Power Load Forecasting Method and System Based on VMD and MoE Systems

Abstract

The present invention relates to the technical field of electric load forecasting, and discloses an electric load forecasting method and system based on a VMD and MoE system. The method includes inputting original electric load data into a trained VMD decomposer, dynamically adjusting the modal number K value and the penalty factor value according to the electric load fluctuation level, and decomposing the data into K modal components. Then, inputting the modal components into the trained MoE system, allocating M groups of expert models and weights to the modal components through a gating network, and adjusting the number of expert models for high-frequency and low-frequency modal components according to the fluctuation level. Finally, selecting at least one group from the M groups of expert models, and obtaining the final electric load forecasting value by fusing the forecasting results through weighted summation based on the selected model and its weight. The accuracy and robustness of electric load forecasting are improved by this method.

Description

Technical Field

[0001] The present invention relates to the technical field of electric load forecasting, and particularly to an electric load forecasting method and system based on VMD and MoE system. Background Art

[0002] Electric load forecasting is the basis and key link in the planning, dispatching and operation of power systems. Accurate load forecasting can provide important support for the safe and stable operation, optimal dispatching, demand-side management and power market transactions of power systems, and is of great significance for improving the reliability, economy and intelligence level of power systems.

[0003] However, electric load is affected by various complex factors, such as meteorological conditions, economic activities, social events, holidays, etc., showing significant non-linear, non-stationary and volatile characteristics. Traditional load forecasting methods, such as statistical models (such as time series model ARIMA, regression model, etc.), are difficult to effectively capture the non-linear, non-stationary characteristics and volatile characteristics of load data, and the forecasting accuracy is poor under complex working conditions. Traditional machine learning models (such as support vector machine SVM, artificial neural network ANN, etc.), although having certain non-linear modeling capabilities, are prone to problems such as overfitting and insufficient generalization ability when dealing with high-dimensional, high-noise and highly volatile load data, and the model interpretability is poor. Signal decomposition methods (such as empirical mode decomposition EMD, ensemble empirical mode decomposition EEMD, etc.) combined with forecasting models: Although methods such as EMD / EEMD can decompose complex signals into multiple modal components, there are problems such as modal aliasing, end effect, and lack of strict mathematical theory support, the decomposition effect is unstable, and the subsequent forecasting models are still difficult to effectively utilize the decomposed information, and the forecasting accuracy improvement is limited.

[0004] Therefore, how to improve the accuracy and robustness of electric load forecasting is an urgent problem to be solved in this field. Summary of the Invention

[0005] The purpose of the present invention is to provide an electric load forecasting method, system, medium and electronic device based on VMD and MoE system, so as to at least partially solve the above problems.

[0006] According to one aspect of the present disclosure, an electric load forecasting method based on VMD and MoE system is proposed, including: Step S1, collecting original electric load data, where the original electric load data at least includes historical electric load data, meteorological data and calendar data.

[0007] Step S2, inputting the original electric load data into a trained variational mode decomposition VMD to decompose the original electric load data into K modal components, where K>1.

[0008] Step S3: Input the K modal components into the trained Mixture of Experts (MoE) system. The MoE system includes a gating network and multiple expert models. Herein, the gating network assigns M groups of expert models and the corresponding weight for each model to the modal components, where each group of expert models includes at least K expert models and M > 0.

[0009] Step S4: Select at least one group of expert models from the M groups of expert models, and based on the selected expert models, fuse the prediction results of multiple expert models to obtain the final power load prediction result.

[0010] Optionally, the method further includes adjusting the values of K and M based on the power load fluctuation level within a preset recent period. Specifically, the higher the fluctuation level, the larger the values of K and M; the lower the fluctuation level, the smaller the values of K and M.

[0011] Optionally, the method further includes that each group of expert models includes at least K expert models, and it includes dividing the K modal components into high-frequency modal components and low-frequency modal components. Specifically, the higher the power load fluctuation level, the more model numbers are assigned to the high-frequency modal components.

[0012] Optionally, the method further includes that the higher the power load fluctuation level, the larger the penalty factor value of the Variational Mode Decomposition (VMD).

[0013] Optionally, the method further includes that the step of fusing the prediction results of multiple expert models based on the selected expert models to obtain the final power load prediction result includes fusing the prediction results of multiple expert models by weighted summation based on the selected expert models and their corresponding weights to obtain the final power load prediction result.

[0014] According to one aspect of the present disclosure, a power load prediction system based on VMD and MoE system is proposed, including: a collection module, configured to collect original power load data, where the original power load data includes at least historical power load data, meteorological data, and calendar data.

