Short-term electricity consumption prediction method and device based on secondary VMD-xLSTM-KAN model, medium and product
By using a decomposition and improvement model based on the quadratic VMD-xLSTM-KAN model to predict short-term electricity consumption, the problem of insufficient prediction accuracy in existing technologies is solved, and higher prediction accuracy and stronger ability to capture long-term dependencies are achieved.
Patent Information
- Application Number
- CN202511709815.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-20
- Publication Date
- 2026-02-13
AI Technical Summary
Existing short-term electricity consumption forecasting methods lack sufficient accuracy, struggle to effectively capture long-term dependencies, and have limited information storage capacity, leading to inaccurate forecast results.
A quadratic VMD-xLSTM-KAN model is adopted, and the electricity consumption sequence is decomposed into multiple modal components and residual components through variational mode decomposition algorithm. The output gate and fully connected layer of the LSTM model are replaced by an improved KAN layer to predict the high-frequency and mid-to-low-frequency components respectively. The results are superimposed and combined to form the total electricity consumption prediction result.
It significantly improves the accuracy of short-term electricity consumption forecasting, enhances the model's ability to capture long-term dependencies and its output expression capabilities, and optimizes the overall forecasting performance.
Smart Images
Figure CN121525976A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to the technical field of power consumption prediction, in particular to a short-term power consumption prediction method, device, medium and product based on a quadratic VMD-xLSTM-KAN model. BACKGROUND
[0002] The rapid development of the state and social economy promotes the increasing demand for electricity, which brings challenges to the safe and stable operation of the power system. Power prediction can provide important data basis for power system dispatching and operation, so continuously improving the accuracy of power prediction is an inevitable requirement for the development of the power system. Among them, short-term power is data with a period of days or smaller time units, which is highly random and strongly nonlinear due to the influence of various external factors, so that short-term power prediction has become the focus and hotspot of time series prediction problems in the power industry. However, the existing short-term power prediction method has insufficient prediction accuracy. SUMMARY
[0003] The purpose of the present application is to provide a short-term power consumption prediction method, device, medium and product based on a quadratic VMD-xLSTM-KAN model, which can effectively improve the accuracy of short-term power consumption prediction.
[0004] To achieve the above purpose, the present application provides the following solutions: In a first aspect, the present application provides a short-term power consumption prediction method based on a quadratic VMD-xLSTM-KAN model, comprising: obtaining an original power sequence; using a variational mode decomposition algorithm to decompose the original power sequence to obtain K modal components and one residual component; using a variational mode decomposition algorithm to further extract data feature information by twice decomposing the residual component to obtain M twice-decomposed residual components; According to a preset threshold, the K modal components and the M twice-decomposed residual components are divided into high-frequency components and medium-low-frequency components; using a trained first improved model to predict the medium-low-frequency components to obtain a first prediction result; the first improved model is a deep learning model obtained by replacing the weight matrix in the output door of the extended long short-term memory network with a KAN layer; using a trained second improved model to predict the high-frequency components to obtain a second prediction result; the second improved model is a deep learning model obtained by replacing the fully connected layer of the extended long short-term memory network with a KAN layer; stacking and combining the first prediction result and the second prediction result to obtain a prediction result of total power data.
[0005] Optionally, the overall structure of the KAN layer includes: ; in, y The output of the KAN layer. x As the input to the KAN layer, This is the inter-layer mapping matrix of the (L-1)th KAN layer. This is the inter-layer mapping matrix of the (L-2)th KAN layer. This is the inter-layer mapping matrix of the 0th KAN layer (initial inter-layer mapping matrix). The expression for calculating the interlayer mapping of the Lth KAN layer is as follows: ; in, Let be the interlayer mapping matrix of the Lth KAN layer, where each element is an internal function of the Lth KAN layer; No. Each inner function in the KAN layer The expression is as follows: ; ; ; Where siLU is the basic activation function and spline is the spline function. for The B-spline basis function is of order G, where G is the number of grid segments in the B-spline function. For learnable weights, This is the internal function in the i-th row and j-th column of the l-th KAN layer. For the first l The spline function used in the internal function of the i-th row and j-th column in the KAN layer. For the first l The coefficients of the spline function used in the internal function of the i-th row and j-th column in the KAN layer.
