Power load prediction method and device, readable storage medium and electronic equipment
By periodically decomposing and recombining historical power load data sequences and extracting features, and using a feature extraction network for multi-layer feature extraction, the problem of low accuracy in power load prediction in existing technologies is solved, and higher prediction accuracy is achieved.
Patent Information
- Application Number
- CN202410586202.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-11
- Publication Date
- 2025-11-11
AI Technical Summary
Existing power load forecasting methods fail to fully utilize the characteristics of power load data, resulting in low forecast accuracy.
By acquiring historical power load data sequences, determining the period of each sequence, performing periodic decomposition and recombination and feature extraction, and using a feature extraction network for multi-layer feature extraction and data mapping, the multi-periodic characteristics of power load are fully demonstrated.
It improves the accuracy of power load forecasting, enables more targeted forecasting, and enhances the reliability of forecast results.
Smart Images
Figure CN120933890A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of energy management technology, and in particular relates to a method, apparatus, computer-readable storage medium and electronic device for predicting power load. Background Technology
[0002] To maximize the benefits of energy management for emerging residential photovoltaic-storage systems, accurate forecasting of their power load is necessary.
[0003] In existing technologies, some general time-series forecasting methods are mainly used for power load forecasting, such as statistical learning model forecasting methods, tree model forecasting methods, and neural network forecasting methods. Although these forecasting methods can complete general time-series forecasting tasks, they cannot make full use of the data characteristics of power load to make more targeted forecasts, resulting in low accuracy of power load forecasting. Summary of the Invention
[0004] In view of this, embodiments of this application provide a power load forecasting method, apparatus, computer-readable storage medium, and electronic device to solve the problem of low accuracy in existing power load forecasting methods.
[0005] A first aspect of this application provides a power load forecasting method, which may include:
[0006] Obtain historical power load data sequences;
[0007] Determine the sequence period of each sequence in the historical power load data sequence;
[0008] The historical power load data sequence is periodically decomposed and recombined according to each sequence period to obtain each group of recombined data;
[0009] Power load forecasting is performed based on the recombined data from each group, resulting in a power load forecasting data sequence.
[0010] By using the above method, the multi-periodic characteristics of power load can be fully utilized. Based on the determined sequence period, the historical power load data sequence can be periodically decomposed and recombined, so that the multi-periodic characteristics of power load can be fully displayed. On this basis, power load forecasting can be carried out on each group of recombined data, which can effectively improve the accuracy of power load forecasting.
[0011] In one specific implementation of the first aspect, determining each sequence period of the historical power load data sequence may include:
[0012] The historical power load data sequence is subjected to time-frequency transformation to obtain frequency domain data corresponding to the historical power load data sequence.
[0013] The k frequencies with the largest amplitudes are determined based on the frequency domain data; where k is a positive integer.
[0014] The periods corresponding to the k frequencies with the largest amplitudes are used as the respective sequence periods of the historical power load data sequence.
[0015] Using the above method, historical power load data sequences can be transformed into frequency domain data. Several frequencies with the largest amplitudes can be selected from these sequences, and the period of each sequence in the historical power load data sequence can be determined accordingly. This allows for a full exploration of the multi-periodic characteristics of power load, providing a foundation for more targeted forecasting and processing.
[0016] In one specific implementation of the first aspect, the step of periodically decomposing and recombining the historical power load data sequence according to each sequence period to obtain each set of recombined data may include:
[0017] The historical power load data sequence is decomposed according to the target sequence period to obtain each decomposed subsequence corresponding to the target sequence period; wherein, the target sequence period is any sequence period;
[0018] The decomposed subsequences are recombined in the form of a two-dimensional tensor to obtain a set of recombined data corresponding to the period of the target sequence.
[0019] Using the above method, the historical power load data sequence can be decomposed into various subsequences according to a certain sequence period, and then recombined in the form of a two-dimensional tensor. The resulting recombined data can fully demonstrate the periodicity of the power load corresponding to the sequence period, while different recombined data can fully demonstrate the multi-periodicity of the power load.
[0020] In one specific implementation of the first aspect, the step of performing power load forecasting based on each set of recombined data to obtain a power load forecasting data sequence may include:
[0021] Feature extraction was performed on each group of recombined data to obtain the corresponding feature data for each group.
[0022] Feature aggregation is performed on each group of feature data to obtain aggregated feature data;
[0023] The aggregated feature data is mapped to obtain the power load prediction data sequence.
[0024] Using the above method, feature extraction can be performed first. Since the sequence periods corresponding to each group of recombined data are different, the resulting feature data can also represent the data characteristics of different periods of power load. After aggregation, aggregated feature data that fully demonstrates the multi-periodic characteristics of power load can be obtained. Based on this, data mapping can effectively improve the accuracy of power load forecasting.
