Short-term power load prediction method and device, equipment and storage medium

Abstract

The invention provides a short-term power load prediction method and device, equipment and a storage medium, and the method comprises the steps: obtaining an original signal of historical power, and decomposing the original signal into a high-frequency component and a low-frequency component; inputting the high-frequency component into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence, inputting the low-frequency component into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence, and superposing and reconstructing the high-frequency prediction sequence and the low-frequency prediction sequence to obtain a preliminary prediction sequence; key factors are obtained through correlation analysis and screening, and feature values of the key factors and the preliminary prediction sequence are spliced into a fusion feature vector; and inputting the fusion feature vector into a full connection layer to obtain a short-term power load prediction result. According to the method, the key factors influencing the power load are obtained through screening and spliced with the preliminary prediction sequence to serve as the input features of the full connection layer, so that unnecessary increase of the model scale is reduced, and the interpretability of the model is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and in particular to a short-term power load forecasting method, device, equipment and storage medium. Background Art

[0002] Electricity provides the basic guarantee for social production and people's lives, so there are high requirements for the safe and stable operation of the power system. Since electric energy storage is difficult and costly, the smooth operation of the power system depends on the production scheduling and planning of the power grid and various power plants, so accurate short-term power load forecasting is needed as a basis.

[0003] In the related technologies, since the power load itself has strong volatility and has characteristics such as nonlinearity and non-stationarity, it is difficult to fully explore its trend by directly predicting the original load data. Empirical mode decomposition (EMD) and its variants, such as ensemble empirical mode decomposition (EEMD), have received widespread attention. However, in short-term load forecasting, in addition to its own load trend, the load is also affected by many related influencing factors. However, the influencing factors are numerous and complex. Directly inputting the influencing factors into the neural network model will cause an unnecessary increase in the model size, resulting in poor interpretability of the model.

[0004] Based on the above analysis of the development status of this technical field, the existing technology lacks a solution for independently analyzing the correlation of influencing factors and then combining it with the prediction results obtained by the neural network model. Summary of the invention

[0005] The purpose of the present invention is to provide a short-term power load forecasting method, device, equipment and storage medium, aiming to solve the above-mentioned problems in the prior art.

[0006] According to a first aspect of an embodiment of the present invention, there is provided a short-term power load forecasting method, comprising:

[0007] Obtaining the original signal of historical power, and decomposing the original signal into high-frequency components and low-frequency components;

[0008] The high-frequency component is input into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence, the low-frequency component is input into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence, and the high-frequency prediction sequence and the low-frequency prediction sequence are superimposed and reconstructed to obtain a preliminary prediction sequence;

[0009] The key factors affecting power load are screened through correlation analysis, and the characteristic values ​​of the key factors and the preliminary prediction sequence are spliced ​​into a fusion feature vector;

[0010] The fused feature vector is input into the fully connected layer to obtain the short-term power load forecast result.

[0011] According to a second aspect of an embodiment of the present invention, there is provided a short-term power load forecasting device, comprising:

[0012] An original signal decomposition module is used to obtain the original signal of historical power and decompose the original signal into a high-frequency component and a low-frequency component;

[0013] An independent prediction module is used to input the high-frequency component into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence, input the low-frequency component into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence, and superimpose and reconstruct the high-frequency prediction sequence and the low-frequency prediction sequence to obtain a preliminary prediction sequence;

[0014] The key factor analysis module is used to screen the key factors affecting the power load through correlation analysis, and to concatenate the eigenvalues ​​of the key factors and the preliminary prediction sequence into a fused feature vector;

[0015] The prediction result correction module is used to input the fused feature vector into the fully connected layer to obtain the short-term power load prediction result.

[0016] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the short-term power load forecasting method provided in the first aspect of the present disclosure are implemented.

[0017] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which a program for implementing information transmission is stored, and when the program is executed by a processor, the steps of the short-term power load forecasting method provided in the first aspect of the present disclosure are implemented.

