Power load prediction method and system based on multi-modal decomposition

Abstract

This invention belongs to the technical field of load forecasting. To address the problems of low accuracy and poor interpretability in existing power load forecasting methods, it proposes a power load forecasting method and system based on multimodal decomposition. The method involves multimodal decomposition of power load data and corresponding temperature data to obtain a base load sequence, a temperature-sensitive sequence, a periodic sequence, and a residual sequence. These sequences are then input into their respective prediction models to obtain prediction results. Finally, the prediction results are superimposed to obtain the final power load forecast. This invention employs a targeted extraction method that sequentially extracts components from the power load sequence and corresponding temperature data, avoiding the mode aliasing problem present in traditional decomposition methods. This allows for the separate prediction of power load influencing factors, enhancing interpretability and reliability.

Description

Technical Field

[0001] This invention belongs to the technical field of load forecasting, and particularly relates to a power load forecasting method and system based on multimodal decomposition. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Current power load forecasting technologies can be broadly categorized into three types: statistical model-based methods, machine learning-based methods, and deep learning-based methods. Statistical model-based methods, particularly in time series forecasting, commonly utilize models such as autoregressive moving average (ARMA), exponential smoothing (ESR), and vector autoregressive (VAR). These methods establish a power load time series and input it into the aforementioned models for time-series forecasting, ultimately yielding the predicted power load. However, these methods often rely on linear assumptions and struggle to handle nonlinear factors. Machine learning-based methods, particularly in regression problems, commonly employ models such as Lasso regression, random forests, gradient boosting decision trees (XGBoost, LightGBM), and support vector regression. These methods treat power load forecasting as a regression problem, using the aforementioned models to perform regression and obtain the predicted power load. They have made significant progress in handling nonlinear relationships, but their ability to handle complex, high-dimensional data is limited, and their generalization ability is weak. Deep learning-based methods, in the field of time series forecasting, commonly utilize models such as RNN, LSTM, TCN, and Transformer. These methods excel at handling high-dimensional and complex data, exhibiting strong adaptability and generalization capabilities, but their parameter tuning is complex and their interpretability is weak. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this invention provides a power load forecasting method and system based on multimodal decomposition. The method decomposes the power load sequence and the corresponding temperature data into modal components with clear physical meanings, and then forecasts each component separately. This enables individual forecasting of factors affecting power load, has physical verifiability, and enhances the interpretability and credibility of the method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a power load forecasting method based on multimodal decomposition, comprising: Acquire power load data and corresponding temperature data; The power load data and the corresponding temperature data are subjected to multimodal decomposition to obtain the base load sequence, temperature sensitive sequence, periodic sequence and residual sequence; The base charge sequence, the temperature-sensitive sequence, the periodic sequence, and the residual sequence are respectively input into the corresponding prediction models to obtain the corresponding prediction results; The power load forecast is obtained by superimposing the various forecast results.

[0006] Secondly, the present invention provides a power load forecasting system based on multimodal decomposition, comprising: The acquisition module is configured to acquire power load data and corresponding temperature data. The multimodal decomposition module is configured to perform multimodal decomposition on the power load data and the corresponding temperature data to obtain a base load sequence, a temperature-sensitive sequence, a periodic sequence, and a residual sequence. The preliminary prediction module is configured to input the base charge sequence, the temperature-sensitive sequence, the periodic sequence, and the residual sequence into the corresponding prediction models to obtain the corresponding prediction results. The prediction module is configured to superimpose the various prediction results to obtain the power load prediction result.

[0007] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, complete the method steps of the first aspect.

[0008] Thirdly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the method steps of the first aspect.

[0009] Fourthly, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the method steps of the first aspect.

[0010] The above one or more technical solutions have the following beneficial effects: In this invention, the power load sequence and corresponding temperature data are decomposed into physically meaningful base load components, temperature-sensitive components, periodic components, and residual components, avoiding the mode aliasing problem in traditional decomposition methods and realizing physical meaning-based mode decomposition of power load sequences. The trained prediction model is used to predict each mode component separately, and the prediction results of all components are superimposed to obtain the accurate prediction result of power load. This enables individual prediction of power load influencing factors, has physical verifiability, avoids the complete black box of deep learning methods, and enhances the interpretability and credibility of the method.

[0011] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0013] Figure 1 This is a flowchart of the power load forecasting method in an embodiment of the present invention. Detailed Implementation

[0014] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0015] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0016] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0017] Example 1 This embodiment discloses a power load forecasting method based on multimodal decomposition, including: Acquire power load data and corresponding temperature data; Multimodal decomposition was performed on the power load data and the corresponding temperature data to obtain the base load sequence, temperature-sensitive sequence, periodic sequence and residual sequence; The base charge sequence, temperature-sensitive sequence, periodic sequence, and residual sequence are input into the corresponding prediction models to obtain the corresponding prediction results. The power load forecast is obtained by superimposing the various forecast results.