[0015] A decomposition module, configured to input the original power load data into the trained VMD to decompose the original power load data into K modal components, where K > 1.

[0016] An assignment module, configured to input the K modal components into the trained MoE system. The MoE system includes a gating network and multiple expert models. Herein, the gating network assigns M groups of expert models and the corresponding weight for each model to the modal components, where each group of expert models includes at least K expert models and M > 0.

[0017] A prediction module, configured to select at least one group of expert models from the M groups of expert models, and based on the selected expert models, fuse the prediction results of multiple expert models to obtain a final power load prediction result.

[0018] Optionally, the system further includes an adjustment module, configured to adjust the K value and the M value based on the power load fluctuation level within a preset recent time period, where the higher the fluctuation level, the larger the K value and the M value, and the lower the fluctuation level, the smaller the K value and the M value.

[0019] Optionally, the system further includes an allocation module, further configured to divide the K modal components into high-frequency modal components and low-frequency modal components, where the higher the power load fluctuation level, the more model numbers are allocated to the high-frequency modal components.

[0020] The present disclosure also provides a computer-readable storage medium storing a computer program, including that the steps in the method described in any one of the above embodiments are executed when the computer program is run by a processor.

[0021] An embodiment of the present application also provides an electronic device, where the electronic device includes a memory and a processor, a computer program is stored in the memory, and the processor executes the steps in the method described in any one of the above embodiments by calling the computer program stored in the memory.

[0022] The present invention discloses a power load prediction method based on variational mode decomposition (VMD) and a mixture of experts (MoE), including inputting original power load data into a trained VMD decomposer, dynamically adjusting the modal number K value and the penalty factor α value according to the power load fluctuation level, and decomposing the data into K modal components. Then, input the modal components into the trained MoE system, allocate M groups of expert models and weights to the modal components through a gating network, and adjust the number of expert models for high-frequency and low-frequency modal components according to the fluctuation level. Finally, select at least one group from the M groups of expert models, and fuse the prediction results through weighted summation based on the selected models and their weights to obtain the final power load prediction value. The accuracy and robustness of power load prediction are improved by this method. Description of the Drawings

[0023] Figure 1 It is a schematic diagram of a power load prediction method based on the VMD and MoE systems provided by an embodiment of the present application.

[0024] Figure 2 It is a schematic diagram of a power load prediction system based on the VMD and MoE systems provided by an embodiment of the present application.

[0025] Figure 3Schematic diagram of the structure of the electronic device provided by the embodiment of the present application. Detailed implementation manners

[0026] The following further describes the detailed implementation manners of the present invention in conjunction with the accompanying drawings. It should be noted here that the description of these implementation manners is used to help understand the present invention, but does not constitute a limitation on the present invention. In addition, the technical features involved in the various implementation manners of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0027] It should be noted that in the present application, "first", "second", and various numerical numbers are used for distinction for the convenience of description, and do not limit the scope of the embodiments of the present application. For example, to distinguish different classification results, etc., rather than for describing a specific order or sequence. It should be understood that the objects described in this way can be interchanged under appropriate circumstances so as to be able to describe the solutions other than the embodiments of the present application.

[0028] Specifically, Figure 1 The specific implementation flowchart of a power load forecasting method based on the VMD and MoE systems provided by an embodiment of the present application is shown. Please refer to Figure 1 , and the specific steps are as follows: Step S1, collect the original power load data, and the original power load data at least includes historical power load data, meteorological data, and calendar data.

[0029] Specifically, use the power system data acquisition device to collect the historical power load data in the target area. These data are from each monitoring node in the power grid and are recorded in hours to ensure the continuity and integrity of the data.

[0030] At the same time, obtain the meteorological data from the database of the local meteorological department, including information such as temperature, humidity, wind speed, weather conditions (sunny, cloudy, rainy, etc.). The time stamps of the meteorological data are accurately matched with the power load data for subsequent analysis of the correlation between the two.

[0031] The calendar data covers information such as weekdays, weekends, holidays, etc. Since different date types have significant differences in power load, it is sorted and marked through the standard calendar library to provide a basis for analyzing the impact of different date types on power load.

[0032] Step S2, input the original power load data into the trained variational mode decomposition (VMD) to decompose the original power load data into K modal components, where K>1.

