulation of new energy and the formulation of power grid operation plans. Therefore, it is only suitable for filling in a small amount of missing data.

2. Reconstruction through supervised learning or deep learning methods: This method can theoretically achieve a higher accuracy, but the large amount of high-resolution data samples required for training is difficult to obtain. If other curves are selected as samples, the reconstructed curve will not be smooth enough when the data scale is small, and a large amount of computing resources will be required when the scale is large.

Summary of the invention

The present invention proposes a power load curve reconstruction method, the purpose of which is to solve the problems of poor accuracy, difficulty in obtaining training data, and consumption of a large amount of computing resources in existing reconstruction methods.

The technical solution of the present invention is as follows:

A method for reconstructing a power load curve, comprising the steps of:

Step 1: preprocessing the original power load curve;

Step 2: Decompose the original power load curve into The intrinsic mode component subsequences;

Step 3, spatially reconstruct all the intrinsic mode component subsequences by splitting them, and obtain the information tables corresponding to the extrinsic mode component subsequences; each row in the information table corresponds to each curve segment divided in sequence in the extrinsic mode component subsequence, and each row in the information table contains all the normalized data points in the corresponding curve segment arranged in sequence, that is, each row corresponds to a row vector composed of data points;

Step 4: For each information table, cluster the row vectors therein to obtain The row vector clustering results corresponding to each of the intrinsic mode component subsequences;

Step 5: Perform cross clustering on the row vector clustering results of all intrinsic mode component subsequences to obtain the final clustering results;

Step 6: Divide each cluster into sets according to the number of elements in each cluster in the final clustering result and collection ： If the number of elements in a cluster in the final clustering result is 1, then the cluster is taken as a set , otherwise the cluster is treated as a set The subclass cluster of

Step 7: Interpolate each row of all information tables. If a row corresponds to a set If a row corresponds to a subclass cluster in the set, the row uses linear interpolation between adjacent data points; If it is an element of a subclass cluster in , the row uses nearest neighbor interpolation between adjacent data points;

Step 8. Denormalize each row in the interpolated information table respectively, then combine the data points of each row in the information table into a curve segment corresponding to the row, and then combine the curve segments of each row in the information table in sequence into a component reconstruction curve of the inherent modal component subsequence corresponding to the information table, and finally superimpose and merge all the component reconstruction curves to obtain the final reconstructed curve.

As a further improvement of the power load curve reconstruction method: in step 1, the preprocessing includes deleting abnormal data and filling in missing data.

As a further improvement of the power load curve reconstruction method: in step 3, it is assumed that each information table contains rows, each containing data points;

The normalization is row-by-row normalization. The first The original value of the data point is , then the normalized value of this point is:

.

As a further improvement of the power load curve reconstruction method: in step 5, the cross-clustering process is: the clustering result obtained by intersecting the row vector clustering result of the first eigenmode component subsequence and the row vector clustering result of the second eigenmode component subsequence is used as the intersected clustering result corresponding to the second eigenmode component subsequence, and starting from the third eigenmode component subsequence, the clustering result of the current eigenmode component subsequence is intersected with the intersected clustering result of the previous eigenmode component subsequence to obtain the intersected clustering result of the current eigenmode component subsequence, and the intersected clustering result of the last eigenmode component subsequence is used as the final clustering result.

As a further improvement of the power load curve reconstruction method: in step 5, for any two clustering results, the intersection process is: traverse the clusters in the first clustering result, for each cluster: intersect it with all the clusters in the second clustering result respectively, and then combine the intersection results into the intersection cluster set corresponding to the cluster; then merge the intersection cluster sets corresponding to all the original clusters in the first clustering result to obtain the intersection clustering result of the two clustering results.

As a further improvement of the power load curve reconstruction method, in step 7, the linear interpolation process is: assuming that the linear interpolation between adjacent data points is data points, the previous data point is the interpolation value before the row data points, with values , the next data point is the first data point before the interpolation of the line data points, with values , then the inserted The value of the new data point is .

