A Short-Term Load Forecasting Method and System Based on Improved GWO-VMD-LSTM

Abstract

The present invention discloses a short-term load forecasting method and system based on improved GWO-VMD-LSTM. The method includes: performing parameter optimization on the number of modal decompositions and penalty factors that need to be determined before VMD feature extraction, performing variational modal decomposition on the load feature time series signal, decomposing it into a series of intrinsic mode function components, using an improved wavelet packet soft threshold function, proposing a sluggish-greed mechanism to improve the function, denoising the IMF component signals, extracting the characteristic coefficients and modal kurtosis values of each component, and evaluating the denoising effect; improving the network structure of the long short-term neural network, proposing a parameter adaptive adjustment and optimization algorithm, dynamically optimizing the learning rate of LSTM, and inputting the data into the network for training and prediction. It can improve the accuracy of short-term load forecasting and the precision of classification, which is beneficial to the short-term accurate forecasting of power load.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric load forecasting, and particularly relates to a short-term load forecasting method and system based on improved GWO-VMD-LSTM. Background Art

[0002] The electric power industry is the basic guarantee for social production and people's daily life. It not only affects the overall development speed of the social economy but also is closely related to everyone's daily life. Compared with other energy sources, electric power energy has many advantages, such as low pollution, high efficiency, easy conversion, and convenient long-distance transmission. Therefore, it has become one of the indispensable important energy sources. The essence of electric load forecasting is to comprehensively consider the operating characteristics of the power system, influencing factors, etc., and use scientific and reasonable methods to explore the laws and development trends of historical electric loads, and estimate and infer the electric load demand values for a period of time in the future.

[0003] At present, the long short-term memory neural network provides a new idea for solving problems. It collects various original electric load data, analyzes the characteristics of load influencing factors, and conducts short-term electric load forecasting through the long short-term memory neural network. Summary of the Invention

[0004] The present invention provides a short-term load forecasting method and system based on improved GWO-VMD-LSTM to solve the technical problems of large forecasting difficulty caused by complex non-linearity and large volatility of electric load data, and insufficient robustness and accuracy of traditional methods.

[0005] In a first aspect, the present invention provides a short-term load forecasting method based on improved GWO-VMD-LSTM, including:

[0006] Determining the number of mode decompositions and the penalty factor of variational mode decomposition according to the improved grey wolf optimization algorithm;

[0007] Performing variational mode decomposition on the load characteristic time series signal according to the number of mode decompositions and the penalty factor to obtain at least one IMF component signal, where the IMF component signal includes a load influencing sub-signal and a load result sub-signal corresponding to the load influencing sub-signal;

[0008] Performing denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal;

[0009] Inputting the at least one target IMF component signal into an improved long short-term neural network, and dynamically optimizing the learning rate of the improved long short-term neural network according to a preset parameter adaptive adjustment optimization algorithm to obtain a short-term load forecasting model;

[0010] Obtain the real-time load impact sub-signal, input the real-time load impact sub-signal into the short-term load prediction model, and the short-term load prediction model outputs the load prediction result.

[0011] In a second aspect, the present invention provides a short-term load prediction system based on improved GWO-VMD-LSTM, including:

[0012] A determination module configured to determine the number of mode decompositions and the penalty factor of variational mode decomposition according to the improved grey wolf optimization algorithm;

[0013] A decomposition module configured to perform variational mode decomposition on the load characteristic time series signal according to the number of mode decompositions and the penalty factor to obtain at least one IMF component signal, where the IMF component signal includes a load impact sub-signal and a load result sub-signal corresponding to the load impact sub-signal;

[0014] A processing module configured to perform denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal;

[0015] A training module configured to input the at least one target IMF component signal into an improved long short-term neural network, and dynamically optimize the learning rate of the improved long short-term neural network according to a preset parameter adaptive adjustment optimization algorithm to obtain a short-term load prediction model;

[0016] An output module configured to obtain a real-time load impact sub-signal, input the real-time load impact sub-signal into the short-term load prediction model, and the short-term load prediction model outputs the load prediction result.

