A migration source selection method and system for short-term load migration prediction

Abstract

The embodiment of the specification provides a migration source selection method and system for short-term load migration prediction, wherein the method comprises the following steps: S1. Preprocessing migration source data according to the length of target source data to obtain a subsequence with the same length as the target source data; S2. Measuring the optimal transmission distance WD between the migration source and the target source by using the Wasserstein distance for each subsequence, and calculating the maximum information coefficient MIC between the migration source and the target source by using the maximum information coefficient method; S3. Constructing a WD-MIC curve with WD as the abscissa and MIC as the ordinate; and S4. Selecting the migration source with the maximum similarity as the final migration source by calculating the area under the WD-MIC curve as the similarity between the migration source and the target source. The application can effectively avoid negative migration in multi-source migration prediction.

Description

Technical Field

[0001] This document relates to the field of short-term load forecasting technology, and in particular to a method and system for selecting migration sources for short-term load migration forecasting. Background Technology

[0002] With the rapid development of distributed energy resources, the temporal and spatial instability of power grid loads makes load forecasting more difficult. Distributed energy resources provide more flexible power sources and can better address load fluctuations and uncertainties, which necessitates accurate short-term load forecasting (STLF) to provide crucial decision-making support. For example, CN115829161A proposes a distributed short-term load forecasting method that considers holiday distinctions. It uses historical load and time information as model inputs, providing reasonable and crucial inputs to the neural network model, and forecasts holiday and non-holiday loads separately to further improve the accuracy of power load forecasting. CN115689026A proposes a residential load short-term forecasting method and system. It decomposes the load value to be predicted into a basic load curve and a difference value. The basic load curve is constructed based on the periodicity and effectiveness of residential loads, effectively reducing the difference between peak and average loads and improving forecast accuracy. By combining self-attention and external attention mechanisms and training with large datasets, the model can more effectively extract time-series features.

[0003] Currently, STLF methods can be divided into traditional methods and machine learning methods. Machine learning methods outperform traditional methods in terms of accuracy and stability in short-term load forecasting. Machine learning models are driven by large amounts of historical load data; however, when it comes to actual forecasting tasks, obtaining complete and large-scale data is difficult due to user data privacy issues and potential system failures of energy companies. This is likely a major obstacle for data-driven machine learning models, thus affecting their accuracy and stability. CN115422994A proposes a cross-city time-series data migration forecasting method and system. It uses a sliding time window technique to process time-series data from both the source and target cities, then uses a two-layer 1DCNN model combined with a two-layer GRU model to extract features from the time-series set. Finally, a pre-trained adaptive forecasting model is used to process the extracted features of the time-series set to obtain the time-series data migration forecasting results.

[0004] However, the selection of migration sources plays a decisive role in the accuracy of multi-source migration forecasting. Due to the complexity of the probability distributions of the time series of the target and source domains, traditional migration source selection algorithms based on linear and nonlinear analysis ignore the probability distribution characteristics of the time series between the target and source domains, inevitably leading to negative migration. For this reason, existing migration source selection algorithms struggle to significantly improve the accuracy of short-term load forecasting when data is missing.

[0005] Multi-source migration forecasting is an effective technique for addressing short-term load forecasting (STLF) in situations where data is scarce, and the selection of migration sources plays a decisive role in its accuracy. Due to the complexity of the probability distributions of the time series in the target and source domains, traditional migration source selection algorithms based on linear and nonlinear analysis ignore the probability distribution characteristics of the time series between the target and source domains, inevitably leading to negative migration. Summary of the Invention

[0006] This specification provides one or more embodiments of a migration source selection method for short-term load migration forecasting, including:

[0007] S1. Preprocess the source data according to the length of the target source data to obtain a subsequence of the same length as the target source data;

[0008] S2. For each subsequence, the optimal transmission distance WD between the migration source and the target source is measured by the Wasserstein distance metric, and the maximum information coefficient MIC between the migration source and the target source is calculated using the maximum information coefficient method.

[0009] S3. Construct the WD-MIC curve with WD as the x-axis and MIC as the y-axis;

[0010] S4. The area under the WD-MIC curve is used as the similarity between the migration source and the target source. The migration source with the highest similarity is selected as the final migration source.

