Load prediction method and system based on data decomposition

Abstract

The invention relates to a load prediction method and system based on data decomposition, and the method comprises the following steps: obtaining historical load data of a target object, carrying out the selection of similar days, and obtaining the historical load data of the similar days similar to a target date; a variational mode decomposition method is utilized to decompose multiple components from similar day historical load data, and a dung beetle algorithm is utilized to optimize the variational mode decomposition method; based on each decomposed component, constructing a parallel CNN-LSTM prediction model respectively, and performing training; and performing load prediction of the target object on the target date through the trained CNN-LSTM prediction model.

Description

Technical Field

[0001] The invention relates to a load forecasting method and system based on data decomposition, belonging to the technical field of load forecasting. Background Art

[0002] In the power industry, load forecasting is crucial to power generation scheduling and energy trading, and can affect the stable operation of the power system. In recent years, hybrid forecasting models have been widely used in the field of load forecasting. However, due to the sharp increase in the proportion of uncertain loads, data decomposition is needed to improve the forecast accuracy. Summary of the invention

[0003] In order to solve the above problems existing in the prior art, the present invention proposes a load forecasting method and system based on data decomposition.

[0004] The technical solution of the present invention is as follows:

[0005] In one aspect, the present invention provides a load forecasting method based on data decomposition, comprising the following steps:

[0006] Obtain historical load data of the target object, and select similar days to obtain historical load data of similar days similar to the target date;

[0007] The variational mode decomposition method is used to decompose multiple components from the historical load data of similar days, and the variational mode decomposition method is optimized using the dung beetle algorithm;

[0008] Based on the decomposed components, parallel CNN-LSTM prediction models are constructed and trained respectively;

[0009] The trained CNN-LSTM prediction model is used to predict the load of the target object on the target date.

[0010] As a preferred implementation, the method for selecting similar days is specifically as follows:

[0011] The weighted n-dimensional Euclidean distance is used as the standard to measure the similarity of load data on two different dates. The specific formula is as follows;

[0012]

[0013] ΔP i =P f -P i ,;

[0014] Where WED is the weighted n-dimensional Euclidean distance, w S represents the weight of the sth load data selected in the stepwise regression, S represents the total number of selected load data, ΔP iIndicates the target date load data P f With historical date load data P i The difference of values;

[0015] According to the target date, a date range of similar days is selected, and within the date range of similar days, similar days are selected according to the similarity of load data.

[0016] As a preferred implementation, in the step of optimizing the variational mode decomposition method using the dung beetle algorithm, the fitness function of the dung beetle algorithm is set to the minimum envelope entropy, specifically:

[0017]

[0018] Among them, p j represents the probability of an event occurring; m is the number of samples; a(j) is the Hilbert demodulated envelope sequence of the jth modal component decomposed by variational mode.

[0019] As a preferred implementation, the CNN-LSTM prediction model includes a CNN layer based on a CNN convolutional neural network, and an LSTM layer based on an LSTM long short-term memory neural network, and an attention layer; the steps of constructing parallel CNN-LSTM prediction models based on the decomposed components and training them include:

[0020] The decomposed components are converted into multivariate arrays and input into the CNN layer and LSTM layer for training respectively;

[0021] The outputs of the CNN layer and the LSTM layer are connected and input into the attention layer. The attention score is calculated through the attention layer and input into the fully connected layer. The load forecast value is output through normalization and the fully connected layer.

[0022] According to the loss between the load prediction value and the true value, the CNN-LSTM prediction model is trained through back propagation.

[0023] On the other hand, the present invention also provides a load forecasting system based on data decomposition, comprising:

[0024] The data acquisition module is used to acquire the historical load data of the target object and select similar days to acquire the historical load data of similar days similar to the target date;

[0025] The data decomposition module uses the variational mode decomposition method to decompose multiple components from the historical load data of similar days, and uses the dung beetle algorithm to optimize the variational mode decomposition method;

[0026] The model training module builds parallel CNN-LSTM prediction models based on the decomposed components and performs training.

[0027] The load forecasting module is used to forecast the load of the target object on the target date through the trained CNN-LSTM forecasting model.

[0028] As a preferred implementation, the method for selecting similar days is specifically as follows:

[0029] The weighted n-dimensional Euclidean distance is used as the standard to measure the similarity of load data on two different dates. The specific formula is as follows;

[0030]

[0031] ΔP i =P f -P i ,;

[0032] Where WED is the weighted n-dimensional Euclidean distance, w S represents the weight of the sth load data selected in the stepwise regression, S represents the total number of selected load data, ΔP i Indicates the target date load data P f With historical date load data P i The difference of values;

[0033] According to the target date, a date range of similar days is selected, and within the date range of similar days, similar days are selected according to the similarity of load data.