[0017] In a third aspect, an electronic device is provided, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the short-term load prediction method based on improved GWO-VMD-LSTM according to any embodiment of the present invention.

[0018] In a fourth aspect, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program instructions are executed by a processor, the processor is enabled to execute the steps of the short-term load prediction method based on improved GWO-VMD-LSTM according to any embodiment of the present invention.

[0019] The short-term load forecasting method and system based on improved GWO-VMD-LSTM of the present application improve the grey wolf optimization algorithm, perform parameter optimization on the number of modal decompositions and penalty factors that need to be determined before VMD feature extraction, perform variational modal decomposition on the load feature time series signal, decompose it into a series of intrinsic mode function components, use an improved wavelet packet soft threshold function, propose a sluggish-greed mechanism to improve the function, perform denoising processing on the IMF component signal, extract the feature coefficients and modal kurtosis values of each component, and evaluate the denoising effect; improve the network structure of the long short-term neural network, propose a parameter adaptive adjustment and optimization algorithm, dynamically optimize the learning rate of the LSTM, and input the data into the network for training and prediction, which can improve the accuracy of short-term load forecasting and the precision of classification, and is conducive to the short-term accurate forecasting of electric power load. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0021] Figure 1 It is a flowchart of a short-term load forecasting method based on improved GWO-VMD-LSTM provided by an embodiment of the present invention;

[0022] Figure 2 It is a structural block diagram of a short-term load forecasting system based on improved GWO-VMD-LSTM provided by an embodiment of the present invention;

[0023] Figure 3 It is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0025] Please refer to Figure 1 , which shows a flowchart of a short-term load forecasting method based on improved GWO-VMD-LSTM of the present application.

[0026] As Figure 1As shown in the figure, the short-term load forecasting method based on the improved GWO-VMD-LSTM specifically includes the following steps:

[0027] Step S101, determine the number of mode decompositions and the penalty factor of variational mode decomposition according to the improved grey wolf optimization algorithm.

[0028] In this step, for the improved grey wolf optimization algorithm (GWO), parameter optimization is carried out on the number of mode decompositions and the penalty factor that need to be determined before variational mode decomposition feature extraction. The specific content is as follows: an envelope exploration weighting factor is introduced, enabling the weighting factor to be appropriately and timely adjusted dynamically based on the change of the wolf pack hunting state, so as to enhance the adaptability and mobility of the algorithm, and then better cope with the complex hunting optimization environment. To expand the global hunting coverage range of the algorithm, a leading wolf hunting and cooperation strategy is proposed. Under the leading wolf hunting and cooperation strategy, the wolf pack continuously adjusts the positions of individuals in the group and approaches the prey orderly for hunting.

[0029] Specifically: calculate the envelope escape coefficient and the exploration weighting factor at the current iteration. The expression is:

[0030] ,

[0031] ,

[0032] In the formula, is the envelope escape coefficient at the t-th iteration, is the exploration weighting factor at the t-th iteration, is the total number of grey wolf individuals, is the probability distribution sequence of the i-th grey wolf individual, is the wolf pack aliasing parameter sequence at the t-th iteration, is the exploration correction coefficient, is the objective weight coefficient, is the spacing between the upper and lower limits of the iteration times, is a non-zero number within 0 to 1, is the wolf pack weighting factor;

[0033] Determine the leading wolf hunting and cooperation strategy of the grey wolf algorithm. The expression is:

[0034] ,

[0035] ,

[0036] In the formula, , , are respectively the leading wolf of the wolf pack, the leading wolf of the wolf pack, The pursuit influence coefficient between the alpha wolf and other gray wolf individuals in the wolf pack is the comprehensive pursuit influence coefficient of the alpha wolf in the wolf pack is the average function 、 、 are respectively the azimuth factor of the alpha wolf in the wolf pack, the azimuth factor of the alpha wolf in the wolf pack, the azimuth factor of the alpha wolf in the wolf pack, 、 、 are respectively the current position of the alpha wolf in the wolf pack, the current position of the alpha wolf in the wolf pack, the current position of the alpha wolf in the wolf pack, 、 、 are respectively the azimuth correlation coefficient of the alpha wolf in the wolf pack, the azimuth correlation coefficient of the alpha wolf in the wolf pack, the azimuth correlation coefficient of the alpha wolf in the wolf pack, is the current position of the prey, 、 、 are respectively the offset coefficient of the wolf pack, the offset coefficient of the wolf pack, the offset coefficient of the wolf pack, is the two-norm operator, is the position of the gray wolf individual at the t-th iteration, is the convergence resolution coefficient, is the azimuth coefficient of the prey at the t-th iteration, is the natural constant, and are respectively the upper and lower limits of the hunting range, is the prey offset probability factor with a range of 0 to 1;