[0025] In one specific implementation of the first aspect, the step of extracting features from each group of recombined data to obtain corresponding feature data may include:
[0026] Determine the time feature embedded data corresponding to the target recombined data; wherein, the target recombined data is any set of recombined data;
[0027] The target reconstructed data is superimposed with the time feature embedded data to obtain superimposed reconstructed data;
[0028] The superimposed and recombined data are subjected to feature extraction using a preset feature extraction network to obtain a set of feature data corresponding to the target recombined data.
[0029] In one specific implementation of the first aspect, determining the time feature embedded data corresponding to the target recombined data may include:
[0030] Obtain time feature data corresponding to the historical power load data sequence;
[0031] The time feature data is mapped to obtain time feature mapped data;
[0032] The time feature mapping data is periodically decomposed and recombined according to the sequence period corresponding to the target recombined data to obtain the time feature embedded data.
[0033] By using the above methods, time-related data can be considered in addition to the existing data, thereby further improving the accuracy of power load forecasting.
[0034] In one specific implementation of the first aspect, the feature extraction network may include an L-layer feature extraction module, where L is a positive integer;
[0035] The step of using a preset feature extraction network to extract features from the superimposed and recombined data to obtain a set of feature data corresponding to the target recombined data may include:
[0036] The L-layer feature extraction module of the feature extraction network is used to extract features from the superimposed and recombined data in sequence to obtain a set of feature data corresponding to the target recombined data.
[0037] By using the above method, multiple feature extraction processes can be performed using the multi-layer feature extraction module of the feature extraction network, thereby extracting deeper feature data and further improving the accuracy of power load forecasting.
[0038] In one specific implementation of the first aspect, the step of aggregating features from each group of feature data to obtain aggregated feature data may include:
[0039] The size of each set of feature data is transformed to obtain feature data of uniform size;
[0040] The aggregated feature data is obtained by superimposing the feature data of each group of features of the same size.
[0041] The above method can unify the different sizes of each group of feature data. Under the same size, the feature data of each group can be easily superimposed, thereby obtaining aggregated feature data that can fully reflect the multi-periodic characteristics of power load.
[0042] A second aspect of the embodiments of this application provides a power load forecasting device, which may include:
[0043] The data sequence acquisition module is used to acquire historical power load data sequences.
[0044] A sequence period determination module is used to determine each sequence period of the historical power load data sequence;
[0045] The periodic decomposition and recombination module is used to periodically decompose and recombine the historical power load data sequence according to each sequence period to obtain each set of recombined data;
[0046] The power load forecasting module is used to forecast power load based on each set of recombined data, and obtain a power load forecasting data sequence.
[0047] The aforementioned device can fully utilize the multi-periodic characteristics of power load, periodically decompose and reassemble the historical power load data sequence according to the determined sequence period, so that the multi-periodic characteristics of power load can be fully displayed. On this basis, power load prediction can be carried out on each set of reassembled data, which can effectively improve the accuracy of power load prediction.
[0048] In one specific implementation of the second aspect, the sequence period determination module may be specifically used to: perform time-frequency transformation on the historical power load data sequence to obtain frequency domain data corresponding to the historical power load data sequence; determine the top k frequencies with the largest amplitudes based on the frequency domain data; where k is a positive integer; and take the periods corresponding to the top k frequencies with the largest amplitudes as the respective sequence periods of the historical power load data sequence.
[0049] The aforementioned device can transform historical power load data sequences into frequency domain data, select several frequencies with the largest amplitudes, and determine the sequence periods of each historical power load data sequence accordingly. This allows for the full exploitation of the multi-periodic characteristics of power load, providing a foundation for more targeted predictive processing.
[0050] In one specific implementation of the second aspect, the periodic decomposition and recombination module can be specifically used to: decompose the historical power load data sequence according to the target sequence period to obtain each decomposed subsequence corresponding to the target sequence period; wherein, the target sequence period is any sequence period; and recombine each decomposed subsequence in the form of a two-dimensional tensor to obtain a set of recombined data corresponding to the target sequence period.
[0051] Using the aforementioned device, the historical power load data sequence can be decomposed into various subsequences according to a certain sequence period, and then recombined in the form of a two-dimensional tensor. The resulting set of recombined data can fully demonstrate the periodicity of the power load corresponding to the sequence period, while different recombined data can fully demonstrate the multi-periodicity of the power load.
[0052] In one specific implementation of the second aspect, the power load forecasting module may include:
[0053] The feature extraction submodule is used to extract features from each group of recombined data to obtain the corresponding feature data for each group.
[0054] The feature aggregation submodule is used to aggregate features from each group of feature data to obtain aggregated feature data.
[0055] The data mapping submodule is used to perform data mapping on the aggregated feature data to obtain the power load prediction data sequence.