[0006] Optionally, K can be 6 and M can be 4.
[0007] Optionally, the high-frequency components include: the first three modal components arranged from high to low frequency and the first three quadratic decomposition residual components arranged from high to low frequency; The low- and mid-frequency components shown include the first three modal components arranged from low to high frequency and the first quadratic decomposition residual component arranged from low to high frequency.
[0008] Optionally, the training process of the first improved model includes: obtain a historical power sequence; the historical power sequence comprises a power sequence at a first time and a power sequence at a second time; the second time is a next time of the first time; train the first improved model with the power sequence at the first time as input and the power sequence at the second time as output, to obtain a trained first improved model.
[0009] Optionally, an output gate expression of the first improved model is ; wherein, indicates an output gate, KAN indicates a KAN layer, x t indicates an input value at t time, h t-1 indicates a hidden state value at t-1 time.
[0010] Optionally, an output gate expression of the second improved model is ; wherein, indicates an output value at t time, KAN indicates a KAN layer, indicates a hidden state value at t time.
[0011] In a second aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the short-term power consumption prediction method based on the secondary VMD-xLSTM-KAN model according to any one of the above.
[0012] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the short-term power consumption prediction method based on the secondary VMD-xLSTM-KAN model according to any one of the above.
[0013] In a fourth aspect, the present application provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the short-term power consumption prediction method based on the secondary VMD-xLSTM-KAN model according to any one of the above.
[0014] According to the embodiments provided in the present application, the following technical effects are disclosed: The application provides a short-term electricity consumption prediction method, device, medium and product based on a quadratic VMD-xLSTM-KAN model, which comprises the following steps: acquiring an original electricity sequence; decomposing the original electricity sequence by using a variational mode decomposition algorithm to obtain K mode components and 1 residual component; further extracting data feature information by performing secondary decomposition on the residual component by using the variational mode decomposition algorithm to obtain M secondary decomposition residual components; dividing the K mode components and the M secondary decomposition residual components into high-frequency components and medium-low-frequency components according to a preset threshold; predicting the medium-low-frequency components by using a trained first improved model to obtain a first prediction result; the first improved model is a deep learning model obtained by replacing a weight matrix in an output gate of an extended long short-term memory network with a KAN (Kolmogorov-Arnold Network) layer; predicting the high-frequency components by using a trained second improved model to obtain a second prediction result; the second improved model is a deep learning model obtained by replacing a fully connected layer of the extended long short-term memory network with the KAN layer; and superimposing and combining the first prediction result and the second prediction result to obtain a prediction result of total electricity data. Firstly, the electricity consumption time sequence is decomposed into mode components and residual components based on the variational mode decomposition algorithm, and the residual components are subjected to secondary variational mode decomposition; secondly, the xLSTM-KAN model is constructed for individual modeling and prediction of each decomposition component by combining the powerful nonlinear modeling capability of the KAN network and the long-term dependence capturing capability of the extended long short-term memory network, thereby improving the problems of limited information storage capacity and insensitivity to long-term dependence relationship of traditional models such as LSTM (Long-Short Term Memory), and two improved models are specifically proposed; the first improved model replaces the weight matrix used in the output gate of the original xLTSM (Extended Long Short-Term Memor) with a KAN layer, thereby improving the fitting modeling effect in the output gate and optimizing the overall prediction result of the model; the second improved model replaces the fully connected layer that maps the hidden state to the output in the original process with a KAN layer, thereby enhancing the output expression capability of the model and improving the overall performance of the model; and finally, the linear superposition of the prediction results of each component is performed to reconstruct the final prediction result of the total electricity consumption, which has higher prediction accuracy than other methods. BRIEF DESCRIPTION OF DRAWINGS
[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort.
[0016] Figure 1 An application environment diagram of a short-term electricity consumption prediction method based on a quadratic VMD-xLSTM-KAN model according to an embodiment of the present application.
[0017] Figure 2 A flowchart of a short-term electricity consumption prediction method based on a quadratic VMD-xLSTM-KAN model according to an embodiment of the present application.
[0018] Figure 3 A first improved model prediction flowchart according to an embodiment of the present application.
[0019] Figure 4 An xLSTM internal structure diagram according to an embodiment of the present application.