[0037] Taking the combination of the number of modal decompositions and penalty factors that need to be optimized as the original input of the gray wolf iteration position, the iterative updated position output, that is, the optimal solution combination, is expressed as:

[0038] ,

[0039] In the formula, is the iterative updated position of the n-th gray wolf in the m-dimensional space, is the matrix window function, is the prey azimuth constant, is the prey space offset factor, is the azimuth inertial matrix of the grey wolf, is the maximum number of iterations, is the determination coefficient of the initial azimuth of the prey.

[0040] Step S102: Perform variational mode decomposition on the load characteristic time series signal according to the number of modal decompositions and the penalty factor to obtain at least one IMF component signal, where the IMF component signal includes a load influence sub-signal and a load result sub-signal corresponding to the load influence sub-signal.

[0041] In this step, variational mode decomposition (VMD) is performed on the load characteristic time series signal, and it is decomposed into a series of intrinsic mode function components (IMF). Specifically, the original time series signal is decomposed into several intrinsic mode functions with different bandwidth constraints and fluctuating around the center frequency. The Wiener filter and the alternating direction multiplier method are used to iteratively update each sub-modal function and the center frequency. The ultimate goal is that the sum of the IMFs is reconstructed as close as possible to the original signal and the sum of the modal bandwidths is minimized. Specifically:

[0042] Construct a variational problem equation to calculate the IMF components decomposed from the original signal , continue to solve the square norm of the gradient of the demodulated signal to obtain the bandwidth of each mode, and construct a constrained variational problem equation, and the expression is:

[0043] ,

[0044] ,

[0045] In the formula, is the amplitude of, is the phase angle of, is the original signal, is the set of IMF components obtained by decomposition, is the set of center frequencies of the is the Tikhonov matrix, is the impulse function, is the imaginary unit, is the number of modal components;

[0046] Convert the constrained variational problem into an unconstrained form, construct a Lagrangian augmented function, and use the alternating direction multiplier method for the th IMF component and the The central frequency of a mode Iteratively update by completing the update for K intrinsic mode functions, that is, for the Lagrange multipliers Perform the update, and the expression is:

[0047] ,

[0048] ,

[0049] ,

[0050] ,

[0051] In the formula, is The augmented Lagrangian function of is the impulse function, is the quadratic penalty term, is the imaginary unit, is the current iteration number, , , are respectively The Fourier transform of The Fourier transform of The Fourier transform of is the frequency, is the Central frequency of the th mode, is the Central frequency of the th mode at the th iteration, is the Central frequency of the th mode at the th iteration, is the Fourier transform of the Lagrange multiplier at the th iteration, is the Fourier transform of the Lagrange multiplier at the th iteration, is the At the th iteration Fourier transform of;

[0052] If the convergence condition is satisfied, stop the iteration and output at least one IMF component signal, otherwise continue the iteration. The expression of the convergence condition is:

[0053] ,

[0054] In the formula, is the determination accuracy and satisfies .

[0055] Step S103: Denoise the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal.