[0056] Using the aforementioned device, feature extraction can be performed on groups. Since the sequence periods corresponding to each group of recombined data are different, the resulting feature data can also represent the data characteristics of different periods of power load. After aggregation, aggregated feature data that fully demonstrates the multi-periodic characteristics of power load can be obtained. Based on this, data mapping can effectively improve the accuracy of power load forecasting.
[0057] In one specific implementation of the second aspect, the feature extraction submodule may include:
[0058] An embedded data determination unit is used to determine the time feature embedded data corresponding to the target recombined data; wherein, the target recombined data is any set of recombined data;
[0059] A data overlay unit is used to overlay the target reconstructed data with the time feature embedded data to obtain overlay reconstructed data;
[0060] The feature extraction unit is used to extract features from the superimposed and recombined data using a preset feature extraction network to obtain a set of feature data corresponding to the target recombined data.
[0061] In one specific implementation of the second aspect, the embedded data determining unit may be specifically used to: acquire time feature data corresponding to the historical power load data sequence; perform data mapping on the time feature data to obtain time feature mapping data; and periodically decompose and reorganize the time feature mapping data according to the sequence period corresponding to the target recombined data to obtain the time feature embedded data.
[0062] The aforementioned device allows for the addition of time-related data to the existing data, thereby further improving the accuracy of power load forecasting.
[0063] In one specific implementation of the second aspect, the feature extraction network may include an L-layer feature extraction module, where L is a positive integer; the feature extraction unit may be specifically used to: use the L-layer feature extraction module of the feature extraction network to sequentially extract features from the superimposed and recombined data to obtain a set of feature data corresponding to the target recombined data.
[0064] The aforementioned device allows for multiple feature extraction processes using a multi-layer feature extraction module of the feature extraction network, thereby extracting deeper feature data and further improving the accuracy of power load forecasting.
[0065] In one specific implementation of the second aspect, the feature aggregation submodule can be specifically used to: transform the size of each group of feature data to obtain each group of feature data of uniform size; and superimpose the each group of feature data of uniform size to obtain the aggregated feature data.
[0066] The aforementioned device can unify the different sizes of various groups of feature data, and under the same size, the feature data of various groups can be conveniently superimposed, thereby obtaining aggregated feature data that can fully reflect the multi-periodic characteristics of power load.
[0067] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described power load forecasting methods.
[0068] A fourth aspect of this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described power load forecasting methods.
[0069] A fifth aspect of this application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the steps of any of the above-described power load forecasting methods. Attached Figure Description
[0070] To more clearly illustrate the technical solutions in the embodiments of this application, 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 of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0071] Figure 1 This is a flowchart of one embodiment of a power load forecasting method in this application.
[0072] Figure 2 This is a schematic diagram illustrating the periodic decomposition and recombination of historical power load data sequences.
[0073] Figure 3 This is a schematic flowchart illustrating the power load forecasting based on the recombined data from each group.
[0074] Figure 4 A flowchart illustrating the process of extracting features from each group of recombined data to obtain the corresponding feature data for each group;
[0075] Figure 5 This is a schematic diagram of a feature extraction network in an embodiment of this application;
[0076] Figure 6 This is a schematic diagram of an end-to-end model processing method;
[0077] Figure 7 This is a structural diagram of one embodiment of a power load forecasting device according to the present application.
[0078] Figure 8 This is a schematic block diagram of an electronic device according to an embodiment of this application. Detailed Implementation
[0079] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0080] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0081] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.
[0082] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0083] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [the described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [the described condition or event] is detected," or "in response to detection of [the described condition or event]."
[0084] Furthermore, in the description of this application, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0085] To maximize the benefits of energy management for emerging residential photovoltaic-storage systems, accurate forecasting of their power load is necessary.
[0086] In existing technologies, some general time-series forecasting methods are mainly used for power load forecasting, such as statistical learning model forecasting methods, tree model forecasting methods, and neural network forecasting methods. Although these forecasting methods can complete general time-series forecasting tasks, they cannot make full use of the data characteristics of power load to make more targeted forecasts, resulting in low accuracy of power load forecasting.
[0087] In view of this, embodiments of this application provide a power load forecasting method, apparatus, computer-readable storage medium, and electronic device to solve the problem of low accuracy in existing power load forecasting methods.
[0088] In this embodiment, the multi-periodic characteristics of power load can be fully utilized. The historical power load data sequence can be periodically decomposed and recombined according to the determined sequence period, so that the multi-periodic characteristics of power load can be fully displayed. Based on this, power load prediction can be performed on each group of recombined data, which can effectively improve the accuracy of power load prediction.
[0089] The subject of this application method can be an electronic device, including but not limited to desktop computers, laptops, handheld computers, servers, and other computing devices.