As a further improvement of the power load curve reconstruction method, in step 7, the linear interpolation process is:

For any information table, assume that the row currently being interpolated corresponds to the set A subclass cluster in An element in the subclass cluster The rows corresponding to all elements in are used as the operation rows of the information table;

When interpolating the area between any two adjacent data points in the above operation row, let the previous data point be the first data point before the interpolation of the row. data points, and the next data point is the first data point before the interpolation of the row data points, then set the search range to the original Data points and The value of the data point;

Along from Data points to When inserting the interpolation point in the direction of the data points, the value with the second closest Euclidean distance to the previous original data point or interpolation point in the search range is used as the value of the current interpolation point until the nearest neighbor interpolation operation of all operation rows is completed.

As a further improvement of the power load curve reconstruction method, in step 8, the denormalization process is:

Suppose that after interpolation of a certain information table The first The data point value is , then the value of the point after denormalization is:

;

in, and They are the maximum and minimum values of the data points in this row before normalization.

Compared with the prior art, the present invention has the following positive effects:

The present invention decomposes the original curve into multiple frequency components through modal decomposition to ensure the non-stationarity and nonlinear adaptability of the curve, and then finely captures the characteristics of each component and each segment through segment processing, and then divides each segment into two cases of linear interpolation and nearest neighbor interpolation through clustering and cross-classification, making full use of the characteristic information of each component to mine the global and local characteristics of the curve, and selecting the most suitable interpolation method for each segment for interpolation. Compared with a single interpolation method, the present invention can better adapt to the complex changes of the curve and greatly improve the accuracy of reconstruction; compared with supervised learning or deep learning methods, the present invention does not require training data, nor does it require complex calculations with reference to other large-scale data samples, which saves more computing resources.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flow chart of the method of the present invention.

DETAILED DESCRIPTION

The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

As shown in FIG1 , a method for reconstructing a power load curve includes the following steps:

Step 1: Preprocess the original power load curve, including deleting abnormal data (such as data with obviously large change amplitude) and filling in missing data (using linear interpolation or other methods).

Generally, the data frequency of the original power load curve can be 15 minutes, 1 hour or 1 day. This embodiment takes the data of about 7 days with a frequency of 1 hour as an example.

Step 2: Decompose the original power load curve into A subsequence of intrinsic mode components: ,in For the In this embodiment, the eigenmode component subsequences are decomposed into 8 components.

The advantage of VMD technology lies in its adaptability. It can determine the number of modal decompositions according to the characteristics of the actual signal, and adaptively match the optimal center frequency and limited bandwidth of each mode during the decomposition process, which can effectively avoid modal aliasing during the modal decomposition process.

Step 3: Reconstruct all intrinsic modal component subsequences by splitting them into two parts, and obtain the information table corresponding to each intrinsic modal component subsequence, which can be recorded as Each row in the information table corresponds to each curve segment divided in sequence in the natural mode component subsequence, and each row in the information table contains all the normalized data points in the corresponding curve segment in sequence.

Assume that each information table contains rows, each containing data points, Indicates the information table In the row The row vector of data points.

The normalization is row-by-row normalization. The first The original value of the data point is , then the normalized value of this point is:

.

Step 4: For each information table, cluster the row vectors therein respectively, so as to obtain the row vector clustering results corresponding to all the intrinsic mode component subsequences.

In this embodiment, there are 8 intrinsic modal component subsequences, each of which contains 9 row vectors. The row vector clustering result of the intrinsic mode component subsequences is , then the row vector clustering results of each intrinsic mode component subsequence can be expressed as:

;

;

;

.....

.

In this embodiment, k-means is used for clustering.

Step 5: Perform cross clustering on the row vector clustering results of all intrinsic mode component subsequences to obtain the final clustering results.

The cross-clustering process is as follows: the clustering result obtained by intersecting the row vector clustering result of the first eigenmodal component subsequence and the row vector clustering result of the second eigenmodal component subsequence is used as the intersected clustering result corresponding to the second eigenmodal component subsequence; starting from the third eigenmodal component subsequence, the clustering result of the current eigenmodal component subsequence is intersected with the intersected clustering result of the previous eigenmodal component subsequence to obtain the intersected clustering result of the current eigenmodal component subsequence; and the intersected clustering result of the last eigenmodal component subsequence is used as the final clustering result.