[0033] Specifically, the collected original power load data is input into the trained variational mode decomposition (VMD). Through iterative optimization of the variational problem, VMD decomposes the original power load data into K modal components. These modal components have different frequency characteristics, and from low frequency to high frequency, they respectively reflect the trend term, periodic term, and some fluctuation details in the power load. For example, the low-frequency modal component may represent the long-term electricity consumption trend, such as the load change caused by seasonal variations; the high-frequency modal component may reflect the rapid load fluctuations in the short term, such as the load change caused by sudden weather changes or special events.

[0034] VMD is a non-recursive and adaptive signal decomposition method, whose goal is to decompose the original signal into a set of intrinsic mode functions (IMFs) with specific bandwidths and ensure that the sum of the decomposed IMF components is equal to the original signal. VMD realizes signal decomposition by constructing and solving a constrained variational problem, and its variational problem can be expressed as: ; where, are the K decomposed IMF components, are the central frequencies of each IMF component, is the Dirac function, represents convolution, represents the partial derivative with respect to time, represents the square of the L2 norm, is the penalty factor, is the original power load data, is the imaginary unit. By using optimization algorithms such as the alternating direction method of multipliers (ADMM) to solve the above variational problem, the K decomposed IMF components can be obtained, which will not be elaborated in this embodiment.

[0035] Optionally, before inputting the original power load data into the trained variational mode decomposition (VMD), preprocessing of the original power load data is also included, including data cleaning (removing outliers and filling missing values), normalization processing (converting data with different dimensions to the same dimension range for subsequent processing), etc. For example, for missing meteorological data, it can be filled by interpolation methods; for abnormal mutation values in power load data, they can be corrected by combining historical data and adjacent data points.

[0036] Step S3, input the K modal components into the trained mixture of experts (MoE) system. The MoE system includes a gating network and multiple expert models. Among them, the gating network assigns M groups of expert models and the corresponding weights for each model to the modal components, where each group of expert models includes at least K expert models and M > 0.

[0037] Specifically, the Mixture of Experts (MoE) system consists of a gating network and multiple expert models. The MoE system contains multiple expert models, and each expert model is good at processing specific types of data or predicting the load at a specific time scale. For example, for high-frequency modal components, models sensitive to high-frequency signals, such as Support Vector Regression (SVR), Radial Basis Function Neural Network (RBFNN), etc., are used to capture the rapid fluctuation changes of the load. For low-frequency modal components, models sensitive to low-frequency signals, such as Long Short-Term Memory Network (LSTM), Gated Recurrent Unit Network (GRU), etc., are used to capture the long-term trends and seasonal changes of the load.

[0038] When K modal components are input into the MoE system, the restricted gating network starts to work. In this embodiment, the gating network dynamically assigns M groups of expert models to them according to information such as the frequency characteristics and energy distribution of the modal components, and determines the corresponding weights for each model. Each group of expert models contains at least K expert models. Different expert models can adopt different types of prediction models to adapt to the characteristics of different modal components. In addition, expert models with different time scales can be set for different prediction time scales. For example, short-term prediction expert models (predicting the next 1 - 24 hours), medium-term prediction expert models (predicting the next 1 - 7 days), etc. This embodiment does not make restrictions.

[0039] Optionally, the Gating Network adopts a three-layer fully connected neural network (MLP). The input layer inputs the feature vector, such as the IMF component features. There are multiple hidden layers, and the number of neurons can be set according to actual needs. The activation function can be ReLU. The output layer has the same number of neurons as the K value, and the activation function can be Softmax, outputting the recommended expert models and their corresponding weights.

[0040] Step S4, select at least one group of expert models from the M groups of expert models, and based on the selected expert models, fuse the prediction results of multiple expert models to obtain the final power load prediction result.

[0041] Specifically, select at least one group of expert models with better performance from the M groups of expert models. The selection basis can be indicators such as the prediction accuracy rate and mean square error of the model on historical data. For example, by comparing the prediction results of each group of expert models on the validation set data, select the group with the smallest mean square error.

[0042] Optionally, the selection-based expert model fuses the prediction results of multiple expert models to obtain the final power load prediction result, including fusing the prediction results of multiple expert models by weighted summation based on the selected expert model and its corresponding weight to obtain the final power load prediction result. Specifically, based on the selected expert model and its corresponding weight, the prediction results of multiple expert models are fused by weighted summation. Suppose a set of selected expert models are E1, E2,..., E6, the corresponding weights are w1, w2,..., w6, and the predicted values of each expert model for the power load are p1, p2,..., p6. Then the final power load prediction result P is: . In this way, the advantages of multiple expert models are integrated to obtain a more accurate final power load prediction result.