[0020] Figure 5 A second improved model prediction flowchart according to an embodiment of the present application.
[0021] Figure 6 A structure diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.
[0023] The present application proposes a quadratic VMD-xLSTM-KAN model to solve the problem of low prediction accuracy of short-term electricity consumption data. Experimental results prove that the prediction accuracy of the model is significantly improved compared to existing models. The present application proposes an xLSTM-KAN fusion network to improve the problems of limited information storage capacity and insensitivity to long-term dependency capture of traditional models such as LSTM. The network has stronger long-term dependency capture ability and short-term decision correction ability compared to the LSTM network, and has performance advantages. Since the nonlinear fitting effect and mathematical interpretability of the KAN network are better than those of traditional MLP, the present application further improves the xLSTM model by using its advantages and proposes two xLSTM-KAN improved models: The first improved model replaces the weight matrix used in the original xLTSM output gate with a KAN layer to improve the fitting modeling effect inside the output gate and optimize the overall prediction results of the model.
[0024] The second improved model uses a KAN layer instead of the fully connected layer that maps hidden states to outputs in the original process, enhancing the model's output expressiveness and improving the overall performance of the model.
[0025] The improved model theoretically addresses the shortcomings of the traditional model, and actual experimental results verify its performance advantages.
[0026] This application proposes a short-term electricity consumption prediction model based on quadratic VMD-xLSTM-KAN. First, the original electricity consumption sequence is decomposed using the variational mode decomposition (VMD) algorithm, and the residual components are decomposed using quadratic VMD. Second, combining the advantages of Kolmogorov-Arnold Network (KAN) and Extended Long Short-Term Memory Network (xLSTM), an xLSTM-KAN model is constructed to predict each component separately. Finally, the models are superimposed to reconstruct the final prediction result of the total electricity consumption.
[0027] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0028] This application provides a short-term electricity consumption prediction method based on a quadratic VMD-xLSTM-KAN model, which can be applied to, for example... Figure 1 The application environment shown is illustrated. Terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server.
[0029] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.
[0030] In one exemplary embodiment, such as Figure 2 As shown, a short-term electricity consumption prediction method based on a quadratic VMD-xLSTM-KAN model is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S1 to S7. Wherein: S1. Obtain the original electricity consumption sequence (or the original daily electricity consumption data sequence).
[0031] S2. The original electrical quantity sequence is decomposed using the variational mode decomposition algorithm to obtain K mode components and 1 residual component.
[0032] The number of modal components in the decomposition result is set based on experience. The original energy sequence is decomposed into a combination of simple modal components such as IMF1~IMF6 and residual component Res using the VMD algorithm to reduce the complexity of the energy data. Specific process: The Variational Mode Decomposition (VMD) algorithm constructs and solves a constrained variational problem. The original signal is adaptively decomposed into a series of intrinsic mode functions (IMFs) with specific center frequencies and finite bandwidths. This effectively extracts signal features and reduces the complexity of time series analysis; the algorithm can adjust the number of modal components. and penalty factor By adjusting parameters such as EMD, the decomposition results are effectively improved, thus addressing the issues of modal aliasing, endpoint effects, and uncontrollable modal numbers in decomposition algorithms.
[0033] S3. Using the variational mode decomposition algorithm, the residual components are decomposed in two stages to further extract data feature information and obtain M secondary decomposition residual components.
[0034] This embodiment considers that the prediction model performs poorly for the residual components, so it uses the VMD decomposition method to perform secondary decomposition on the residual components to further extract data feature information. In this embodiment, the number of secondary decomposition modal components is 3, resulting in 3 components and one new residual (named Res1-Res4), that is, 4 secondary decomposition residual components.
[0035] S4. According to a preset threshold, the K modal components and M secondary decomposition residual components are divided into high-frequency components and medium-low frequency components.
[0036] The high-frequency components include: the first three modal components arranged from high to low frequency and the first three quadratic decomposition residual components arranged from high to low frequency.
[0037] The low- and mid-frequency components shown include the first three modal components arranged from low to high frequency and the first quadratic decomposition residual component arranged from low to high frequency.
[0038] S5. Using the trained first improved model, predict the mid-to-low frequency components to obtain the first prediction result; the first improved model is a deep learning model obtained by replacing the weight matrix in the output gate of the extended long short-term memory network with a KAN layer.