[0018] The technical solution provided by the embodiment of the present invention includes the following beneficial effects: based on the effective decomposition of the original signal into high-frequency components and low-frequency components, the corresponding neural network model is used to predict the prediction sequence of different components, and the key factors affecting the power load are screened through correlation analysis, which are directly spliced ​​with the preliminary prediction sequence as the input feature of the fully connected layer, and the key factor processing module is not added to the neural network model to reduce the unnecessary increase in the model scale, thereby enhancing the interpretability of the model.

[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0021] Figure 1 is a flow chart of a short-term power load forecasting method according to an embodiment of the present invention;

[0022] Figure 2 is a schematic diagram of CEEMDAN signal processing according to an embodiment of the present invention;

[0023] Figure 3 is a schematic diagram of a design framework of an embodiment of the present invention;

[0024] Figure 4 is a schematic diagram of a short-term power load forecasting device according to an embodiment of the present invention;

[0025] Figure 5 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0027] Method Embodiment

[0028] According to an embodiment of the present invention, a short-term power load forecasting method is provided. Figure 1 is a flow chart of a short-term power load forecasting method according to an embodiment of the present invention. Figure 1 As shown, the short-term power load forecasting method according to an embodiment of the present invention specifically includes:

[0029] In step S110, the original signal of historical power is obtained, and the original signal is decomposed into a high-frequency component and a low-frequency component, specifically including:

[0030] The power load is abstracted into a signal that varies over time, usually expressed as a time series;

[0031] The signal processing technology CEEMDAN is used to decompose the original signal into high-frequency components and low-frequency components, including:

[0032] Determine the average number I, add white noise with different amplitudes ε to the original signal, obtain each new signal sequence, and obtain the new signal sequence based on formula 1:

[0033] x i (t) = x(t) + εn i (t) Formula 1;

[0034] Among them, x i (t)(i=1,2,3,...,I) represents the new signal sequence after adding Gaussian white noise, that is, the sequence to be decomposed, x(t) represents the original signal, n i (t) represents the Gaussian white noise processed for the i-th time.

[0035] For each new signal sequence x i (t) Perform EMD modal decomposition to obtain the IMF signal corresponding to each new signal sequence, and calculate the average value of all IMF signals based on formula 2:

[0036]

[0037] in, represents the IMF signal corresponding to each new signal sequence in the first iteration, IMF1(t) represents the average value of the first iteration, the subscript represents the corresponding number of iterations, and i represents the sequence number;

[0038] Based on Formula 3, the mean value is removed from the original signal to obtain the residual signal, and the residual signal is used as the original signal for the next iteration:

[0039] RES1=x(t)-IMF1(t) Formula 3;

[0040] Among them, RES1 represents the residual signal after the first decomposition, and the subscript represents the corresponding number of iterations;

[0041] Repeat the iteration T times until the residual signal is a monotonic function or a constant, then the iteration stops. The average value of the EMD modal decomposition of the Tth iteration and the calculation of the residual signal are shown in Formula 4 and Formula 5. The total CEEMDAN decomposition of the original signal is shown in Formula 6, and x is the decomposition result of the original signal:

[0042]

[0043] RES T =RES T-1 -IMF T (t) Formula 5;

[0044]

[0045] Figure 2 Schematic diagram of CEEMDAN signal processing according to an embodiment of the present invention. Figure 2 As shown, the algorithm flow of CEEMDAN is demonstrated.

[0046] After the original load data is decomposed by CEEMDAN, multiple IMFs will be obtained. Different components represent the information of different frequency components in the original load. The high-frequency component reflects the fluctuation degree of the data, and the low-frequency component reflects the change trend of the data. The IMF components should be reasonably divided.