[0034] As a preferred implementation, in the step of optimizing the variational mode decomposition method using the dung beetle algorithm, the fitness function of the dung beetle algorithm is set to the minimum envelope entropy, specifically:

[0035]

[0036] Among them, p j represents the probability of an event occurring; m is the number of samples; a(j) is the Hilbert demodulated envelope sequence of the jth modal component decomposed by variational mode.

[0037] As a preferred implementation, the CNN-LSTM prediction model includes a CNN layer based on a CNN convolutional neural network, and an LSTM layer based on an LSTM long short-term memory neural network, and an attention layer; the steps of constructing parallel CNN-LSTM prediction models based on the decomposed components and training them include:

[0038] The decomposed components are converted into multivariate arrays and input into the CNN layer and LSTM layer for training respectively;

[0039] The outputs of the CNN layer and the LSTM layer are connected and input into the attention layer. The attention score is calculated through the attention layer and input into the fully connected layer. The load forecast value is output through normalization and the fully connected layer.

[0040] According to the loss between the load prediction value and the true value, the CNN-LSTM prediction model is trained through back propagation.

[0041] On the other hand, the present invention further proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the load forecasting method based on data decomposition as described in any embodiment of the present invention is implemented.

[0042] On the other hand, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the load forecasting method based on data decomposition as described in any embodiment of the present invention.

[0043] Additional aspects and advantages of the present invention will be set forth in the following description, and some of them will be apparent from the description, or may be understood by practicing the present invention. In addition, the various aspects and advantages of the present invention may be realized and obtained by the method steps and combinations particularly pointed out in the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of a method flow chart of Embodiment 1 of the present invention;

[0045] Figure 2 Schematic diagram of optimizing VMD decomposition by using DBO algorithm in an embodiment of the present invention;

[0046] Figure 3 Schematic diagram of the working principle of the CNN-LSTM model in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0049] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0050] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0051] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0052] Embodiment 1:

[0053] As the proportion of uncertain loads increases dramatically, data decomposition is needed to improve prediction accuracy. However, when the original data contains more complex nonlinear and nonstationary components, the appropriate decomposition level of variational mode decomposition (VMD) is more difficult to determine. If the decomposition level is set too low, multiple components of the time series data will exist in one mode at the same time. On the other hand, if the decomposition level is set too high, one component of the time series data will exist in multiple modes. In the serial CNN-LSTM model architecture, the CNN output is used as the LSTM input, and the prediction results are accurate; however, the features extracted from the CNN layer affect the training output of the LSTM, and its ability to extract temporal features is not effectively utilized.

[0054] To address the above problems, this embodiment proposes a parallel CNN-LSTM load forecasting method based on data decomposition, which can effectively improve the accuracy and reliability of load forecasting. Figure 1 , the method comprises the following steps:

[0055] S100, obtaining historical load data of the target object, and selecting similar days to obtain historical load data of similar days similar to the target date;

[0056] S200, using the variational mode decomposition (VMD) method to decompose multiple components from the historical load data of similar days, and using the dung beetle optimizer (DBO) algorithm to optimize the variational mode decomposition method;

[0057] S300, based on the decomposed components, respectively construct parallel CNN-LSTM prediction models and perform training;

[0058] S400, using the trained CNN-LSTM prediction model to predict the load of the target object on the target date.

[0059] As a preferred implementation of this embodiment, in step S100, the method for selecting similar days is specifically as follows:

[0060] S101. The weighted n-dimensional Euclidean distance is used as a standard for measuring the similarity of load data on two different dates. The specific formula is as follows:

[0061]

[0062] ΔP i =P f -P i , (1)

[0063] Where WED is the weighted n-dimensional Euclidean distance, w S represents the weight of the sth load data selected in the stepwise regression, S represents the total number of selected load data, ΔP i Indicates the target date load data P f With historical date load data P i The difference of values;

[0064] S102. Select a date range of similar days based on the target date. In this embodiment, the date range of similar days is from four weeks before the target date to the target date, and two weeks before and two weeks after the target date of the previous year. Select similar days within the date range of similar days based on the similarity of load data.