[0039] In this embodiment, two improved models are used to predict the different components after decomposition; the first improved model is used to predict the mid- and low-frequency components (IMF1-IMF3, Res1); The second improved model was used to complete the prediction of the remaining high-frequency components (IMF4-IMF6, Res2-Res4).
[0040] This embodiment combines the superior long-term modeling capability of xLSTM compared to LSTM with the nonlinear modeling advantage of KAN compared to MLP, and constructs two improved xLSTM-KAN neural networks.
[0041] First improved model: The model's final prediction output From hidden state The result is obtained from the dimensionality reduction mapping, and according to the expression, The output gate Therefore, this embodiment replaces the weight matrix used in the original output gate with a KAN layer, constructing the first improved model. The output gate expression is derived from... Improved to , ( It is the Sigmoid activation function. W 0 This is the weight matrix of the fully connected layer with output gates. x t Let be the input value at time t. b 0 This is the bias vector for the fully connected layer with output gates. h t-1 (The hidden state value at time t-1) By improving the fitting modeling effect inside the output gate, the overall prediction result of the model is optimized.
[0042] Based on the above improvements, the prediction process of the model is as follows: after the input data is normalized, it is input into the first improved model (xLSTM-KAN improved model 1) to obtain the high-dimensional hidden state. The output is then processed using a fully connected layer (MLP). The data is mapped to one-dimensional data, and finally, after inverse normalization, the final prediction result is output. The overall prediction process is as follows: Figure 3 As shown.
[0043] Specifically, Extended Long Short-Term Memory (xLSTM) networks: The xLSTM approach overcomes these limitations. Two novel memory units are designed within xLSTM: sLSTM, which employs a hybrid scalar and memory mechanism, and mLSTM, which uses matrix memory and covariance updates and supports parallelization. The sLSTM unit makes the following improvements: 1) Exponential gating was introduced to replace the input gate. (Forgotten Gate) (Optional) Sigmoid gated function, whose larger output value range enhances the ability to capture new information and correct historical storage.
[0044] 2) A normalized state was designed. and steady state To stabilize the exponential gating, avoid numerical overflow and gradient explosion problems caused by excessively large exponential gating output values.
[0045] The expressions for each gate and state within the sLSTM cell are as follows: ; ; ; ; ; ; in The input gate is the sigmoid activation function, and exp is the exponential activation function. To control the storage of new information, the gating function is changed to exponential gating; forget gate The retention of historical information can be controlled, and exponential gating can be selected to adapt to different application scenarios; cell state. Used for storing long-term information, output gate The amount of control output information remains constant; normalized state Used for hiding state Cellular state Normalization is performed. This can be considered as the normalized cell state; W i This is the weight matrix of the fully connected layer in the input gate; h t-1 The hidden state value at time t-1; x t The input value at time t; b i The bias vector of the fully connected layer in the input gate; W fThis is the weight matrix of the fully connected layer in the forget gate; b f This is the bias vector of the fully connected layer in the forget gate; W 0 This is the weight matrix of the fully connected layer in the output gate; b 0 This is the bias vector of the fully connected layer in the output gate; c t-1 The cell state at time t-1; Z t As an intermediate variable, it represents the state of candidate cells, that is, the newly generated candidate information at the current moment; W z This is the weight matrix of the fully connected layer in the candidate cell state; b z is the bias vector of the fully connected layer in the candidate cell state; n t-1 t is the normalized state value at time t-1; tanh is a commonly used activation function that maps the input axis to (-1, 1).
[0046] Furthermore, the stable function The relevant calculation formulas are as follows: ; ; ; in , It is used to replace the original forget gate and input gate functions during forward propagation, thereby stabilizing the gated computation. m t-1 This represents the stable function value at time t-1. In this embodiment, all xLSTM neural networks use sLSTM as their internal unit modules. See the internal structure diagram of xLSTM. Figure 4 .
[0047] Specifically, the KAN network: KAN networks are a novel neural network architecture based on the Kolmogorov-Arnold representation theorem, which proves that for any multivariate continuous function defined on a bounded field... Each can be represented as a finite nested combination of several one-dimensional continuous functions, and its mathematical expression is as follows: ; in, for Dimensional input variables, inner functions external functions All are univariate continuous functions; x represents the input variable of the function; q represents the external summation index and the intermediate layer dimension; p represents the input variable dimension.