[0047] The IMF signals in the last round of iterations are obtained as judgment signals. The signal zero-crossing rate is used to measure the severity of data changes to distinguish high-frequency and low-frequency components. The signal zero-crossing rate Z is calculated using formula 7:

[0048]

[0049] Among them, N represents the number of points contained in the component, n zero It indicates the number of times a component crosses the horizontal axis;

[0050] The judgment signal whose zero-crossing rate is greater than the preset threshold is taken as the high-frequency sub-component. In the embodiment of the present invention, the preset threshold is 0.01, so the judgment signal whose zero-crossing rate is greater than 0.01 is the high-frequency sub-component, and the remaining judgment signals are taken as low-frequency sub-components, that is, the judgment signals whose zero-crossing rates are between 0-0.01. All high-frequency sub-components are superimposed to obtain the high-frequency component, and all low-frequency sub-components are superimposed to obtain the low-frequency component. Both the high-frequency component and the low-frequency component are sequences.

[0051] In step S120, the high frequency component is input into a pre-established high frequency prediction model to obtain a high frequency prediction sequence, the low frequency component is input into a pre-established low frequency prediction model to obtain a low frequency prediction sequence, and the high frequency prediction sequence and the low frequency prediction sequence are superimposed and reconstructed to obtain a preliminary prediction sequence, specifically including:

[0052] Two independent load forecasting models are established in advance, namely, a high-frequency forecasting model and a low-frequency forecasting model;

[0053] The BiGRU model is used to obtain the high-frequency prediction sequence, and the multivariate linear regression model is used to obtain the low-frequency prediction sequence. This is because the multivariate linear regression model does not require tedious iteration and parameter adjustment processes, has fast execution speed and high prediction accuracy, and is suitable for stable and smooth sequences such as low-frequency components, while high-frequency components with large fluctuations are more suitable for complex models that can effectively capture fluctuations;

[0054] The BiGRU model consists of two independent GRU networks, combining a bidirectional recurrent neural network and a gated recurrent unit. Its output is determined by the states of a forward GRU and a reverse GRU. Its bidirectionality and gating mechanism can capture past and future information in time series, which helps improve sequence modeling capabilities. The output of the BiGRU model is shown in Formula 8:

[0055]

[0056] in, They represent the output of the forward hidden layer and the reverse hidden layer at time t, respectively, and x t represents the sequence of high-frequency components, u t 、v t Respectively The corresponding trainable weight coefficient, b t Represents the bias corresponding to the hidden layer state at time t;

[0057] The expression of the multivariate linear regression model is shown in Formula 9 and Formula 10:

[0058] Y=X×β+μ Formula 9;

[0059]

[0060] Among them, Y represents the power load value, X represents the value in the low-frequency component, β represents the coefficient obtained by pre-training, among which β0 represents the constant term, β i (i=1,2,...,n) represents the regression coefficient, μ and μ i represents random fluctuations, y i represents the predicted value of the i-th time node, x ij represents the value of the jth independent variable selected in the low-frequency component at the i-th time node;

[0061] The prediction process of the low-frequency prediction sequence is as follows:

[0062] Filter the preset time nodes in the low-frequency component, and obtain the value of the low-frequency component corresponding to each time node, use the value of the low-frequency component as the independent variable in the multiple linear regression model, and use the multiple linear regression model to obtain the low-frequency prediction sequence;

[0063] If the sequence of low-frequency components is abcd, and the low-frequency prediction sequence of future ef needs to be predicted, ab can be used as the independent variable for predicting e, cd can be used as the independent variable for predicting f, and so on. That is, according to the time distance scale, the corresponding independent variable is selected for each node to be predicted, and then each prediction value is continuously formed into a low-frequency prediction sequence;

[0064] The load values ​​of the corresponding time nodes of the high-frequency prediction sequence and the low-frequency prediction sequence are averaged and used as the preliminary prediction sequence.