[0065] See Figure 2 In one embodiment, the steps of optimizing the variational mode decomposition method by using the dung beetle optimizer (DBO) are as follows:

[0066] First, the original signal is obtained, the parameter range of the variational mode decomposition is set, and the dung beetle algorithm is initialized;

[0067] Set the fitness function of the dung beetle algorithm and execute the algorithm. By calculating the fitness function value and iteration, record the best individual position (dung beetle position).

[0068] It is determined whether the maximum number of iterations has been reached. If not, iteration is performed to update the position of the dung beetle. If the maximum number of iterations has been reached, the globally optimal position of the dung beetle is output to optimize the parameters of the variational mode decomposition.

[0069] Variational Mode Decomposition (VMD) is a new, non-recursive, fully intrinsic and adaptive signal processing tool that is able to decompose a signal into sub-signals or IMFs. These IMFs are spectral components of the original signal that have a finite bandwidth and are compact around a central pulsation. Each IMF represents the characteristics of the original signal at a different frequency range, and by combining all the IMFs, the original signal can be reconstructed. In order to detect the bandwidth and center frequency of each IMF, VMD uses the optimal results of the evolutionary variation problem as a theoretical solution and iteratively maintains a constant optimization strategy. The decomposition problem of any signal can be defined as:

[0070]

[0071] Where: u k is the kth modal component; ω k is the kth center frequency; f(t) is the original load sequence; is the partial derivative operator; δ t is the unit pulse function.

[0072] The sum of all modes reconstructs the original signal f(t). This embodiment introduces a quadratic penalty term α and a Lagrange multiplier λ to solve this constrained variation in the signal reconstruction process. Formula (2) can be modified as follows:

[0073]

[0074] During each shift operation, an alternating direction method for multipliers (ADMMs) is introduced to generate different modes and center frequencies during each shift process. Using ADMMs, Equation (3) can be effectively minimized, and the center frequency and estimated mode display in the frequency domain are updated accordingly, as shown in Equations (4) and (5).

[0075]

[0076] In the formula is the Wiener filter of the current margin; is the center frequency of the current modal component; are the Fourier transforms of f(t), λ(t), and u(t) respectively.

[0077] The Dung Beetle Optimization (DBO) algorithm is a new optimization algorithm based on the behavior of dung beetles, which has the characteristics of fast convergence, high precision and high stability. The DBO algorithm mainly obtains the global optimal position and fitness value through the position change of dung beetles. It has five behaviors: dancing, rolling, stealing, foraging, and reproduction. Assuming that n represents the population size of the beetle, the specific steps are as follows:

[0078] (1) When the dung beetle rolls the ball forward, its position is updated and defined:

[0079]

[0080] Where: is the position of the ith dung beetle after t iterations; g is a random number -1 or 1; G is a random number within (0,0.2); is the global worst position; h is a random number in (0,1).

[0081] (2) When a dung beetle encounters an obstacle and cannot move forward, it needs to readjust itself by dancing to find a new route.

[0082]

[0083] Where: θ is a random number in [0,π], when When the dung beetle is in the same position, it remains unchanged.

[0084] (3) The reproduction process of dung beetles:

[0085]

[0086] Where: b1 and b2 are both independent 1*b dimensional random matrices;

[0087] are the upper and lower boundaries of the breeding area, respectively; is the current global best position.

[0088] (4) Update the current position of a dung beetle while foraging and define it as:

[0089]

[0090] Where: are the upper and lower bounds of the foraging area respectively; C1 is a 1*d dimensional matrix that obeys the normal distribution; C2 is a 1*d dimensional random matrix.

[0091] (5) Some dung beetles will steal dung balls from other dung beetles. At this time, their position update formula is:

[0092]

[0093] Where: g is a 1*d matrix that obeys the normal distribution; is the local optimal position; S is a constant.

[0094] Finally, the dung beetles that dance with the ball are grouped together, and the remaining dung beetles in each behavior are grouped separately. The position of each group is updated according to the behavior of the group to which it belongs, and multiple iterations are performed to obtain the global optimal position and its fitness value.

[0095] According to the VMD process and decomposition principle, it can be seen that the number of decompositions K and the penalty factor α have a significant impact on the result. As a preferred implementation of this embodiment, in step S200, in the step of optimizing the variational mode decomposition method using the dung beetle algorithm, the fitness function of the dung beetle algorithm is set to the minimum envelope entropy, specifically:

[0096]

[0097] Among them, E p is the minimum envelope entropy, p j represents the probability of an event occurring; m is the number of samples; a(j) is the Hilbert demodulated envelope sequence of the jth modal component decomposed by variational mode.