[0011] This specification provides one or more embodiments of a migration source selection system for short-term load migration forecasting, including:

[0012] Data processing module: used to preprocess the source data to obtain a subsequence of the same length as the target source data, based on the length of the target source data.

[0013] Data calculation module: used to calculate the maximum information coefficient (MIC) between the migration source and the target source for each subsequence using the Wasserstein distance metric (WD) and the maximum information coefficient method.

[0014] Curve Construction Module: Used to construct a WD-MIC curve with WD as the x-axis and MIC as the y-axis;

[0015] Migration source selection module: This module calculates the area under the WD-MIC curve as the similarity between the migration source and the target source, and selects the migration source with the highest similarity as the final migration source.

[0016] This specification provides one or more embodiments of an electronic device, including:

[0017] Processor; and,

[0018] A memory is configured to store computer-executable instructions, which, when executed, cause the processor to implement the steps of the migration source selection method described above for short-term load migration forecasting.

[0019] This specification provides one or more embodiments of a storage medium for storing computer-executable instructions that, when executed, implement the steps of the migration source selection method described above for short-term load migration forecasting.

[0020] The embodiments of this invention can be used to accurately screen similar nodes from migration sources, train machine learning models, and propose the WM algorithm by combining similarity analysis algorithms of Wasserstein distance and maximum information coefficient (MIC). By integrating the similarity of nonlinearity and probability distribution, the similarity between time series variables is measured. Using the WM algorithm to determine migration sources can effectively avoid negative migration in multi-source migration prediction. The WM algorithm can improve the robustness of multi-source migration models.

[0021] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating a migration source selection method for short-term load migration forecasting, provided for one or more embodiments of this specification;

[0024] Figure 2A detailed flowchart illustrating a migration source selection method for short-term load migration forecasting provided in one or more embodiments of this specification;

[0025] Figure 3 A schematic diagram of the load sequence of Greek natural gas pipeline network nodes in the specific embodiments provided in this specification;

[0026] Figure 4 The specific embodiments provided show the fitting degree of the prediction results of the top one to top ten nodes ranked by different similarity measurement algorithms.

[0027] Figure 5 The diagram shows the prediction result curves of each model in the specific embodiments provided in this specification.

[0028] Figure 6 A schematic diagram illustrating the composition of a migration source selection system for short-term load migration forecasting, provided for one or more embodiments of this specification;

[0029] Figure 7 This is a schematic diagram of the structure of an electronic device provided for one or more embodiments of this specification. Detailed Implementation

[0030] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0031] Method Implementation Examples

[0032] According to embodiments of the present invention, a method for selecting migration sources for short-term load migration forecasting is provided. Figure 1 This document provides a flowchart of a migration source selection method for short-term load migration forecasting, as illustrated in one or more embodiments. Figure 2 A detailed flowchart illustrating a migration source selection method for short-term load migration forecasting, provided in one or more embodiments of this specification, is shown below. Figure 1 and Figure 2 As shown, a migration source selection method for short-term load migration forecasting according to an embodiment of the present invention specifically includes:

[0033] S1. Preprocess the source data to obtain a subsequence of the same length as the target source data based on the length of the target source data.

[0034] S2. For each subsequence, the optimal transmission distance WD between the migration source and the target source is measured by the Wasserstein distance metric, and the maximum information coefficient MIC between the migration source and the target source is calculated using the maximum information coefficient method.

[0035] Wasserstein distance (WD) is a method for measuring the distance between two probability distributions. The specific method for measuring the optimal transmission distance (WD) between the migration source and the target source for each subsequence using Wasserstein distance is as follows:

[0036] The minimum average distance required to move a subsequence from the migration source to the target source is calculated as follows:

[0037]

[0038] Where μ is the probability distribution of the source subsequence sample x, ν is the probability distribution of the target source sequence sample y, X is the source space, and Y is the target source space; f(x,y) represents the distance function between x and y, ∏(μ,υ) is the set of probabilistic couplings of π pairs (μ,υ); p is the order of WD, and when p=1, WD is Earth Mover's distance (EMD);

[0039] The WD can be optimized using linear programming or gradient descent. When p=1, the WD is as follows:

[0040] W1(μ,υ)=infE X×Y～π [||xy||];

[0041] Among all possible probabilistic couplings, the expected lower bound of the distance between samples x and y is the Wasserstein-1 distance between μ and υ; W1(μ,υ) is the average EMD of the probability mass function;

[0042] To evaluate the similarity of the load series, the similarity between the migration source and target source sequences is evaluated by calculating the normalized WD. The calculation method is as follows:

[0043]

[0044] Among them, W norm Let WD be the normalized optimal transmission distance, and maxW1 be the x and y values ​​of the maximum EMD of the source and target source sequences.