[0090] Please see Figure 1 One embodiment of the power load forecasting method in this application may include:
[0091] Step S101: Obtain historical power load data sequence.
[0092] Electricity load forecasting refers to predicting unknown electricity load data for a future period based on known electricity load data from a previous period. For example, known electricity load data from January 1st to January 4th can be used to predict unknown electricity load data for January 5th. Electricity load data can refer to power, measured in kilowatts (kW), or electricity, measured in kilowatt-hours (kWh).
[0093] Electricity load data is typically a time-series sequence obtained by sampling at regular time intervals. For example, assuming the data sampling interval is 5 minutes, one electricity load data point can be sampled every 5 minutes. These electricity load data points can be arranged sequentially according to the order of sampling and presented in the form of a time-series sequence.
[0094] In this embodiment, the input to the power load forecasting process can be denoted as the historical power load data sequence, and the output of the power load forecasting process can be denoted as the power load forecasting data sequence. Taking the example of forecasting unknown power load data for January 5th based on known power load data from January 1st to January 4th, the input to the power load forecasting process is the power load data from January 1st to January 4th. If the data sampling interval is 5 minutes, then 288 power load data points can be sampled each day, totaling 1152 power load data points over 4 days. The time series sequence formed by these power load data points is the historical power load data sequence, which can be represented in the following form: [X1, X2, ..., X... n ], where n is the number of known power load data, X1 is the first known power load data, X2 is the second known power load data, ..., X n Given the known nth power load data point, and so on, in this example, n is 1152. Correspondingly, the output of the power load forecasting process is 288 power load data points for January 5th. The time series formed by these power load data points is the power load forecasting data sequence, which can be represented in the following form: [X] n+1 ,X n+2 ,…,X n+m ], where m is the number of these predicted power load data, X n+1 For the first predicted electricity load data, X n+2 For the second predicted power load data, ..., X n+m For the m-th predicted power load data, the same logic applies. In this example, m is 288.
[0095] The time period length corresponding to the historical power load data sequence (denoted as the historical time period length) and the time period length corresponding to the power load forecast data sequence (denoted as the forecast time period length) can both be flexibly adjusted according to actual conditions. For example, when the historical time period length is 7 days and the forecast time period length is 2 days, it means that the power load data for the next 2 days can be predicted based on the power load data of the previous 7 days; when the historical time period length is 4 days and the forecast time period length is 4 days, it means that the power load data for the next 4 days can be predicted based on the power load data of the previous 4 days. The data sampling time interval can also be flexibly adjusted according to actual conditions, for example, it can be adjusted to 1 minute, 2 minutes, 3 minutes, 10 minutes or other values.
[0096] In one specific implementation of this application, after obtaining the historical power load data sequence, data preprocessing can be performed, and subsequent power load forecasting can be made based on the results of the data preprocessing. Data preprocessing may include, but is not limited to, quality inspection and normalization.
[0097] Quality inspection can identify defective values in historical power load data sequences, and interpolation algorithms can be used to correct these defective values. Quality inspection can include, but is not limited to, outlier detection and missing value detection; correspondingly, defective values can include, but are not limited to, outliers and missing values; interpolation algorithms can include, but are not limited to, linear interpolation, quadratic interpolation, nearest neighbor interpolation, bilinear interpolation, and cubic spline interpolation.
[0098] Normalization can include, but is not limited to, Z-Score normalization, min-max normalization, and decimal scaling normalization. Taking Z-Score normalization as an example, it can be normalized based on the mean and standard deviation of historical power load data sequences, as shown in the following formula:
[0099]
[0100] Where X is the historical power load data series, mean(X) is the mean of the historical power load data series, and std(X) is the standard deviation of the historical power load data series. This is the normalized result of the historical power load data sequence.
[0101] Step S102: Determine the sequence period of each sequence in the historical power load data sequence.
[0102] In one specific implementation of this application, the historical power load data sequence can first undergo time-frequency transformation to obtain frequency domain data corresponding to the historical power load data sequence. Then, the top k frequencies with the largest amplitudes can be determined based on the frequency domain data, and the periods corresponding to the top k frequencies with the largest amplitudes can be used as the sequence periods of the historical power load data sequence. The time-frequency transformation can include, but is not limited to, Fourier transform and wavelet transform.
[0103] Taking the time-frequency transformation using Fast Fourier Transform (FFT) as an example, the process of determining each sequence period of the historical power load data sequence is shown in the following equation:
[0104]
[0105] Where FFT is the Fast Fourier Transform function, Amp is the amplitude calculation function, A is the amplitude corresponding to each frequency after the time-frequency transformation, arg maxTopke is the function to select the k frequencies with the largest amplitudes, k is a positive integer, and {f1,...,f...} k} represents the k frequencies with the largest amplitude, {p1,..,p k} represents the various sequence periods of the historical power load data sequence.