[0056] In this step, a wavelet packet soft threshold function is constructed to enable the IMF components to better reflect the characteristics of the original signal at different frequencies. A sluggish-greed mechanism is proposed to denoise the IMF component signals, extract the characteristic coefficients and modal kurtosis values of each component, and evaluate the denoising effect, effectively suppressing the "over-killing" phenomenon, improving the signal-to-noise ratio of the denoised signal, facilitating the extraction of the influencing characteristics for power load forecasting, and enhancing the analysis accuracy and reliability of the power load data time series. Specifically:

[0057] Construct a wavelet packet soft threshold function to obtain the IMF components after wavelet packet transform, and calculate the wavelet packet soft threshold. The expression is:

[0058] ,

[0059] ,

[0060] In the formula, is the k-th IMF component after wavelet packet transform, is the spectral flatness coefficient, is the number of IMF components, is the signal-to-noise ratio of the i-th IMF component, is the frequency parameter, is the sliding parameter, is the wavelet basis constant, is the wavelet noise factor, is the projection value of the original signal of the i-th IMF component, is the characteristic information parameter, is the wavelet packet soft threshold, is the sign function, is the sign function, is the local threshold;

[0061] Considering that there are many influencing factors for power load forecasting and the complexity of the noise signal is relatively high, to improve the prediction accuracy of the algorithm, an IMF component sluggish-greed denoising mechanism is proposed. This mechanism obtains the greed trade-off factor according to the frequency characteristics and perturbation characteristics of the noise, which is used as the judgment consideration parameter for the sluggish function, calculates the wavelet packet soft threshold coefficient to denoise the IMF components, and screens out the interference noise signals that are not conducive to the accurate prediction of the algorithm. The expression is:

[0062] ,

[0063] ,

[0064] ,

[0065] In the formula, is the greed trade-off factor, is the activation function, is the Sluggish parsing function, is the IMF component weight matrix, is the IMF component bias matrix, is the wavelet gradient state parameter, is the interference weight, is the th wavelet packet soft threshold coefficient of the kth IMF component, is the th execution threshold of the kth IMF component, is the co-frequency perturbation factor, is the th gamma weight of the kth IMF component, is the kth IMF component after wavelet packet soft threshold denoising.

[0066] Step S104: Input the at least one target IMF component signal into the improved long short-term neural network, and dynamically optimize the learning rate of the improved long short-term neural network according to the preset parameter adaptive adjustment optimization algorithm to obtain a short-term load prediction model.

[0067] In this step, the expression of the loss function of the improved long short-term neural network is:

[0068] ,

[0069] In the formula, is the Boltzmann-cross entropy loss function, is the sample, is the true value, is the predicted value, is the cross entropy quantization factor, is the generalization coefficient.

[0070] The learning rate is an important hyperparameter that controls the parameter update step size of the LSTM model, and determines the size of the parameter adjustment along the opposite direction of the gradient during each parameter update. If the learning rate is too large or too small, the model cannot converge, the training process is slow, the convergence speed is slow, and it may fall into a local optimal solution. Therefore, a Levy-Adam parameter optimization algorithm is introduced to automatically adjust the learning rate according to the gradient of the parameters to adapt to the update requirements of different parameters, so as to accelerate the training process of the model and improve the performance of the model. The expression of the parameter adaptive adjustment optimization algorithm is:

[0071] ,

[0072] ,

[0073] ,

[0074] ,

[0075] ,

[0076] wherein, is the Adam first-order momentum at the t-th iteration, is the Adam first-order momentum at the (t-1)-th iteration, is the first-order momentum hyperparameter, is the Levy distribution coefficient, is the Adam second-order momentum at the t-th iteration, is the Adam second-order momentum at the (t-1)-th iteration, is the second-order momentum hyperparameter, is the Levy update value of the first-order momentum, is the momentum gradient descent rate, is the Levy correction exponent, is the Levy update value of the second-order momentum, is the hyperparameter update amount of the learning rate at the t-th iteration, is the learning rate at the t-th iteration, is the learning rate at the (t-1)-th iteration.

[0077] In step S105, obtain the real-time load impact sub-signal, input the real-time load impact sub-signal into the short-term load prediction model, and the short-term load prediction model outputs a load prediction result.

[0078] In summary, the method and system of the present application improve the gray wolf optimization algorithm, perform parameter optimization on the number of modal decompositions and penalty factors that need to be determined before VMD feature extraction, perform variational modal decomposition on the load feature time series signal, decompose it into a series of intrinsic mode function components, use an improved wavelet packet soft threshold function, propose a sluggish-greed mechanism to improve the function, perform denoising processing on the IMF component signals, extract the feature coefficients and modal kurtosis values of each component, and evaluate the denoising effect; improve the network structure of the long short-term neural network, propose a parameter adaptive adjustment and optimization algorithm, dynamically optimize the learning rate of the LSTM, and input the data into the network for training and prediction, which can improve the accuracy of short-term load prediction and the precision of classification, and is beneficial to the short-term accurate prediction of power load.