[0045] The maximum information coefficient (MIC) between the migration source and the target source is calculated using the maximum information coefficient method. The specific method is as follows:

[0046] Given two load sequences with n elements: based on the source subsequence sample x and the target source sequence sample y, obtain a finite set D = {(x...} i ,y i |i=1,...,n};

[0047] Based on grid G, x of D i Divide the values ​​into x groups, and assign the y values ​​of D to each group. i The values ​​are divided into y groups, and the mutual information MI between the two variables in each group is calculated. The maximum MI value is selected as the MIC value.

[0048] The specific calculation method for MIC is as follows:

[0049]

[0050] Where D = (x, y) is a set of n ordered pairs X and Y, x ∈ X and y ∈ Y; the data space is divided into XY grids, grouping the values ​​of x and y, and B(n) = n α It is a function of the n sample size, representing the upper limit of the search grid size; α is usually set to 0.6; I(D, X, Y) is the maximum mutual information of all XY grids; MIC values ​​range from 0 to 1, with higher values ​​indicating stronger correlation between load series, a value of 0 indicating no correlation, and a value of 1 indicating perfect correlation.

[0051] S3. Construct the WD-MIC curve with WD as the x-axis and MIC as the y-axis.

[0052] The specific method is as follows:

[0053] Let the normalized WD and MIC be denoted as x. i and y i This yields the sequence {(x1, y1), (x2, y2), ..., (x... n y n )}, sort this sequence by x i Sort the elements (i = 1, 2, ..., n) in ascending order. The sorted sequence is denoted as {(X1, Y1), (X2, Y2), ..., (X... n Y n Using WD as the x-axis and MIC as the y-axis, a two-dimensional curve is plotted on the coordinates constructed by WD and MIC to construct the WD-MIC curve.

[0054] S4. The area under the WD-MIC curve is used as the similarity between the migration source and the target source. The migration source with the highest similarity is selected as the final migration source.

[0055] The specific method for calculating the area under the WD-MIC curve is as follows:

[0056]

[0057] Where AUWM is the area under the WD-MIC curve; X and Y represent the x and y axes, J is a coefficient equal to n+1, X0 and Y0 are equal to 0, and X i and Y i Represents the WD and MIC values, where i = 1, 2, ..., n, X n+1 and Y n+1 It equals 1.

[0058] The similarity between the source and target sequences is ranked according to the area under the WD-MIC curve (AUWM), and the source with the largest AUWM is selected as the final source.

[0059] The following describes the specific embodiments of the present invention in further detail using the Greek natural gas pipeline network as an example:

[0060] Daily natural gas load data from 25 nodes within the Greek natural gas pipeline network were collected as the base data, such as... Figure 3 As shown, OIN is the target source for migration prediction; other nodes serve as migration sources.

[0061] Use transfer learning to observe the transfer effect on each node. For example... Figure 4 As shown, this embodiment analyzes the goodness of fit (R2) of the prediction results of the top one to the top ten nodes using different similarity measurement algorithms.

[0062] analyze Figure 4 As can be seen, the WM algorithm did not exhibit any negative transitions in the multi-source migration prediction of the target node, indicating that the algorithm is stable and can effectively reduce the risk of negative migrations during the multi-source migration prediction process. Based on the sensitivity analysis of the prediction results, the first four nodes were selected as the selection nodes for multi-source migration prediction.

[0063] After determining the source nodes for multi-source migration, the multi-source migration prediction results for the target nodes are shown in Table 1:

[0064] Table 1 Predictive performance of the prediction model

[0065]

[0066] Compared with the single LSTM model, the multi-source transfer model obtained by the WM algorithm improves the R2, MAE and MARNE by 22.65%, 2.42% and 2.60%, respectively.