[0043] Optionally, based on the power load fluctuation level within a preset recent period, adjust the important parameter K value of VDM and value, and the penalty factor is used to balance data fidelity and the sparsity of modal components. The larger the value, the narrower the bandwidth of the decomposed modal components and the better the sparsity. Specifically, the preset recent period is the past week, and the fluctuation situation of the power load within this period is calculated. By calculating statistical quantities such as the standard deviation and range of the load data and combining the fluctuation range of historical data, the power load fluctuation level is determined. For example, the fluctuation level is divided into three levels: low, medium, and high. When the standard deviation is less than a certain threshold and the range is within a small range, it is determined as a low fluctuation level; when the standard deviation and range are in a medium range, it is a medium fluctuation level; when the standard deviation is greater than a large threshold and the range is large, it is a high fluctuation level.

[0044] Determine the modal number K value and value of VMD according to the power load fluctuation level in the past period of time. Among them, the higher the power load fluctuation level, the larger the penalty factor value corresponding to the variational mode decomposition VMD. At the low fluctuation level, is set to a relatively small value. For example, when the recent power load fluctuation level is high, a larger K value (such as K = 5) is selected to more finely decompose modal components of different frequencies and improve the algorithm robustness in the case of high power load fluctuations; when the fluctuation level is low, a smaller K value (such as K = 3) is selected. At the same time, adjust the penalty factor value of VMD according to the fluctuation level. The higher the fluctuation level, the larger the value. For example, when the fluctuation level is high, can be set to a relatively large value, such as 3000. The larger The value makes VMD emphasize more on capturing the details of the signal during decomposition, enabling it to better decompose different modal components under complex fluctuation conditions and providing richer and more accurate information for subsequent prediction. When the fluctuation level is low, it can be set to a relatively small value, such as 1000. At this time, VMD pays more attention to smoothness during decomposition to extract relatively stable modal components.

[0045] Optionally, based on the preset power load fluctuation level within a recent period, adjust the value of M. Specifically, adjust the value of M according to the power load fluctuation level. When the fluctuation level is low, reduce the value of M. For example, adjust M to 2. Because in the case of low fluctuations, the power load data is relatively stable, and there is no need for too many modal components and expert model groups for analysis and prediction to improve the efficiency of the system. When the fluctuation level is high, increase the value of M, such as increasing M to 4. This is because in the case of high fluctuations, the changes in the power load data are complex, and more modal components are needed to decompose the data. At the same time, more groups of expert models are needed to analyze and predict the data from different perspectives.

[0046] Optionally, divide the K modal components into high-frequency modal components and low-frequency modal components. Among them, the higher the power load fluctuation level, the more model numbers are allocated to the high-frequency modal components. Specifically, divide the K modal components into high-frequency modal components and low-frequency modal components. For the low fluctuation level, since the data is relatively stable, there are fewer high-frequency modal components, and the number of model numbers allocated to the high-frequency modal components is relatively small. For example, allocate 1 model to each high-frequency modal component in each group of expert models, and more expert models can be selected to be allocated to the low-frequency modal components to enhance the prediction ability for long-term trends. When the fluctuation level increases, such as in the case of high fluctuations, the power load changes violently, the high-frequency modal components increase, and the number of model numbers allocated to the high-frequency modal components increases accordingly. In each group of expert models, 2 - 3 models may be allocated to each high-frequency modal component to enhance the modeling ability for short-term fluctuations and better capture the load change information in the high-frequency part.

[0047] Through this embodiment, the original power load data is decomposed into modal components with different frequency characteristics by VMD, which can more meticulously display the trend items, periodic items, and fluctuation details in the power load, providing rich and accurate information for subsequent prediction. The gating network in MoE allocates multiple groups of expert models to different modal components, and each expert model adopts different prediction algorithms. Then, the prediction results are fused through weighted summation to give full play to the advantages of each model; at the same time, adjust the values of K, M, and the penalty factor according to the preset power load fluctuation level within a recent period . When the load fluctuation is low, reduce the number of modal components and expert model groups to reduce the computational complexity; when the fluctuation is high, increase the corresponding number and Values are adjusted to adapt to complex and changing data. In addition, different numbers of expert models are allocated to high-frequency and low-frequency modal components according to the load fluctuation level. When the fluctuation level is high, more models are allocated to the high-frequency modal components to better capture rapid load fluctuation information, and vice versa when the fluctuation level is low. Through the above solutions, the accuracy and robustness of power load forecasting are improved.