[0106] Using the above method, historical power load data sequences can be transformed into frequency domain data. Several frequencies with the largest amplitudes can be selected from these sequences, and the period of each sequence in the historical power load data sequence can be determined accordingly. This allows for a full exploration of the multi-periodic characteristics of power load, providing a foundation for more targeted forecasting and processing.
[0107] In another specific implementation of this application, the various sequence periods of the historical power load data sequence do not need to be calculated; instead, pre-set historical statistical values can be used directly. When setting these historical statistical values, multiple different historical power load data sequences can be collected. The periods corresponding to the k frequencies with the largest amplitudes in each sequence are selected using the method described above. These selected periods are then statistically analyzed, and the k periods that appear most frequently are determined as the historical statistical values for each sequence period. These historical statistical values can then be directly used in subsequent power load forecasting.
[0108] Step S103: Periodically decompose and reassemble the historical power load data sequence according to each sequence period to obtain each group of reassembled data.
[0109] Each sequence period can correspond to a set of recombined data. Since k sequence periods are determined in step S102, k sets of recombined data can be obtained after periodic decomposition and recombination.
[0110] Taking any given sequence period (denoted as the target sequence period) as an example, we can first decompose the historical power load data sequence according to the target sequence period to obtain the various decomposed subsequences corresponding to the target sequence period. Then, we can reassemble these decomposed subsequences in the form of a two-dimensional tensor to obtain a set of reconstructed data corresponding to the target sequence period.
[0111] like Figure 2As shown, assuming there are 1152 power load data points in the historical power load data sequence, and the target sequence period is 3, the historical power load data sequence can be decomposed into three subsequences, each containing 384 power load data points. Specifically, the first subsequence consists of the time series from the 1st to the 384th power load data point; the second subsequence consists of the time series from the 385th to the 768th power load data point; and the third subsequence consists of the time series from the 769th to the 1152nd power load data point. These three subsequences can be reconstructed using a two-dimensional tensor. For example, each of the three subsequences can be used as a row in the reconstructed data, resulting in a set of reconstructed data with a size of 3×384.
[0112] By traversing each sequence period in the manner described above, the historical power load data sequence X∈R can be obtained. n Reconstructed data in the form of a series of two-dimensional tensors
[0113] Using the above method, the historical power load data sequence can be decomposed into various subsequences according to a certain sequence period, and then recombined in the form of a two-dimensional tensor. The resulting recombined data can fully demonstrate the periodicity of the power load corresponding to the sequence period, while different recombined data can fully demonstrate the multi-periodicity of the power load.
[0114] Step S104: Perform power load forecasting based on the recombined data of each group to obtain the power load forecasting data sequence.
[0115] Figure 3 The diagram shown is a schematic flowchart for power load forecasting based on the recombined data of each group. As shown in the figure, step S104 may specifically include the following process:
[0116] Step S1041: Extract features from each group of recombined data to obtain the corresponding feature data for each group.
[0117] Each set of recombined data corresponds to a set of feature data. Since k sets of recombined data were obtained through decomposition and recombination in step S103, k sets of feature data can be obtained after feature extraction.
[0118] Taking any set of reconstructed data (denoted as target reconstructed data) as an example, in one specific implementation of this application embodiment, a preset feature extraction network can be used to extract features from the target reconstructed data to obtain a set of feature data corresponding to the target reconstructed data. The feature extraction network can adopt any existing network structure capable of feature extraction, depending on the actual situation.
[0119] In another specific implementation of this application's embodiments, time characteristic data can be considered in addition to the original target reconstructed data, thereby further improving the accuracy of power load forecasting. The specific process is as follows: Figure 4 As shown:
[0120] Step S10411: Determine the time feature embedded data corresponding to the target recombined data.
[0121] First, time feature data corresponding to the historical power load data sequence can be obtained. For example, the number of hours per day, days per week, days per month, and days per year can be calculated for the time of the historical power load data sequence, thus obtaining time feature data of size n×4.
[0122] Next, the time feature data can be mapped to obtain time feature mapped data. For example, a Multi-Layer Perceptron (MLP) can be used to map the time feature data from an n×4 size to an n×D size, thus obtaining the time feature mapped data. Here, D is a hyperparameter, and its value is a positive integer.
[0123] Then, the time feature mapping data can be periodically decomposed and recombined according to the sequence period corresponding to the target recombined data, thereby obtaining the time feature embedding data corresponding to the target recombined data. The specific process of periodic decomposition and recombination can be found in the detailed description of step S103, and will not be repeated here. By traversing each sequence period in the above manner, the time feature mapping data can be transformed into each set of time feature embedding data corresponding to each set of recombined data.