[0098] See Figure 3 As a preferred implementation of this embodiment, in step S300, the CNN-LSTM prediction model includes a CNN layer based on a CNN convolutional neural network, and an LSTM layer based on a LSTM long short-term memory neural network, and an attention layer; the steps of constructing parallel CNN-LSTM prediction models based on the decomposed components and training them include:

[0099] The decomposed components are converted into multivariate arrays and input into the CNN layer and LSTM layer for training respectively;

[0100] The outputs of the CNN layer and the LSTM layer are connected and input into the attention layer. The attention score is calculated through the attention layer and input into the fully connected layer. The load forecast value is output through normalization and two fully connected layers.

[0101] According to the loss between the load prediction value and the true value, the CNN-LSTM prediction model is trained through back propagation.

[0102] In this embodiment, the number of layers of the CNN and LSTM are both 3, and the output dimension is smaller than the input dimension.

[0103] In this embodiment, the specific steps of the attention mechanism are:

[0104] Step 1: Input the feature matrix I and obtain the query matrix Q, key matrix K and value matrix V through different matrix transformations.

[0105] Step 2: Calculate the attention score matrix A:

[0106] A=softmax(K T Q) (11)

[0107] Step 3: Calculate the output matrix O:

[0108] O=VA (12)

[0109] Based on the above implementation methods,

[0110] Embodiment 2:

[0111] This embodiment provides a load forecasting system based on data decomposition, including:

[0112] A data acquisition module is used to acquire historical load data of the target object, and select similar days to acquire historical load data of similar days similar to the target date; this module is used to implement the function of step S100 in the first embodiment, and will not be described in detail here;

[0113] A data decomposition module, which uses a variational mode decomposition method to decompose multiple components from the historical load data of similar days, and uses a dung beetle algorithm to optimize the variational mode decomposition method; this module is used to implement the function of step S200 in the first embodiment, and will not be repeated here;

[0114] A model training module, based on the decomposed components, respectively constructs parallel CNN-LSTM prediction models and performs training; this module is used to implement the function of step S300 in the first embodiment, and will not be described in detail here;

[0115] The load forecasting module is used to forecast the load of the target object on the target date through the trained CNN-LSTM forecasting model; this module is used to implement the function of step S400 in the first embodiment, which will not be repeated here.

[0116] As a preferred implementation of this embodiment, the method for selecting similar days is specifically as follows:

[0117] The weighted n-dimensional Euclidean distance is used as the standard to measure the similarity of load data on two different dates. The specific formula is as follows;

[0118]

[0119] ΔP i =P f -P i ,;

[0120] Where WED is the weighted n-dimensional Euclidean distance, w S represents the weight of the sth load data selected in the stepwise regression, S represents the total number of selected load data, ΔP i Indicates the target date load data P f With historical date load data P i The difference of values;

[0121] According to the target date, a date range of similar days is selected, and within the date range of similar days, similar days are selected according to the similarity of load data.

[0122] As a preferred implementation of this embodiment, in the step of optimizing the variational mode decomposition method using the dung beetle algorithm, the fitness function of the dung beetle algorithm is set to the minimum envelope entropy, specifically:

[0123]

[0124] Among them, p j represents the probability of an event occurring; m is the number of samples; a(j) is the Hilbert demodulated envelope sequence of the jth modal component decomposed by variational mode.

[0125] As a preferred implementation of this embodiment, the CNN-LSTM prediction model includes a CNN layer based on a CNN convolutional neural network, and an LSTM layer based on an LSTM long short-term memory neural network, and an attention layer; the steps of constructing parallel CNN-LSTM prediction models based on the decomposed components and training them include:

[0126] The decomposed components are converted into multivariate arrays and input into the CNN layer and LSTM layer for training respectively;

[0127] The outputs of the CNN layer and the LSTM layer are connected and input into the attention layer. The attention score is calculated through the attention layer and input into the fully connected layer. The load forecast value is output through normalization and the fully connected layer.

[0128] According to the loss between the load prediction value and the true value, the CNN-LSTM prediction model is trained through back propagation.

[0129] Embodiment three:

[0130] This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the load forecasting method based on data decomposition as described in any embodiment of the present invention is implemented.

[0131] Embodiment 4:

[0132] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the load forecasting method based on data decomposition as described in any embodiment of the present invention is implemented.

[0133] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0134] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0135] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0136] In several embodiments provided in the present application, if any function is implemented in the form of 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, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk and other media that can store program codes.