[0079] Please refer to Figure 2 , which shows a structural block diagram of a short-term load forecasting system based on improved GWO-VMD-LSTM of the present application.

[0080] As shown in Figure 2 , the short-term load forecasting system 200 includes a determination module 210, a decomposition module 220, a processing module 230, a training module 240, and an output module 250.

[0081] Among them, the determination module 210 is configured to determine the number of mode decompositions and the penalty factor of variational mode decomposition according to the improved grey wolf optimization algorithm; the decomposition module 220 is configured to perform variational mode decomposition on the load characteristic time series signal according to the number of mode decompositions and the penalty factor to obtain at least one IMF component signal, where the IMF component signal includes a load influence sub-signal and a load result sub-signal corresponding to the load influence sub-signal; the processing module 230 is configured to perform denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal; the training module 240 is configured to input the at least one target IMF component signal into an improved long short-term neural network, and dynamically optimize the learning rate of the improved long short-term neural network according to a preset parameter adaptive adjustment optimization algorithm to obtain a short-term load forecasting model; the output module 250 is configured to obtain a real-time load influence sub-signal, input the real-time load influence sub-signal into the short-term load forecasting model, and the short-term load forecasting model outputs a load forecasting result.

[0082] It should be understood that Figure 2 the modules described in Figure 1 correspond to the respective steps in the method described with reference to Figure 2 . Thus, the operations and features described above for the method and the corresponding technical effects also apply to the modules in

[0083] and will not be elaborated herein.

[0084] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium, on which a computer program is stored. When the program instructions are executed by a processor, the processor is caused to execute the short-term load forecasting method based on improved GWO-VMD-LSTM in any of the above method embodiments;

[0084] As an implementation manner, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are set as:

[0085] Determine the number of mode decompositions and the penalty factor of variational mode decomposition according to the improved grey wolf optimization algorithm;

[0086] Perform variational mode decomposition on the load characteristic time series signal according to the modal decomposition quantity and the penalty factor to obtain at least one IMF component signal, where the IMF component signal includes a load influence sub-signal and a load result sub-signal corresponding to the load influence sub-signal;

[0087] Perform denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal;

[0088] Input the at least one target IMF component signal into an improved long short-term neural network, and dynamically optimize the learning rate of the improved long short-term neural network according to a preset parameter adaptive adjustment and optimization algorithm to obtain a short-term load prediction model;

[0089] Obtain a real-time load influence sub-signal, input the real-time load influence sub-signal into the short-term load prediction model, and the short-term load prediction model outputs a load prediction result.

[0090] A computer-readable storage medium may include a storage program area and a storage data area. Among them, the storage program area may store an operating system and application programs required for at least one function; the storage data area may store data created according to the use of the short-term load prediction system based on improved GWO-VMD-LSTM, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the computer-readable storage medium may optionally include a memory remotely provided with respect to the processor, and these remote memories may be connected to the short-term load prediction system based on improved GWO-VMD-LSTM through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0091] Figure 3 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention, as Figure 3 shown. The device includes: a processor 310 and a memory 320. The electronic device may further include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330, and the output device 340 may be connected through a bus or other means. Figure 3Take the bus connection as an example. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 320, that is, implements the short-term load forecasting method based on the improved GWO-VMD-LSTM in the above method embodiment. The input device 330 can receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the short-term load forecasting system based on the improved GWO-VMD-LSTM. The output device 340 can include display devices such as a display screen.

[0092] The above electronic device can execute the method provided by the embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiment of the present invention.