[0048] The original theorem lacks interpretability because univariate continuous functions cannot be directly learned. To address this deficiency, the KAN network introduces a B-spline function to achieve a smooth curve approximation of univariate functions, while also extending the two-layer structure of the original theorem to arbitrary depths, increasing its future development potential. For depths of... The overall structure of the KAN network is as follows: ; in, y The output of the KAN layer. x As the input to the KAN layer, This is the inter-layer mapping matrix of the (L-1)th KAN layer. This is the inter-layer mapping matrix of the (L-2)th KAN layer. This is the inter-layer mapping matrix of the 0th KAN layer (initial inter-layer mapping matrix). The expression for calculating the interlayer mapping of the Lth KAN layer is as follows: ; in, Let be the inter-layer mapping matrix of the Lth KAN layer, where each element is an internal function of the Lth KAN layer. The expression is shown below; No. Each inner function in the KAN layer The expression is as follows: ; ; ; Where siLU is the basic activation function and spline is the spline function. for The B-spline basis function is of order G, where G is the number of grid segments in the B-spline function. For learnable weights, This is the internal function in the i-th row and j-th column of the l-th KAN layer. For the first l The spline function used in the internal function of the i-th row and j-th column in the KAN layer. For the first l The coefficients of the spline function used in the internal function of the i-th row and j-th column in the KAN layer.
[0049] The training process of the first improved model includes: Obtain historical power consumption sequences; the historical power consumption sequences include: power consumption sequences at a first time point and power consumption sequences at a second time point; the second time point is the time point following the first time point.
[0050] Using the energy sequence at the first time step as input and the energy sequence at the second time step as output, the first improved model is trained to obtain the trained first improved model.
[0051] In practice, the time window width is set to 8. Therefore, the input is a series of continuous data segments of length 8, and the output is the model's prediction at the next time point after the input sequence. The label is the true value at the next time point after the input sequence, and the loss function is the mean squared error (MSE).
[0052] In actual prediction, the input is a continuous data window of length 8 in the test set, and the output is the model's prediction for the next time point (i.e. the first step in the future).
[0053] S6. Using the trained second improved model, predict the high-frequency components to obtain a second prediction result; the second improved model is a deep learning model obtained by replacing the fully connected layer of the extended long short-term memory network with a KAN layer.
[0054] The training process for the second improved model is the same as that for the first improved model.
[0055] Second improved model (xLSTM-KAN improved model 2): The original prediction process used a fully connected layer to hide the state. Dimensionality reduction mapping to output result The output expression is The second improved xLSTM-KAN model replaces the fully connected layer in the original process with a KAN layer, and the output expression is improved as follows: ;in, W yh To calculate the weight matrix of the fully connected layer when outputting y, b y Calculate the paralysis function of the fully connected layer when outputting y.
[0056] The overall prediction process of the second improved model is as follows: Normalized data is used as input to the xLSTM neural network, and the high-dimensional hidden state output of the xLSTM... The data is reduced to one dimension by a KAN layer, and the prediction result is finally obtained through an inverse normalization layer. The second improved model, leveraging the learnable activation function in the KAN layer, possesses stronger expressive power, thereby improving the overall prediction performance. The overall prediction flowchart of the model is as follows... Figure 5 As shown.
[0057] S7. The first prediction result and the second prediction result are superimposed and combined to obtain the prediction result of the total power consumption data.
[0058] Since the decomposition algorithm decomposes the original sequence into an additive combination of its components, the outputs of each component are additively superimposed to obtain the total predicted value.
[0059] Furthermore, the electricity consumption dataset used in this embodiment is daily electricity consumption data for the entire province of a certain province in southern my country, spanning from January 1, 2019 to July 31, 2024. To verify the advantages of the model proposed in this paper, this embodiment divides the above dataset into a training set and a test set in an 8:2 ratio.