[0067] Plot the prediction curves of each model, and analyze the superiority of the WM algorithm in selecting migration sources by combining the prediction curve graphs of the WM-LSTM-TL multi-source migration prediction model generated by the WM algorithm.

[0068] like Figure 5 As shown, for OIN nodes, the load curve predicted by the multi-source transmission model obtained by the WM algorithm fluctuates less compared with the single LSTM prediction. The prediction curve is relatively smooth in the low load segment, and its prediction curve can better describe the changing trend of the short-term load curve.

[0069] The beneficial effects of this invention are as follows:

[0070] This invention can be used to accurately screen similar nodes from migration sources, train machine learning models, and propose the WM algorithm, which combines similarity analysis algorithms based on Wasserstein distance and maximum information coefficient (MIC). By integrating the similarity of nonlinearity and probability distribution, the similarity between time series variables is measured. Using the WM algorithm to determine migration sources can effectively avoid negative migration in multi-source migration prediction and improve the robustness of multi-source migration models.

[0071] System Implementation Examples

[0072] According to embodiments of the present invention, a migration source selection system for short-term load migration forecasting is provided. Figure 6 A schematic diagram illustrating the composition of a migration source selection system for short-term load migration forecasting, provided for one or more embodiments of this specification, is shown below. Figure 6 As shown, the migration source selection system for short-term load migration forecasting according to an embodiment of the present invention specifically includes:

[0073] Data processing module 60: used to preprocess the source data of the migration according to the length of the target source data to obtain a subsequence of the same length as the target source data.

[0074] Data calculation module 62: for each of the subsequences, the maximum information coefficient (MIC) between the migration source and the target source is calculated using the Wasserstein distance metric (WD) and the maximum information coefficient method.

[0075] Curve Construction Module 64: Used to construct a WD-MIC curve with WD as the x-axis and MIC as the y-axis.

[0076] The curve construction module 64 is specifically used for:

[0077] With WD as the x-axis and MIC as the y-axis, construct the WD-MIC curve by sorting x in ascending order.

[0078] Migration source selection module 66: Used to select the migration source with the highest similarity as the final migration source by calculating the area under the WD-MIC curve as the similarity between the migration source and the target source.

[0079] The embodiments of the present invention are system embodiments corresponding to the above method embodiments. The specific operation of each module can be understood by referring to the description of the method embodiments, and will not be repeated here.

[0080] Device Example 1

[0081] This invention provides an electronic device, such as... Figure 7 As shown, it includes: a memory 70, a processor 72, and a computer program stored in the memory 70 and executable on the processor 72. When the computer program is executed by the processor 72, it performs the following method steps:

[0082] S1. Preprocess the source data according to the length of the target source data to obtain a subsequence of the same length as the target source data;

[0083] S2. For each subsequence, the optimal transmission distance WD between the migration source and the target source is measured by the Wasserstein distance metric, and the maximum information coefficient MIC between the migration source and the target source is calculated using the maximum information coefficient method.

[0084] S3. Construct the WD-MIC curve with WD as the x-axis and MIC as the y-axis;

[0085] S4. The area under the WD-MIC curve is used as the similarity between the migration source and the target source. The migration source with the highest similarity is selected as the final migration source.

[0086] Device Example 2

[0087] This invention provides a computer-readable storage medium storing an information transmission implementation program. When executed by a processor 72, the program performs the following method steps:

[0088] S1. Preprocess the source data according to the length of the target source data to obtain a subsequence of the same length as the target source data;

[0089] S2. For each subsequence, the optimal transmission distance WD between the migration source and the target source is measured by the Wasserstein distance metric, and the maximum information coefficient MIC between the migration source and the target source is calculated using the maximum information coefficient method.

[0090] S3. Construct the WD-MIC curve with WD as the x-axis and MIC as the y-axis;

[0091] S4. The area under the WD-MIC curve is used as the similarity between the migration source and the target source. The migration source with the highest similarity is selected as the final migration source.