[0093] As an implementation manner, the above electronic device is applied to a short-term load forecasting system based on the improved GWO-VMD-LSTM and is used for a client, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to:

[0094] Determine the number of mode decompositions and the penalty factor of variational mode decomposition according to the improved grey wolf optimization algorithm;

[0095] Perform variational mode decomposition on the load characteristic time series signal according to the number of mode decompositions and the penalty factor to obtain at least one IMF component signal, wherein the IMF component signal includes a load impact sub-signal and a load result sub-signal corresponding to the load impact sub-signal;

[0096] Perform denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal;

[0097] Input the at least one target IMF component signal into an improved long short-term neural network, and dynamically optimize the learning rate of the improved long short-term neural network according to a preset parameter adaptive adjustment optimization algorithm to obtain a short-term load forecasting model;

[0098] Obtain a real-time load impact sub-signal, input the real-time load impact sub-signal into the short-term load forecasting model, and the short-term load forecasting model outputs a load forecasting result.

[0099] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A short-term load forecasting method based on improved GWO-VMD-LSTM, characterized in that: include: Determine the modal decomposition quantity and penalty factor of the variational modal decomposition according to the improved grey wolf optimization algorithm, wherein the modal decomposition quantity and penalty factor of the variational modal decomposition according to the improved grey wolf optimization algorithm comprises: Calculate the envelope escape coefficient and exploration weight factor under the current iteration. The expression is: ， ， In the formula, is the envelope escape coefficient at the tth iteration, is the exploration weight factor at the tth iteration, is the total number of gray wolf individuals, is the probability distribution sequence of the i-th gray wolf individual, is the wolf pack aliasing parameter sequence at the tth iteration, is the exploration correction factor, is the objective weight coefficient, is the upper and lower limit interval of the number of iterations, is a non-zero number between 0 and 1. Empowerment factor for the wolf pack; Determine the leader hunting leadership cooperation strategy of the gray wolf algorithm, the expression is: ， ， In the formula, , , They are Wolf pack leader, Wolf pack leader, The hunting influence coefficient between the wolf pack leader and other gray wolves, is the comprehensive hunting influence coefficient of the wolf pack leader, is the mean function, , , They are The orientation factor of the wolf pack leader, The orientation factor of the wolf pack leader, The orientation factor of the wolf pack leader, , , They are The current location of the wolf pack leader, The current location of the wolf pack leader, The current location of the wolf pack leader. , , They are The orientation correlation coefficient of the wolf pack leader, The orientation correlation coefficient of the wolf pack leader, The orientation correlation coefficient of the wolf pack leader, is the current position of the prey, , , They are The deviation coefficient of the wolf pack, The deviation coefficient of the wolf pack, The deviation coefficient of the wolf pack, is the two-norm operator, is the position of the gray wolf individual at the tth iteration, is the convergence resolution coefficient, is the orientation coefficient of the prey at the tth iteration, is a natural constant, and are the upper and lower limits of the hunting range, is the prey displacement probability factor ranging from 0 to 1; The number of modal decompositions and the penalty factor combination that need to be optimized are used as the original input of the gray wolf iterative position, and the iterative update position output, that is, the optimal solution combination, is obtained. The expression is: ， In the formula, is the iterative update position of the nth gray wolf in the m-dimensional space, is the matrix window function, is the prey position constant, is the prey space offset factor, is the gray wolf's azimuth inertia matrix, is the maximum number of iterations, is the determination coefficient of the initial orientation of the prey; Performing variational mode decomposition on the load characteristic time series signal according to the mode decomposition quantity and the penalty factor to obtain at least one IMF component signal, wherein the IMF component signal includes a load influence sub-signal and a load result sub-signal corresponding to the load influence sub-signal; Performing denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal; Inputting the at least one target IMF component signal into an improved long-term and short-term neural network, dynamically optimizing the learning rate of the improved long-term and short-term neural network according to a preset parameter adaptive adjustment optimization algorithm, and obtaining a short-term load forecasting model; A real-time load impact sub-signal is obtained, and the real-time load impact sub-signal is input into the short-term load forecasting model, and the short-term load forecasting model outputs a load forecasting result.