[0060] This embodiment uses four error evaluation metrics—Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Symmetric Mean Absolute Percentage Error (SMAPE)—to assess model performance. The smaller the error calculated by each metric, the higher the model's prediction accuracy and the better its performance. The calculation formulas for each metric are as follows: ; ; ; ; Where n is the number of samples. Let be the predicted value for the i-th sample. y i Let be the true value of the i-th sample, and the rest are related mathematical calculations such as absolute value, summation, squaring, square root, etc.
[0061] Among the indicators, RMSE and MAE are in the same unit as the electricity consumption data in the dataset, which is kilowatt-hours, while MAPE and SMAPE are in percentage.
[0062] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a short-term electricity consumption prediction method based on a quadratic VMD-xLSTM-KAN model.
[0063] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0064] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0065] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0066] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0067] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0068] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0069] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0070] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0071] This embodiment uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application; at the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. In summary, the content of this specification should not be construed as a limitation of this application.
Claims
1. A short-term electricity consumption prediction method based on a quadratic VMD-xLSTM-KAN model, characterized in that, include: Obtain the original battery level sequence; The original electrical quantity sequence is decomposed using a variational mode decomposition algorithm to obtain K mode components and 1 residual component; The variational mode decomposition algorithm is used to perform secondary decomposition on the residual components, and further extract data feature information to obtain M secondary decomposition residual components. According to a preset threshold, the K modal components and M secondary decomposition residual components are divided into high-frequency components and medium-low frequency components. Using the trained first improved model, the mid-to-low frequency components are predicted to obtain a first prediction result; the first improved model is a deep learning model obtained by replacing the weight matrix in the output gate of the extended long short-term memory network with a KAN layer. The high-frequency components are predicted using the trained second improved model to obtain a second prediction result; the second improved model is a deep learning model obtained by replacing the fully connected layer of the extended long short-term memory network with a KAN layer. The first prediction result and the second prediction result are superimposed and combined to obtain the prediction result of the total power consumption data.
2. The short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model according to claim 1, characterized in that, The overall structure of the KAN layer includes: ； in, y The output of the KAN layer. x As the input to the KAN layer, This is the inter-layer mapping matrix of the (L-1)th KAN layer. This is the inter-layer mapping matrix of the (L-2)th KAN layer. This is the inter-layer mapping matrix of the 0th KAN layer; The expression for calculating the interlayer mapping of the Lth KAN layer is as follows: ； in, Let be the interlayer mapping matrix of the Lth KAN layer, where each element is an internal function of the Lth KAN layer; No. Each inner function in the KAN layer The expression is as follows: ； ； ； Where siLU is the basic activation function and spline is the spline function. for The B-spline basis function is of order G, where G is the number of grid segments in the B-spline function. For learnable weights, This is the internal function in the i-th row and j-th column of the l-th KAN layer. For the first l The spline function used in the internal function of the i-th row and j-th column in the KAN layer. For the first l The coefficients of the spline function used in the internal function of the i-th row and j-th column in the KAN layer.
3. The short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model according to claim 1, characterized in that, K takes the value of 6, and M takes the value of 4.
4. The short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model according to claim 3, characterized in that, The high-frequency components include: the first three modal components arranged from high to low frequency and the first three quadratic decomposition residual components arranged from high to low frequency. The low- and mid-frequency components shown include the first three modal components arranged from low to high frequency and the first quadratic decomposition residual component arranged from low to high frequency.
5. The short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model according to claim 1, characterized in that, The training process of the first improved model includes: Obtain historical power consumption sequences; the historical power consumption sequences include: power consumption sequences at a first time point and power consumption sequences at a second time point; the second time point is the time point following the first time point; Using the energy sequence at the first time step as input and the energy sequence at the second time step as output, the first improved model is trained to obtain the trained first improved model.
6. The short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model according to claim 1, characterized in that, The output gate expression of the first improved model is: ； in, Indicates the output gate, KAN indicates the KAN layer. x t This represents the input value at time t. h t-1 This represents the hidden state value at time t-1.
7. The short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model according to claim 1, characterized in that, The output gate expression of the second improved model is: ； in, This represents the output value at time t, and KAN represents the KAN layer. This represents the hidden state value at time t.
8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model as described in any one of claims 1-7.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model as described in any one of claims 1-7.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the short-term electricity consumption prediction method based on the quadratic VMD-xLSTM-KAN model as described in any one of claims 1-7.