[0065] In step S130, the key factors affecting the power load are screened through correlation analysis, and the characteristic values ​​of the key factors and the preliminary prediction sequence are spliced ​​into a fusion feature vector, which specifically includes:

[0066] Based on formula 11, the key factors were screened using the Pearson correlation analysis method:

[0067]

[0068] Among them, r represents the Pearson correlation result, N represents the total number of data, e represents the power load sequence, and m i Indicates different meteorological factors;

[0069] Preferably, due to the inconsistency of the dimensions of the power load and the influencing factor sequence, all the selected variables need to be normalized before the final prediction. In the training stage of the fully connected layer, normalization is also performed to avoid the difference in the dimensions of the data leading to slow model convergence, gradient disappearance or gradient explosion, etc. Min-Max normalization is a linear transformation of the data, mapping the data value to [0.1], as shown in Formula 12:

[0070]

[0071] Among them, z* represents the normalized value of the sample, z represents the original value of the sample data, and z max Represents the maximum value of the sample data, z min Indicates the minimum value of sample data;

[0072] Preferably, for date data, the seven days of the week are converted into numbers 1-7, and the dates are encoded using one-hot encoding, with working days represented by 1 and holidays represented by 0.

[0073] In step S140, the fused feature vector is input into the fully connected layer to obtain the short-term power load forecast result, which specifically includes:

[0074] The fully connected layer can directly generate prediction values. Since the prediction error is calculated by setting the target value to correct the model parameters during training, the fused feature vector can be directly input into the trained fully connected layer.

[0075] Since the fusion feature vector includes the eigenvalues ​​of the key factors and the preliminary prediction sequence, the preliminary prediction sequence is also input into the fully connected layer as a feature, and the fully connected layer can learn the relationship between each feature. Therefore, the process of using the fully connected layer to obtain the short-term power load forecast result is also the process of using the key factors to correct the preliminary prediction sequence to get closer to the actual prediction result;

[0076] In an embodiment of the present invention, the activation function used in the linear layer is linear, and the model is compiled using an Adam optimizer and a mean square error loss function. Preferably, a regularization technique is used to prevent overfitting.

[0077] Figure 3 Schematic diagram of the design framework of an embodiment of the present invention, such as Figure 3 As shown, the complete framework of short-term power load forecasting is presented.

[0078] To sum up, in response to the existing problems, the short-term power load forecasting method invented this time is based on the effective decomposition of the original signal into high-frequency components and low-frequency components, and the corresponding neural network model is used to predict the prediction sequence of different components. The BiGRU model can effectively capture the fluctuations in the high-frequency components, and the MLR model is more suitable for processing the prediction of stable sequences to improve the accuracy of the preliminary prediction sequence prediction; at the same time, the key factors affecting the power load are screened through correlation analysis and directly concatenated with the preliminary prediction sequence as the input features of the fully connected layer, and the short-term power load prediction results are directly obtained through the fully connected layer, so as to optimize the results of the preliminary prediction sequence with the help of key factors and obtain more accurate short-term prediction values; no key factor processing module is added to the neural network model to reduce the unnecessary increase in the model scale, thereby enhancing the interpretability of the model.

[0079] Device Embodiment

[0080] According to an embodiment of the present invention, a short-term power load forecasting device is provided. Figure 4 Schematic diagram of a short-term power load forecasting device according to an embodiment of the present invention. Figure 4 As shown, the short-term power load forecasting device according to an embodiment of the present invention specifically includes:

[0081] The original signal decomposition module 40 is used to obtain the original signal of historical power and decompose the original signal into a high-frequency component and a low-frequency component, specifically for:

[0082] The signal processing technology CEEMDAN is used to decompose the original signal into high-frequency components and low-frequency components, including:

[0083] Add white noise of different amplitudes to the original signal to obtain each new signal sequence;

[0084] Perform EMD modal decomposition on each new signal sequence to obtain the IMF signal corresponding to each new signal sequence, calculate the average value of all IMF signals, remove the average value from the original signal to obtain the residual signal, and use the residual signal as the original signal for the next round of iteration;

[0085] Each IMF signal in the last round of iteration is obtained as a judgment signal, and the judgment signal whose signal zero-crossing rate is greater than a preset threshold is used as a high-frequency sub-component, and the remaining judgment signals are used as low-frequency sub-components. All high-frequency sub-components are superimposed to obtain a high-frequency component, and all low-frequency sub-components are superimposed to obtain a low-frequency component.