2. According to the method for short-term load forecasting based on improved GWO-VMD-LSTM in claim 1, it is characterized in that: The performing variational modal decomposition on the load characteristic time series signal according to the modal decomposition quantity and the penalty factor to obtain at least one IMF component signal comprises: Construct the variational problem equation and calculate the IMF components decomposed from the original signal , continue to solve the square norm of the demodulated signal gradient to obtain the bandwidth of each mode, and construct the constrained variational problem equation, the expression is: ， ， In the formula, for The amplitude of for The phase angle, is the original signal, For decomposition A set of IMF components, for The set of center frequencies of the modes, is the Tikhonov matrix, is the impulse function, is an imaginary unit, is the number of modal components; The constrained variational problem is transformed into an unconstrained form, the Lagrangian augmented function is constructed, and the alternating direction multiplier method is used to solve the IMF Components and The center frequency of the mode Iterative update is completed by updating K intrinsic mode functions, that is, updating the Lagrange multiplier To update, the expression is: ， ， ， ， In the formula, for The Lagrangian augmented function of is the pulse function, is the quadratic penalty term, is an imaginary unit, is the current iteration number, , , They are The Fourier transform of The Fourier transform of The Fourier transform of is the frequency, For the The center frequency of the mode, For the iteration The center frequency of the mode, is the Fourier transform of the Lagrange multiplier at the iteration, is the Fourier transform of the Lagrange multiplier at the iteration, For the first iteration Fourier transform of If the convergence condition is met, the iteration is stopped and at least one IMF component signal is output, otherwise the iteration continues. The expression of the convergence condition is: ， In the formula, To determine the accuracy and meet .

3. The short-term load forecasting method based on improved GWO-VMD-LSTM according to claim 1 is characterized in that: The performing denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal comprises: Construct the wavelet packet soft threshold function, obtain the IMF component after wavelet packet transformation, and calculate the wavelet packet soft threshold. The expression is: ， ， In the formula, is the IMF component after the kth wavelet packet transform, is the spectral flatness coefficient, is the number of IMF components, For the The signal-to-noise ratio of the IMF components, is the frequency parameter, is the sliding parameter, is the wavelet basis constant, is the wavelet noise factor, For the The original signal projection value of the IMF component, is the characteristic information parameter, is the wavelet packet soft threshold, is the symbolic function, is the local threshold; Get the greed trade-off factor , as the judgment parameter of the sluggish function, calculate the wavelet packet soft threshold coefficient to denoise the IMF component. The expression is: ， ， ， In the formula, is the greed trade-off factor, is the activation function, For Sluggish parsing function, is the IMF component weight matrix, is the IMF component bias matrix, is the wavelet gradient state parameter, is the interference weight, For the The wavelet packet soft threshold coefficients of the IMF components are For the The execution threshold value of each IMF component, is the same frequency disturbance factor, For the The gamma weights of the IMF components, is the kth IMF component denoised by wavelet packet soft thresholding.

4. The short-term load forecasting method based on improved GWO-VMD-LSTM according to claim 1 is characterized in that: After performing denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function, the method further includes: According to the wavelet packet soft threshold coefficient obtained by the sluggish-greed mechanism ,Will As the input parameter of IMF judgment index, each IMF component decomposed by VMD is reconstructed and demodulated, and the denoising effect judgment index of each IMF component, namely the modal kurtosis value and characteristic coefficient, is calculated. The expression is: ， ， ， ， In the formula, is the noise signal sensitive period, is the clockwise contour integral, For the The signal sampling frequency of the IMF component is is the signal noise frequency, is the maximum frequency of the IMF component, For the The modal kurtosis value of the IMF component, is the mixing kurtosis parameter, For the The covariance of the IMF components, is the global interference coefficient of white noise, For the The shifted mean of the IMF components, is the shift coefficient, For the The ambient noise parameters of the IMF components are: For the The characteristic coefficients of the IMF components are is the signal strength coefficient.