Among them, for any two clustering results, the intersection process is: traverse the clusters in the first clustering result, for each cluster: intersect it with all the clusters in the second clustering result respectively, and then combine the intersection results into the intersection cluster set corresponding to the cluster; then merge the intersection cluster sets corresponding to all the original clusters in the first clustering result to obtain the intersection clustering result of the two clustering results.

For example, the above and The intersection process is: first The first cluster in and The three clusters in intersect with each other, and the three intersection results are , empty set, empty set, The set of clusters after intersection corresponding to the first cluster in is Similarly, The second cluster in and The three clusters in intersect with each other, and the obtained cluster set after intersection is , The set of clusters after the intersection of the second cluster in is , and The clustering result after intersection is . Then and Intersect ,Will and Intersect , and so on to get the final clustering result .

Step 6: Divide each cluster into sets according to the number of elements in each cluster in the final clustering result and collection ： If the number of elements in a cluster in the final clustering result is 1, then the cluster is taken as a set , otherwise the cluster is treated as a set subclass cluster of .

In this embodiment, set ,gather Obviously, the set The union of all subclass clusters in is .

Step 7: Interpolate each row of all information tables. If a row corresponds to a set If a row corresponds to a subclass cluster in the set, the row uses linear interpolation between adjacent data points; If a row is an element of a subclass cluster in , the nearest neighbor interpolation is used between adjacent data points.

(1) The linear interpolation process is:

Assume that we interpolate between adjacent data points data points, the previous data point is the interpolation value before the row data points, with values , the next data point is the first data point before the interpolation of the line data points, with values , then the inserted The value of the new data point is .

(2) The nearest neighbor interpolation process is:

For any information table, assume that the row currently being interpolated corresponds to the set A subclass cluster in An element in the subclass cluster The rows corresponding to all elements in the table are used as operation rows of the information table. For example, the subclass cluster , then the 2nd and 6th rows are used as operation rows.

When interpolating the area between any two adjacent data points in the above operation row, let the previous data point be the first data point before the interpolation of the row. data points, and the next data point is the first data point before the interpolation of the row data points, then set the search range to the original Data points and In this embodiment, it is assumed that the interpolation is performed between the third and fourth data points, and the search range is the values of the third and fourth data points in the second and sixth rows, for a total of four values. Data points to When inserting interpolation points in the direction of the data points, the value in the search range that is the second closest to the value of the previous original data point or interpolation point in the Euclidean distance is used as the value of the current interpolation point until the nearest neighbor interpolation operation of all operation rows is completed. For example, when inserting the first interpolation point between the third and fourth data points in the second row, the value in the search range that is the second closest to the value of the third data point (original) in the row is used as the value of the current interpolation point. When inserting the second interpolation point, the value in the search range that is the second closest to the value of the first interpolation point in the Euclidean distance is used as the value of the current interpolation point, and so on.

Step 8. Denormalize each row in the interpolated information table respectively, then combine the data points of each row in the information table into a curve segment corresponding to the row, and then combine the curve segments of each row in the information table in sequence into a component reconstruction curve of the inherent modal component subsequence corresponding to the information table, and finally superimpose and merge all the component reconstruction curves to obtain the final reconstructed curve.

The denormalization process is:

Suppose that after interpolation of a certain information table The first The data point value is , then the value of the point after denormalization is:

.

in, and They are the maximum and minimum values of the data points in this row before normalization.

It should be noted that, for those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention.

Claims (8)

1. A method of reconstructing an electrical load curve, comprising the steps of:

Step 1, preprocessing an original power load curve;

step 2, decomposing the original power load curve into components through variation modal decomposition A sub-sequence of intrinsic modal components;

step 3, carrying out space reconstruction on all the intrinsic mode component subsequences in a splitting mode to obtain information tables corresponding to the intrinsic mode component subsequences respectively; each row in the information table sequentially corresponds to each curve segment which is divided in sequence in the inherent modal component subsequence, and each row in the information table comprises all sequentially arranged normalized data points in the corresponding curve segment, namely, each row corresponds to a row vector consisting of the data points;