[0086] The independent prediction module 42 is used to input the high-frequency component into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence, input the low-frequency component into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence, and superimpose and reconstruct the high-frequency prediction sequence and the low-frequency prediction sequence to obtain a preliminary prediction sequence, specifically for:

[0087] Use the bidirectional gated recurrent unit BiGRU model to obtain high-frequency prediction sequences;

[0088] The preset time nodes in the low-frequency components are screened, and the values ​​of the low-frequency components corresponding to each time node are obtained. The values ​​of the low-frequency components are used as independent variables in the multiple linear regression model, and the multiple linear regression model is used to obtain the low-frequency prediction sequence.

[0089] The load values ​​of the corresponding time nodes of the high-frequency prediction sequence and the low-frequency prediction sequence are averaged and used as the preliminary prediction sequence.

[0090] The key factor analysis module 44 is used to screen the key factors affecting the power load through correlation analysis, and to concatenate the characteristic values ​​of the key factors and the preliminary prediction sequence into a fusion characteristic vector, which is specifically used for:

[0091] Pearson correlation analysis was used to screen the key factors.

[0092] The prediction result correction module 46 is used to input the fused feature vector into the fully connected layer to obtain the short-term power load prediction result.

[0093] To sum up, in response to the existing problems, the short-term power load forecasting device invented this time is based on the effective decomposition of the original signal into high-frequency components and low-frequency components, and the corresponding neural network model is used to predict the prediction sequence of different components. The BiGRU model can effectively capture the fluctuations in the high-frequency components, and the MLR model is more suitable for processing the prediction of stable sequences to improve the accuracy of the preliminary prediction sequence prediction; at the same time, the key factors affecting the power load are screened through correlation analysis and directly concatenated with the preliminary prediction sequence as the input features of the fully connected layer, and the prediction results of the short-term power load are directly obtained through the fully connected layer, so as to optimize the results of the preliminary prediction sequence with the help of key factors and obtain more accurate short-term prediction values; no key factor processing module is added to the neural network model to reduce the unnecessary increase in the model scale, thereby enhancing the interpretability of the model.

[0094] Electronic device embodiment

[0095] Figure 5 Schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device 500 may include at least one processor 510 and a memory 520. The processor 510 may execute instructions stored in the memory 520. The processor 510 is connected to the memory 520 through a data bus. In addition to the memory 520, the processor 510 may also be connected to an input device 530, an output device 540, and a communication device 550 through a data bus.

[0096] The processor 510 may be any conventional processor, such as a commercially available CPU. The processor may also include a graphics processor (Graphic Process Unit, GPU), a field programmable gate array (Field Programmable Gate Array, FPGA), a system on chip (System on Chip, SOC), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC) or a combination thereof.

[0097] The memory 520 may be implemented by any type of volatile or nonvolatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0098] In the embodiment of the present disclosure, executable instructions are stored in the memory 520, and the processor 510 can read the executable instructions from the memory 520 and execute the instructions to implement all or part of the steps of any short-term power load forecasting method in the above exemplary embodiments.

[0099] Computer Readable Storage Medium Embodiments

[0100] In addition to the above-mentioned methods and devices, an exemplary embodiment of the present disclosure may also be a computer program product or a computer-readable storage medium storing the computer program product, wherein the computer product includes computer program instructions that can be executed by a processor to implement all or part of the steps described in any of the short-term power load forecasting methods in the above-mentioned exemplary embodiments.