[0048] A power load forecasting method based on VMD and MoE system corresponding to the above embodiment Figure 2 The structural block diagram of a power load forecasting system based on VMD and MoE system provided by an embodiment of the present application is shown. For ease of description, only parts related to the embodiment of the present application are shown.

[0049] See Figure 2 As shown, a power load forecasting system 200 based on VMD and MoE system provided by an embodiment of the present application includes: a collection module for collecting original power load data, where the original power load data at least includes historical power load data, meteorological data, and calendar data.

[0050] A decomposition module for inputting the original power load data into a trained variational mode decomposition (VMD) to decompose the original power load data into K modal components, where K > 1.

[0051] An allocation module for inputting the K modal components into a trained mixture of experts (MoE) system. The MoE system includes a gating network and multiple expert models. Among them, the gating network allocates M groups of expert models and the corresponding weights for each model to the modal components. Each group of expert models includes at least K expert models, where M > 0.

[0052] A prediction module for selecting at least one group of expert models from the M groups of expert models, and based on the selected expert models, fusing the prediction results of multiple expert models to obtain the final power load forecasting result.

[0053] Optionally, the system further includes an adjustment module for adjusting the values of K and M based on the preset power load fluctuation level within a recent period. Among them, the higher the fluctuation level, the larger the values of K and M, and the lower the fluctuation level, the smaller the values of K and M.

[0054] Optionally, the system further includes an allocation module, which is further configured to divide the K modal components into high-frequency modal components and low-frequency modal components. Among them, the higher the power load fluctuation level, the more the number of models allocated to the high-frequency modal components.

[0055] Correspondingly, an embodiment of the present application further provides an electronic device, which can be a mobile terminal or a server. As Figure 3 shownFigure 3 This is a schematic structural diagram of the electronic device provided by the embodiment of the present application. For the convenience of description, only the parts related to the embodiment of the present application are shown.

[0056] The electronic device 300 includes a processor 301 having one or more processing cores, a memory 302 having one or more computer-readable storage media, and a computer program stored on the memory 302 and executable on the processor. Among them, the processor 301 is electrically connected to the memory 302. Those skilled in the art can understand that the structural diagram of the electronic device shown in the figure does not constitute a limitation on the electronic device, and it may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0057] The processor 301 is the control center of the electronic device 300, and connects various parts of the entire electronic device 300 through various interfaces and lines. By running or loading the software program (computer program) and / or unit stored in the memory 302, and calling the data stored in the memory 302, it executes various functions of the electronic device 300 and processes data, thereby monitoring the entire electronic device 300.

[0058] In the embodiment of the present application, the processor 301 in the electronic device 300 will load the instructions corresponding to the processes of one or more application programs into the memory 302 according to the following steps, and the processor 301 will run the application programs stored in the memory 302 to implement various functions: Step S1, collect the original power load data, and the original power load data at least includes historical power load data, meteorological data, and calendar data.

[0059] Step S2, input the original power load data into the trained variational mode decomposition (VMD) to decompose the original power load data into K modal components, where K>1.

[0060] Step S3, input the K modal components into the trained mixture of experts (MoE), and the mixture of experts (MoE) includes a gating network and multiple expert models. Among them, the gating network assigns M groups of expert models and the corresponding weights for each model to the modal components, where each group of expert models includes at least K expert models, and M>0.

[0061] Step S4, select at least one group of expert models from the M groups of expert models, and based on the selected expert models, fuse the prediction results of multiple expert models to obtain the final power load prediction result.

[0062] For the specific implementation of each of the above operations, reference can be made to the foregoing embodiments, and details will not be described herein again.

[0063] Optionally, asFigure 3 As shown, the electronic device 300 further includes: a prediction unit 303, a communication unit 304, an input unit 305, and a power supply 306. Among them, the processor 301 is electrically connected to the prediction unit 303, the communication unit 304, the input unit 305, and the power supply 306 respectively. Those skilled in the art can understand that Figure 3 the structure of the electronic device shown in does not constitute a limitation on the electronic device, and it may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements.

[0064] The prediction unit 303 can be used for power load prediction based on the VMD and MoE systems.

[0065] The communication unit 304 can be used for communicating with other devices.

[0066] The input unit 305 can be used to receive input digital, character information, or user feature information (such as fingerprints, irises, facial information, etc.), and generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function controls.