[0124] Step S10412: Overlay the target reconstructed data with the time feature embedded data to obtain overlaid reconstructed data.
[0125] In one specific implementation of this application, the target reconstructed data can be superimposed with the corresponding time feature embedded data according to the following formula to obtain superimposed reconstructed data corresponding to the target reconstructed data:
[0126]
[0127] By iterating through each group of recombination data in the manner described above, we can obtain the superimposed recombination data corresponding to each group of recombination data.
[0128] Step S10413: Use a preset feature extraction network to extract features from the superimposed and recombined data to obtain a set of feature data corresponding to the target recombined data.
[0129] As an example, Figure 5 A schematic diagram of a feature extraction network is shown in the figure. The feature extraction network can include, but is not limited to, L-layer feature extraction modules, where L is a positive integer. Layers can be directly connected or use residual connections. Each feature extraction module can include, but is not limited to, depth-wise convolution (DWConv) and point-wise convolution (PWConv). Depth-wise convolution can extract intra-period features (short-term features) and cross-period features (long-term features), while point-wise convolution can fuse temporal features. During feature extraction, the L-layer feature extraction modules of the feature extraction network can be used sequentially to extract features from the superimposed and reconstructed data, thereby obtaining a set of feature data corresponding to the target reconstructed data.
[0130] according to Figure 4 By iterating through each group of recombined data in the manner shown, we can obtain the feature data corresponding to each group of recombined data.
[0131] By using the above method, multiple feature extraction processes can be performed using the multi-layer feature extraction module of the feature extraction network, thereby extracting deeper feature data and further improving the accuracy of power load forecasting.
[0132] Step S1042: Perform feature aggregation on each group of feature data to obtain aggregated feature data.
[0133] In one specific implementation of this application, the size of each set of feature data can be transformed to obtain each set of feature data of the same size, and then the set of feature data of the same size can be superimposed to obtain aggregated feature data.
[0134] As an example, each set of feature data can be represented by p. i ×f i The size of ×D is transformed into the size of n×D, thus obtaining the characteristic data F′ of each set with uniform size. i ∈R n×D(i = 1, ..., k), and then the feature data of each group of uniform size can be superimposed according to the following formula: Thus, aggregated feature data F∈R is obtained. n×D .
[0135] The above method can unify the different sizes of each group of feature data. Under the same size, the feature data of each group can be easily superimposed, thereby obtaining aggregated feature data that can fully reflect the multi-periodic characteristics of power load.
[0136] Step S1043: Perform data mapping on the aggregated feature data to obtain the power load prediction data sequence.
[0137] In one specific implementation of this application, a multilayer perceptron (MLP) can be used to map the aggregated feature data, transforming it from an n×D size to a sequence of length m, thereby obtaining the power load forecast data sequence Y∈R. m .
[0138] Using the above method, feature extraction can be performed first. Since the sequence periods corresponding to each group of recombined data are different, the resulting feature data can also represent the data characteristics of different periods of power load. After aggregation, aggregated feature data that fully demonstrates the multi-periodic characteristics of power load can be obtained. Based on this, data mapping can effectively improve the accuracy of power load forecasting.
[0139] In one specific implementation of the embodiments of this application, the following can be adopted: Figure 6 The end-to-end model processing method shown above is used to achieve the above-mentioned power load forecasting. That is, the above-mentioned decomposition and recombination, feature extraction, feature aggregation and data mapping are achieved through a preset power load forecasting model. The input of the power load forecasting model is the historical power load data sequence, and the output is the power load forecasting data sequence.
[0140] Before using the power load forecasting model, it can be trained using a pre-set training dataset.
[0141] The training dataset can be constructed from electricity load data over a historical period, such as a quarterly, semi-annual, or annual period. Each training dataset includes a length m of electricity load data as input to the model and a length n of electricity load data as the actual result. The specific training method can be flexibly configured according to actual conditions. For example, the Adam optimizer can be used with a learning rate of 0.0001, a batch size of 64, and 100 training cycles, employing an appropriate early shutdown strategy. In each training cycle, the training dataset for the current cycle is first obtained and input into the electricity load prediction model to obtain the corresponding prediction results. Then, based on the prediction and actual results of the current training cycle, the loss function for that cycle can be determined. Finally, the parameters of the electricity load prediction model can be adjusted according to the loss function.
[0142] In the embodiments of this application, the loss function used may include, but is not limited to, the Mean Squared Error (MSE) loss function.
[0143] The mean squared error loss function can be calculated using the following formula:
[0144]
[0145] Among them, X n+1:n+m For actual results, For the prediction result, Loss is the mean squared error loss function.
[0146] In the embodiments of this application, the evaluation metrics used to measure the prediction results of the power load forecasting model may include, but are not limited to, mean absolute error (MAE) and root mean squared error (RMSE).