[0137] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A load forecasting method based on data decomposition, characterized in that: The following steps are involved: Obtain historical load data of the target object, and select similar days to obtain historical load data of similar days similar to the target date; The variational mode decomposition method is used to decompose multiple components from the historical load data of similar days, and the variational mode decomposition method is optimized using the dung beetle algorithm; Based on the decomposed components, parallel CNN-LSTM prediction models are constructed and trained respectively; The trained CNN-LSTM prediction model is used to predict the load of the target object on the target date.

2. A load forecasting method based on data decomposition according to claim 1, characterized in that: The method for selecting similar days is specifically as follows: The weighted n-dimensional Euclidean distance is used as the standard to measure the similarity of load data on two different dates. The specific formula is as follows; ΔP i =P f -P i ,; Where WED is the weighted n-dimensional Euclidean distance, w S represents the weight of the sth load data selected in the stepwise regression, S represents the total number of selected load data, ΔP i Indicates the target date load data P f With historical date load data P i The difference of values; According to the target date, a date range of similar days is selected, and within the date range of similar days, similar days are selected according to the similarity of load data.

3. A load forecasting method based on data decomposition according to claim 1, characterized in that: In the step of optimizing the variational mode decomposition method using the dung beetle algorithm, the fitness function of the dung beetle algorithm is set to the minimum envelope entropy, specifically: Where pj represents the probability of an event occurring; m is the number of samples; and a(j) is the Hilbert demodulated envelope sequence of the jth modal component decomposed by variational mode.

4. A load forecasting method based on data decomposition according to claim 1, characterized in that: The CNN-LSTM prediction model includes a CNN layer based on a CNN convolutional neural network, and an LSTM layer based on a LSTM long short-term memory neural network and an attention layer; The steps of constructing parallel CNN-LSTM prediction models based on the decomposed components and training them include: The decomposed components are converted into multivariate arrays and input into the CNN layer and LSTM layer for training respectively; The outputs of the CNN layer and the LSTM layer are connected and input into the attention layer. The attention score is calculated through the attention layer and input into the fully connected layer. The load forecast value is output through normalization and the fully connected layer. According to the loss between the load prediction value and the true value, the CNN-LSTM prediction model is trained through back propagation.

5. A load forecasting system based on data decomposition, characterized in that: include: The data acquisition module is used to acquire the historical load data of the target object and select similar days to acquire the historical load data of similar days similar to the target date; The data decomposition module uses the variational mode decomposition method to decompose multiple components from the historical load data of similar days, and uses the dung beetle algorithm to optimize the variational mode decomposition method; The model training module builds parallel CNN-LSTM prediction models based on the decomposed components and performs training. The load forecasting module is used to forecast the load of the target object on the target date through the trained CNN-LSTM forecasting model.

6. A load forecasting system based on data decomposition and seasonal capture according to claim 5, characterized in that: The method for selecting similar days is specifically as follows: The weighted n-dimensional Euclidean distance is used as the standard to measure the similarity of load data on two different dates. The specific formula is as follows; ΔP i =P f -P i ,; Where WED is the weighted n-dimensional Euclidean distance, w S represents the weight of the sth load data selected in the stepwise regression, S represents the total number of selected load data, ΔP i Indicates the target date load data P f With historical date load data P i The difference of values; According to the target date, a date range of similar days is selected, and within the date range of similar days, similar days are selected according to the similarity of load data.

7. A load forecasting system based on data decomposition and seasonal capture according to claim 5, characterized in that: In the step of optimizing the variational mode decomposition method using the dung beetle algorithm, the fitness function of the dung beetle algorithm is set to the minimum envelope entropy, specifically: Where pj represents the probability of an event occurring; m is the number of samples; and a(j) is the Hilbert demodulated envelope sequence of the jth modal component decomposed by variational mode.

8. The load forecasting system based on data decomposition and seasonal capture according to claim 5, characterized in that: The CNN-LSTM prediction model includes a CNN layer based on a CNN convolutional neural network, and an LSTM layer based on a LSTM long short-term memory neural network and an attention layer; The steps of constructing parallel CNN-LSTM prediction models based on the decomposed components and training them include: The decomposed components are converted into multivariate arrays and input into the CNN layer and LSTM layer for training respectively; The outputs of the CNN layer and the LSTM layer are connected and input into the attention layer. The attention score is calculated through the attention layer and input into the fully connected layer. The load forecast value is output through normalization and the fully connected layer. According to the loss between the load prediction value and the true value, the CNN-LSTM prediction model is trained through back propagation.

9. 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 program, the load forecasting method based on data decomposition as described in any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the load forecasting method based on data decomposition as described in any one of claims 1 to 4 is implemented.

Images