[0067] The power supply 306 is used to supply power to each component of the electronic device 300. Optionally, the power supply 306 can be logically connected to the processor 301 through a power management system, so as to implement functions such as management of charging, discharging, and power consumption management through the power management system. The power supply 306 can also include any components such as one or more DC or AC power supplies, a recharge system, a power failure detection circuit, a power converter or inverter, and a power status indicator.

[0068] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0069] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructions, or by controlling relevant hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0070] For this reason, an embodiment of the present application provides a computer-readable storage medium, in which multiple computer programs are stored. These computer programs can be loaded by a processor to execute the steps of a power load prediction method based on the VMD and MoE systems provided by the embodiments of the present application.

[0071] For the specific implementation of each of the above operations, reference can be made to the previous embodiments, and details will not be repeated here.

[0072] Among them, the computer-readable storage medium may include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk, optical disk, etc.

[0073] Since the computer program stored in the storage medium can execute the steps in any of the power load prediction methods based on the VMD and MoE systems provided in the embodiments of the present application, the beneficial effects of any of the power load prediction methods based on the VMD and MoE systems provided in the embodiments of the present application can be achieved. For details, see the previous embodiments and will not be elaborated here.

[0074] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0075] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0076] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate computer-implemented processing, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0077] The above has described the embodiments of the present invention in detail in conjunction with the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, without departing from the principles and spirit of the present invention, various changes, modifications, substitutions, and variations made to these embodiments still fall within the protection scope of the present invention.

Claims

1. A power load forecasting method based on the VMD and MoE systems, characterized in that, Including: Step S1: Collect the original power load data, where the original power load data at least includes historical power load data, meteorological data, and calendar data; Step S2: Input the original power load data into the trained variational mode decomposition (VMD) to decompose the original power load data into K modal components, where K > 1; Step S3: Input the K modal components into the trained mixture of experts (MoE) system. The MoE system includes a gating network and multiple expert models. Among them, the gating network assigns M groups of expert models and the corresponding weights for each model to the modal components. Each group of expert models includes at least K expert models, where M > 0. Different time-scale expert models are set for different prediction time scales; Step S4: Select at least one group of expert models from the M groups of expert models, and based on the selected expert models, fuse the prediction results of multiple expert models to obtain the final power load prediction result; Based on the preset power load fluctuation level within a recent period, adjust the values of K and M. Among them, the higher the fluctuation level, the larger the values of K and M, and the lower the fluctuation level, the smaller the values of K and M; Each group of expert models includes at least K expert models, including: dividing the K modal components into high-frequency modal components and low-frequency modal components. Among them, the higher the power load fluctuation level, the more model numbers are assigned to the high-frequency modal components; The higher the power load fluctuation level, the larger the penalty factor α value corresponding to the variational mode decomposition (VMD); 2. The power load forecasting method based on the VMD and MoE systems according to claim 1, wherein The step of fusing the prediction results of multiple expert models based on the selected expert models to obtain the final power load prediction result includes: based on the selected expert models and their corresponding weights, fuse the prediction results of multiple expert models by weighted summation to obtain the final power load prediction result; 3. A power load forecasting system based on the VMD and MoE systems, characterized in that, Including: A collection module for collecting the original power load data, where the original power load data at least includes historical power load data, meteorological data, and calendar data; A decomposition module for inputting the original power load data into the trained variational mode decomposition (VMD) to decompose the original power load data into K modal components, where K > 1; An assignment module for inputting the K modal components into the trained mixture of experts (MoE) system. The MoE system includes a gating network and multiple expert models. Among them, the gating network assigns M groups of expert models and the corresponding weights for each model to the modal components. Each group of expert models includes at least K expert models, where M > 0; A prediction module for selecting at least one group of expert models from the M groups of expert models, and based on the selected expert models, fusing the prediction results of multiple expert models to obtain the final power load prediction result; A first adjustment module for adjusting the values of K and M based on the preset power load fluctuation level within a recent period. Among them, the higher the fluctuation level, the larger the values of K and M, and the lower the fluctuation level, the smaller the values of K and M; The allocation module is further configured to divide the K modal components into high-frequency modal components and low-frequency modal components, wherein the higher the power load fluctuation level is, the more model numbers are allocated to the high-frequency modal components; The second adjustment module is configured such that the higher the power load fluctuation level is, the larger the penalty factor α value corresponding to the variational mode decomposition (VMD) is.

4. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is run by a processor, it executes the method according to any one of claims 1-2.

5. An electronic device, characterized in that: It includes a memory storing executable program code and a processor coupled to the memory; wherein, the processor calls the executable program code stored in the memory and executes the method according to any one of claims 1-2.

Images