[0147] The mean absolute error can be calculated using the following formula:
[0148]
[0149] The mean square error can be calculated using the following formula:
[0150]
[0151] Among them, X n+i For actual results, For the prediction results, MAE is the mean absolute error and RMSE is the mean squared error.
[0152] After training is complete, the power load forecasting model can be used for end-to-end forecasting. By inputting the historical power load data sequence into the power load forecasting model, the power load forecasting data sequence output by the power load forecasting model can be obtained.
[0153] In summary, in the embodiments of this application, the multi-periodic characteristics of power load can be fully utilized. The historical data sequence of power load can be periodically decomposed and recombined according to the determined sequence period, so that the multi-periodic characteristics of power load can be fully displayed. On this basis, power load prediction can be performed on each group of recombined data, which can effectively improve the accuracy of power load prediction.
[0154] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0155] Corresponding to the power load forecasting method described in the above embodiments, Figure 7 This illustration shows a structural diagram of an embodiment of a power load forecasting device provided in this application.
[0156] In this embodiment, a power load forecasting device may include:
[0157] The data sequence acquisition module 701 is used to acquire historical power load data sequences.
[0158] The sequence period determination module 702 is used to determine each sequence period of the historical power load data sequence;
[0159] The periodic decomposition and recombination module 703 is used to periodically decompose and recombine the historical power load data sequence according to each sequence period to obtain each group of recombined data.
[0160] The power load forecasting module 704 is used to forecast power load based on each set of recombined data to obtain a power load forecasting data sequence.
[0161] In one specific implementation of this application, the sequence period determination module can be specifically used to: perform time-frequency transformation on the historical power load data sequence to obtain frequency domain data corresponding to the historical power load data sequence; determine the top k frequencies with the largest amplitudes based on the frequency domain data; where k is a positive integer; and take the periods corresponding to the top k frequencies with the largest amplitudes as the respective sequence periods of the historical power load data sequence.
[0162] In one specific implementation of this application, the periodic decomposition and recombination module can be specifically used to: decompose the historical power load data sequence according to the target sequence period to obtain each decomposed subsequence corresponding to the target sequence period; wherein, the target sequence period is any sequence period; and recombine each decomposed subsequence in the form of a two-dimensional tensor to obtain a set of recombined data corresponding to the target sequence period.
[0163] In one specific implementation of this application embodiment, the power load forecasting module may include:
[0164] The feature extraction submodule is used to extract features from each group of recombined data to obtain the corresponding feature data for each group.
[0165] The feature aggregation submodule is used to aggregate features from each group of feature data to obtain aggregated feature data.
[0166] The data mapping submodule is used to perform data mapping on the aggregated feature data to obtain the power load prediction data sequence.
[0167] In one specific implementation of this application embodiment, the feature extraction submodule may include:
[0168] An embedded data determination unit is used to determine the time feature embedded data corresponding to the target recombined data; wherein, the target recombined data is any set of recombined data;
[0169] A data overlay unit is used to overlay the target reconstructed data with the time feature embedded data to obtain overlay reconstructed data;
[0170] The feature extraction unit is used to extract features from the superimposed and recombined data using a preset feature extraction network to obtain a set of feature data corresponding to the target recombined data.
[0171] In one specific implementation of this application, the embedded data determining unit may be specifically used to: acquire time feature data corresponding to the historical power load data sequence; perform data mapping on the time feature data to obtain time feature mapping data; and periodically decompose and reorganize the time feature mapping data according to the sequence period corresponding to the target recombined data to obtain the time feature embedded data.
[0172] In one specific implementation of this application embodiment, the feature extraction network may include an L-layer feature extraction module, where L is a positive integer; the feature extraction unit may be specifically used to: use the L-layer feature extraction module of the feature extraction network to sequentially extract features from the superimposed and recombined data to obtain a set of feature data corresponding to the target recombined data.
[0173] In one specific implementation of this application, the feature aggregation submodule can be specifically used to: transform the size of each group of feature data to obtain each group of feature data of uniform size; and superimpose the each group of feature data of uniform size to obtain the aggregated feature data.
[0174] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0175] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0176] Figure 8 A schematic block diagram of an electronic device provided in an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown.
[0177] like Figure 8 As shown, the electronic device 8 in this embodiment includes: a processor 80, a memory 81, and a computer program 82 stored in the memory 81 and executable on the processor 80. When the processor 80 executes the computer program 82, it implements the steps in the various power load forecasting method embodiments described above, for example... Figure 1 Steps S101 to S104 are shown. Alternatively, when the processor 80 executes the computer program 82, it implements the functions of each module / unit in the above-described device embodiments, for example... Figure 4 The functions of the data sequence acquisition module 701 to the power load prediction module 704 are shown.