Step 4, clustering the row vectors in each information table to obtain Row vector clustering results corresponding to the intrinsic mode component sub-sequences respectively;

step 5, cross-clustering is carried out on row vector clustering results of all the intrinsic mode component subsequences to obtain a final clustering result;

step 6, dividing each cluster into sets according to the element number of each cluster in the final clustering result Sum set: If the element number of a cluster of a certain class in the final clustering result is 1, the cluster of the class is taken as a setIf not, the cluster is used as a setIs a sub-cluster of (2);

Step 7, respectively interpolating each row of all information tables, if a row corresponds to a set If the row is a sub-cluster element, then the row adopts linear interpolation between adjacent data points; if a certain line corresponds to a setIf one element of a sub-cluster in the row is a sub-cluster, the row adopts neighbor interpolation between adjacent data points;

And 8, respectively carrying out inverse normalization on each row in the information table after interpolation, then forming the data points of each row in the information table into curve segments corresponding to the row, sequentially forming the curve segments of each row in the information table into component reconstruction curves of the intrinsic mode component subsequence corresponding to the information table, and finally superposing and merging all the component reconstruction curves to obtain a final reconstruction curve.

2. The electrical load curve reconstruction method as set forth in claim 1, wherein: in step 1, preprocessing includes deleting abnormal data and filling the deleted data.

3. The electrical load curve reconstruction method as set forth in claim 1, wherein: in step 3, each information table is provided to includeRows, each row comprisingData points;

The normalization refers to normalization according to rows, and the first is set Line 1The original value of the data point isThe normalized value at this point is:

。

4. The electrical load curve reconstruction method as set forth in claim 1, wherein: in step 5, the cross clustering process is as follows: and taking a clustering result obtained by intersecting the row vector clustering result of the first natural mode component sub-sequence and the row vector clustering result of the second natural mode component sub-sequence as an intersecting clustering result corresponding to the second natural mode component sub-sequence, starting from the third natural mode component sub-sequence, intersecting the clustering result of the current natural mode component sub-sequence with the intersecting clustering result of the previous natural mode component sub-sequence to obtain an intersecting clustering result of the current natural mode component sub-sequence, and taking the intersecting clustering result of the last natural mode component sub-sequence as the final clustering result.

5. The electrical load curve reconstruction method as set forth in claim 4, wherein: in step 5, for any two clustering results, the intersecting process is as follows: traversing the class clusters in the first clustering result, and for each class cluster: intersecting the clustering result with all class clusters in the second clustering result respectively, and combining the intersecting results into intersected class clusters corresponding to the class clusters; and then merging the intersected cluster groups corresponding to all the original cluster groups in the first clustering result to obtain intersected clustering results of the two clustering results.

6. The method of reconstructing an electrical load curve according to claim 1, wherein in step 7, the linear interpolation process is: suppose that interpolation between adjacent data pointsData point, the previous data point is the first data point before interpolation of the lineThe data points and values areThe next data point is the first data point before the row interpolationThe data points and values areThen insert the firstThe value of each new data point is。

7. The method of reconstructing an electrical load curve according to claim 1, wherein in step 7, the linear interpolation process is:

For any information table, assume that the row currently being interpolated corresponds to a set Some sub-cluster of (a)One element of the subclass clusterThe row corresponding to all elements in the information table is used as an operation row of the information table;

in the case of interpolating the region between any two adjacent data points in the operation line, the previous data point is set to be the first data point before the line is interpolated Data point, the latter data point is the first data point before the line interpolationData points, the searching range is set as the original first operation line in the information tableData point and the firstValues of the individual data points;

Edge slave first Data points to the firstWhen the direction of the data points is sequentially inserted into the interpolation points, the value which is the second closest to the Euclidean distance of the value of the previous original data point or the interpolation point in the search range is taken as the value of the current interpolation point until the neighbor interpolation operation of all operation lines is completed.

8. The method for reconstructing an electrical load curve according to any one of claims 1 to 7, wherein in step 8, the inverse normalization process is:

Interpolation of a certain information table is set Line 1Data point value isThe value after the inverse normalization of the point is:

；

Wherein, AndThe maximum and minimum values of the data points before normalization of the line, respectively.