[0092] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for selecting migration sources for short-term load migration forecasting, characterized in that, include: S1. Preprocess the source data according to the length of the target source data to obtain a subsequence of the same length as the target source data; S2. For each subsequence, the optimal transmission distance WD between the migration source and the target source is measured by the Wasserstein distance metric, and the maximum information coefficient MIC between the migration source and the target source is calculated using the maximum information coefficient method. S3. Construct the WD-MIC curve with WD as the x-axis and MIC as the y-axis; S4. The area under the WD-MIC curve is used as the similarity between the migration source and the target source. The migration source with the highest similarity is selected as the final migration source. The specific method for calculating the area under the WD-MIC curve is as follows: ； in The area under the WD-MIC curve; and Representing the x and y axes, For a = coefficient, and Equal to 0, and Represents WD and MIC values, where , and It equals 1; The specific method for selecting the migration source with the highest similarity as the final migration source is as follows: Based on the area under the WD-MIC curve The similarity between the source and target subsequences is ranked, and the following sequences are selected. The largest source of migration is used as the final source of migration.

2. The method according to claim 1, characterized in that, The optimal transport distance WD between the migration source and the target source for each subsequence is specifically calculated using the Wasserstein distance metric: the minimum average distance required to move the subsequence from the migration source to the target source is calculated, as follows: ； in, For migration source subsequence samples The probability distribution, For target source sequence samples The probability distribution of X is the migration source space and Y is the target source space; express and The distance function, for right The probabilistic coupling set; For the order of WD, when At that time, WD represents the distance between the bulldozer and EMD; Optimize the calculation for WD, when At that time, WD is as follows: ； Among all possible probabilistic couplings, the sample and The expected lower bound of the distance between them is and Wasserstein-1 distance; The average EMD of the probability mass function; The similarity between the migration source and target source sequences is evaluated by calculating the normalized WD. The calculation method is as follows: ； in, The maximum EMD of the source and target source sequences and .

3. The method according to claim 2, characterized in that, The specific method for calculating the maximum information coefficient (MIC) between the migration source and the target source using the maximum information coefficient method is as follows: Based on migration source subsequence samples and target source sequence samples This yields a finite set of ordered pairs. ; According to the grid G, D will be... Values ​​are divided into Group, will D's Values ​​are divided into Grouping, calculating the mutual information (MI) between the two variables in each group; The MIC value is calculated based on the maximum MI value, as follows: ； in yes A set of ordered pairs and ; The data space is divided into XY grids, and Grouping values yes A function of the sample size, representing the upper limit of the search grid size; It is usually set to 0.6; It represents the maximum mutual information among all XY grids; a higher MIC value indicates a stronger correlation between load series.

4. The method according to claim 1, characterized in that, The specific method for constructing the WD-MIC curve with WD as the horizontal axis and MIC as the vertical axis is as follows: With WD as the x-axis and MIC as the y-axis, construct the WD-MIC curve by sorting x in ascending order.

5. A migration source selection system for short-term load migration forecasting, characterized in that, include: Data processing module: used to preprocess the source data to obtain a subsequence of the same length as the target source data, based on the length of the target source data. Data calculation module: used to calculate the maximum information coefficient (MIC) between the migration source and the target source for each subsequence using the Wasserstein distance metric (WD) and the maximum information coefficient method. Curve Construction Module: Used to construct a WD-MIC curve with WD as the x-axis and MIC as the y-axis; Migration source selection module: This module calculates the area under the WD-MIC curve as the similarity between the migration source and the target source, and selects the migration source with the highest similarity as the final migration source. The specific method for calculating the area under the WD-MIC curve is as follows: ； in The area under the WD-MIC curve; and Representing the x and y axes, For a = coefficient, and Equal to 0, and Represents WD and MIC values, where , and It equals 1; The specific method for selecting the migration source with the highest similarity as the final migration source is as follows: Based on the area under the WD-MIC curve The similarity between the source and target subsequences is ranked, and the following sequences are selected. The largest source of migration is used as the final source of migration.

6. The system according to claim 5, characterized in that, The curve construction module is specifically used for: With WD as the x-axis and MIC as the y-axis, construct the WD-MIC curve by sorting x in ascending order.

7. An electronic device, characterized in that, include: processor; as well as, A memory configured to store computer-executable instructions, which, when executed, cause the processor to implement the steps of any of the migration source selection methods for short-term load migration forecasting as described in claims 1 to 4.

8. A storage medium, characterized in that, Used to store computer-executable instructions, which, when executed, implement the steps of any of the migration source selection methods for short-term load migration forecasting as described in claims 1 to 4.

Images