[0178] For example, the computer program 82 may be divided into one or more modules / units, which are stored in the memory 81 and executed by the processor 80 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program 82 in the electronic device 8.
[0179] The electronic device 8 can be a desktop computer, laptop, handheld computer, server, or other computing device. Those skilled in the art will understand that... Figure 8 This is merely an example of electronic device 8 and does not constitute a limitation on electronic device 8. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 8 may also include input / output devices, network access devices, buses, etc.
[0180] The processor 80 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.
[0181] The memory 81 can be an internal storage unit of the electronic device 8, such as a hard disk or memory. The memory 81 can also be an external storage device of the electronic device 8, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 81 can include both internal and external storage units of the electronic device 8. The memory 81 is used to store the computer program and other programs and data required by the electronic device 8. The memory 81 can also be used to temporarily store data that has been output or will be output.
[0182] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0183] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0184] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0185] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.
[0186] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0187] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0188] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.
[0189] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A method for predicting electricity load, characterized in that, include: Obtain historical power load data sequences; Determine the sequence period of each sequence in the historical power load data sequence; The historical power load data sequence is periodically decomposed and recombined according to each sequence period to obtain each group of recombined data; Power load forecasting is performed based on the recombined data from each group, resulting in a power load forecasting data sequence.
2. The power load forecasting method according to claim 1, characterized in that, Determining each sequence period of the historical power load data sequence includes: The historical power load data sequence is subjected to time-frequency transformation to obtain frequency domain data corresponding to the historical power load data sequence. The k frequencies with the largest amplitudes are determined based on the frequency domain data; where k is a positive integer. The periods corresponding to the k frequencies with the largest amplitudes are used as the respective sequence periods of the historical power load data sequence.
3. The power load forecasting method according to any one of claims 1 to 2, characterized in that, The historical power load data sequence is periodically decomposed and recombined according to each sequence period to obtain each set of recombined data, including: The historical power load data sequence is decomposed according to the target sequence period to obtain each decomposed subsequence corresponding to the target sequence period; wherein, the target sequence period is any sequence period; The decomposed subsequences are recombined in the form of a two-dimensional tensor to obtain a set of recombined data corresponding to the period of the target sequence.
4. The power load forecasting method according to any one of claims 1 to 3, characterized in that, The process of forecasting power load based on each set of recombined data to obtain a power load forecasting data sequence includes: Feature extraction was performed on each group of recombined data to obtain the corresponding feature data for each group. Feature aggregation is performed on each group of feature data to obtain aggregated feature data; The aggregated feature data is mapped to obtain the power load prediction data sequence.
5. The power load forecasting method according to claim 4, characterized in that, The step of extracting features from each group of recombined data to obtain corresponding feature data for each group includes: Determine the time feature embedded data corresponding to the target recombined data; wherein, the target recombined data is any set of recombined data; The target reconstructed data is superimposed with the time feature embedded data to obtain superimposed reconstructed data; The superimposed and recombined data are subjected to feature extraction using a preset feature extraction network to obtain a set of feature data corresponding to the target recombined data.
6. The power load forecasting method according to claim 5, characterized in that, The determination of the time feature embedded data corresponding to the target recombined data includes: Obtain time feature data corresponding to the historical power load data sequence; The time feature data is mapped to obtain time feature mapped data; The time feature mapping data is periodically decomposed and recombined according to the sequence period corresponding to the target recombined data to obtain the time feature embedded data.
7. The power load forecasting method according to claim 5, characterized in that, The feature extraction network includes L layers of feature extraction modules, where L is a positive integer; The step of using a preset feature extraction network to extract features from the superimposed and recombined data to obtain a set of feature data corresponding to the target recombined data includes: The L-layer feature extraction module of the feature extraction network is used to extract features from the superimposed and recombined data in sequence to obtain a set of feature data corresponding to the target recombined data.
8. The power load forecasting method according to any one of claims 4 to 7, characterized in that, The process of aggregating features from each group of feature data to obtain aggregated feature data includes: The size of each set of feature data is transformed to obtain feature data of uniform size; The aggregated feature data is obtained by superimposing the feature data of each group of features of the same size.
9. A power load forecasting device, characterized in that, include: The data sequence acquisition module is used to acquire historical power load data sequences. A sequence period determination module is used to determine each sequence period of the historical power load data sequence; The periodic decomposition and recombination module is used to periodically decompose and recombine the historical power load data sequence according to each sequence period to obtain each set of recombined data; The power load forecasting module is used to forecast power load based on each set of recombined data, and obtain a power load forecasting data sequence.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the power load forecasting method as described in any one of claims 1 to 8.
11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the power load forecasting method as described in any one of claims 1 to 8.