[0101] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, etc., and also conventional procedural programming languages ​​such as "C" language or similar programming languages ​​and scripting languages ​​(e.g., Python). The program code may be executed entirely on the user computing device, partially on the user computing device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0102] Computer readable storage media can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of readable storage media include: a static random access memory (SRAM) with one or more wires electrically connected, an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk, or any suitable combination of the above.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements 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 short-term power load forecasting method, characterized in that: include: Acquire an original signal of historical power, and decompose the original signal into a high-frequency component and a low-frequency component; Inputting the high-frequency component into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence, inputting the low-frequency component into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence, and superimposing and reconstructing the high-frequency prediction sequence and the low-frequency prediction sequence to obtain a preliminary prediction sequence; The key factors affecting the power load are screened by correlation analysis, and the characteristic values ​​of the key factors and the preliminary prediction sequence are spliced ​​into a fusion characteristic vector; The fused feature vector is input into the fully connected layer to obtain the short-term power load forecast result.

2. The method according to claim 1, characterized in that Decomposing the original signal into high-frequency components and low-frequency components specifically includes: using the signal processing technology CEEMDAN to decompose the original signal into high-frequency components and low-frequency components.

3. The method according to claim 2, characterized in that The use of the signal processing technology CEEMDAN to decompose the original signal into high-frequency components and low-frequency components specifically includes: Adding white noise of different amplitudes to the original signal to obtain new signal sequences; Perform EMD modal decomposition on each new signal sequence to obtain an IMF signal corresponding to each new signal sequence, calculate the average value of all the IMF signals, remove the average value from the original signal to obtain a residual signal, and use the residual signal as the original signal for the next round of iteration; Each IMF signal in the last round of iteration is obtained as a judgment signal, and the judgment signal whose signal zero-crossing rate is greater than a preset threshold is used as a high-frequency sub-component, and the remaining judgment signals are used as low-frequency sub-components. All high-frequency sub-components are superimposed to obtain a high-frequency component, and all low-frequency sub-components are superimposed to obtain a low-frequency component.

4. The method according to claim 1, characterized in that: The step of inputting the high-frequency component into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence specifically includes: using a bidirectional gated recurrent unit BiGRU model to obtain a high-frequency prediction sequence.

5. The method according to claim 1, characterized in that The step of inputting the low-frequency component into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence specifically includes: A preset number of time nodes in the low-frequency component are screened, and the value of the low-frequency component corresponding to each time node is obtained, the value of the low-frequency component is used as an independent variable in a multiple linear regression model, and the multiple linear regression model is used to obtain a low-frequency prediction sequence.

6. The method according to claim 1, characterized in that The superimposing and reconstructing the high-frequency prediction sequence and the low-frequency prediction sequence to obtain a preliminary prediction sequence includes: calculating an average of the load values ​​of the high-frequency prediction sequence and the low-frequency prediction sequence at corresponding time nodes to obtain the preliminary prediction sequence.

7. The method according to claim 1, characterized in that The method of obtaining the key factors affecting the power load by screening through correlation analysis specifically includes: using the Pearson correlation analysis method to screen the key factors.

8. A short-term power load forecasting device, characterized in that: include: An original signal decomposition module, used for obtaining the original signal of historical power and decomposing the original signal into a high-frequency component and a low-frequency component; An independent prediction module, used for inputting the high-frequency component into a pre-established high-frequency prediction model to obtain a high-frequency prediction sequence, inputting the low-frequency component into a pre-established low-frequency prediction model to obtain a low-frequency prediction sequence, and superimposing and reconstructing the high-frequency prediction sequence and the low-frequency prediction sequence to obtain a preliminary prediction sequence; A key factor analysis module, used for screening key factors affecting power load through correlation analysis, and splicing the characteristic values ​​of the key factors and the preliminary prediction sequence into a fusion characteristic vector; The prediction result correction module is used to input the fused feature vector into the fully connected layer to obtain a short-term power load prediction result.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the short-term power load forecasting method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an implementation program for information transmission, and when the program is executed by a processor, the steps of the short-term power load forecasting method according to any one of claims 1 to 7 are implemented.

Images