5. The short-term load forecasting method based on improved GWO-VMD-LSTM according to claim 1 is characterized in that: The expression of the parameter adaptive adjustment optimization algorithm is: ， ， ， ， In the formula, is the first-order Adam momentum at the t-th iteration, is the first-order Adam momentum at the t-1th iteration, is the first-order momentum hyperparameter, is the Levy distribution coefficient, is the Adam second-order momentum at the t-th iteration, is the Adam second-order momentum at the t-1th iteration, is the second-order momentum hyperparameter, is the Levy update value of the first-order momentum, is the momentum gradient descent rate, is the Levy modified index, is the Levy update value of the second-order momentum, is the hyperparameter update of the learning rate at the tth iteration, is the learning rate at the tth iteration, is the learning rate at the t-1th iteration.

6. The short-term load forecasting method based on improved GWO-VMD-LSTM according to claim 1 is characterized in that: The expression of the loss function of the improved long short-term neural network is: ， In the formula, is the Boltzmann-cross entropy loss function, For the sample, is the true value, is the predicted value, is the cross entropy quantization factor, is the generalization coefficient.

7. A short-term load forecasting system based on improved GWO-VMD-LSTM, characterized in that: include: A determination module is configured to determine the modal decomposition quantity and penalty factor of the variational modal decomposition according to the improved grey wolf optimization algorithm, wherein the determination of the modal decomposition quantity and penalty factor of the variational modal decomposition according to the improved grey wolf optimization algorithm includes: Calculate the envelope escape coefficient and exploration weight factor under the current iteration. The expression is: ， ， In the formula, is the envelope escape coefficient at the tth iteration, is the exploration weight factor at the tth iteration, is the total number of gray wolf individuals, is the probability distribution sequence of the i-th gray wolf individual, is the wolf pack aliasing parameter sequence at the tth iteration, is the exploration correction factor, is the objective weight coefficient, is the upper and lower limit interval of the number of iterations, is a non-zero number between 0 and 1. Empowerment factor for the wolf pack; Determine the leader hunting leadership cooperation strategy of the gray wolf algorithm, the expression is: ， ， In the formula, , , They are Wolf pack leader, Wolf pack leader, The hunting influence coefficient between the wolf pack leader and other gray wolves, is the comprehensive hunting influence coefficient of the wolf pack leader, is the mean function, , , They are The orientation factor of the wolf pack leader, The orientation factor of the wolf pack leader, The orientation factor of the wolf pack leader, , , They are The current location of the wolf pack leader, The current location of the wolf pack leader, The current location of the wolf pack leader. , , They are The orientation correlation coefficient of the wolf pack leader, The orientation correlation coefficient of the wolf pack leader, The orientation correlation coefficient of the wolf pack leader, is the current position of the prey, , , They are The deviation coefficient of the wolf pack, The deviation coefficient of the wolf pack, The deviation coefficient of the wolf pack, is the two-norm operator, is the position of the gray wolf individual at the tth iteration, is the convergence resolution coefficient, is the orientation coefficient of the prey at the tth iteration, is a natural constant, and are the upper and lower limits of the hunting range, is the prey displacement probability factor ranging from 0 to 1; The number of modal decompositions and the penalty factor combination that need to be optimized are used as the original input of the gray wolf iterative position, and the iterative update position output, that is, the optimal solution combination, is obtained. The expression is: ， In the formula, is the iterative update position of the nth gray wolf in the m-dimensional space, is the matrix window function, is the prey position constant, is the prey space offset factor, is the gray wolf's azimuth inertia matrix, is the maximum number of iterations, is the determination coefficient of the initial orientation of the prey; a decomposition module configured to perform variational mode decomposition on the load characteristic time series signal according to the mode decomposition quantity and the penalty factor to obtain at least one IMF component signal, wherein the IMF component signal includes a load impact sub-signal and a load result sub-signal corresponding to the load impact sub-signal; A processing module, configured to perform denoising processing on the at least one IMF component signal according to a preset wavelet packet soft threshold function to obtain at least one target IMF component signal; A training module is configured to input the at least one target IMF component signal into an improved long-term and short-term neural network, dynamically optimize the learning rate of the improved long-term and short-term neural network according to a preset parameter adaptive adjustment optimization algorithm, and obtain a short-term load forecasting model; The output module is configured to obtain a real-time load impact sub-signal, input the real-time load impact sub-signal into the short-term load forecasting model, and the short-term load forecasting model outputs a load forecasting result.

8. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images