[0018] This embodiment decomposes the power load sequence and corresponding temperature data into physically meaningful base load components, temperature-sensitive components, periodic components, and residual components. A trained prediction model is used to predict each modal component separately, and the prediction results of all components are superimposed to obtain an accurate prediction of the power load. A targeted extraction method sequentially extracts each component from the power load sequence and corresponding temperature data, integrating physical mechanisms with data-driven approaches. This avoids the modal aliasing problem present in traditional decomposition methods, achieving physically meaningful modal decomposition of the power load sequence. This embodiment's method of decomposing the power load sequence into physically meaningful modal components and then predicting each component separately enables individual prediction of factors influencing the power load, possesses physical verifiability, avoids the complete black box nature of deep learning methods, and enhances the interpretability and credibility of the method.

[0019] The following is combined Figure 1 The power load forecasting method based on multimodal decomposition proposed in this embodiment will be described in detail below: S1: Obtain power load data and corresponding temperature data.

[0020] The data to be acquired includes power load data of a power system over a period of time, which should be recorded at least hourly, with the load value in kilowatts; in addition to the power load data, temperature data of the same sampling frequency for the same period in the region where the power system is located is also required.

[0021] S2: Perform multimodal decomposition on the power load data and the corresponding temperature data to obtain the base load sequence, temperature sensitive sequence, periodic sequence and residual sequence.

[0022] Multimodal decomposition of power load data yields four modal components: base load sequence, temperature-sensitive sequence, periodic sequence, and residual sequence.

[0023] The multimodal decomposition of power load data includes the following steps: S21: Assume a discrete sequence of power loads ,in, N For the total sampling time points, It is the power load at the nth time point, with a sampling interval of . .

[0024] The base charge is the lowest frequency and slowest changing component in the load signal. An adaptive bandwidth Gaussian filter is used to extract the base charge sequence, and the calculation formula is as follows:

[0025]

[0026] Among them, the base charge sequence , For load The base charge component; The power load value after boundary processing of its index, i.e., the subscript; The standard deviation of the Gaussian kernel; Basic bandwidth; The curvature coefficient; The standard deviation of the global load; It is the first Electricity load at a specific point in time; It is the first Electricity load at a specific point in time; The sampling interval; k The integer index offset represents the positional offset of each data point within the sliding window relative to the center point during Gaussian filtering, used to traverse all sampling points in the neighborhood of the center point; This is a boundary handling function that performs mirror-symmetric processing on indices that exceed the boundary. Its calculation formula is as follows:

[0027] Where N represents the total number of sampling time points. This is the original index.

[0028] S22: From the discrete sequence of power load Extracting the base charge sequence Then, the first residual was obtained. The calculation formula is as follows:

[0029] in, For electrical load The base charge component; It is the power load at the nth time point.

[0030] S23: Set temperature data sequence .

[0031] Temperature-sensitive components are the temperature-affected parts of the load signal. They can be extracted by minimizing an objective function that simultaneously considers the modal characteristics and the temperature response.

[0032] This problem can be solved using the gradient descent method with alternating optimization.

[0033] in, For electrical load Temperature-sensitive components, temperature-sensitive sequences , For the optimal balance parameters of grid search, For smoothing terms, The formula for calculating the fit term is:

[0034]

[0035] in, For smoothing constraint coefficients; These are the regression coefficients; For electrical load Temperature-sensitive components; For the first Temperature data at various time points; For the first Temperature data at 10 time points; N is the total number of sampling time points.

[0036] S24: From the power load sequence After extracting the temperature-sensitive sequence T, the second residual is obtained. The calculation formula is as follows:

[0037] in, For electrical load Temperature-sensitive components; This is the first residual.

[0038] S25: For the second residual Performing the Discrete Fourier Transform:

[0039] in, The complex amplitude of the m-th frequency component.

[0040]

[0041] in, The amplitude of the m-th frequency component; Represents the real part of a complex number; It represents the imaginary part of a complex number.

[0042] S26: Select the k most significant frequency components from the amplitude values. The selection of k is based on the minimum number of frequency components that account for 90% of the energy. The calculation formula is as follows:

[0043] in, Amplitude values ​​arranged in descending order , It is the set of positive integers. This is an iterative variable used to test different component quantities until the energy percentage condition is met.

[0044] S27: Reconstruct the signal using only significant frequency components:

[0045] in, The number of significant frequency components obtained in S26 The physical frequency corresponding to the j-th significant component. The phase angle, For electrical load Periodic components, periodic sequences .

[0046] S28: From discrete power load sequences Extracting periodic sequences Then, the residual term is obtained. The calculation formula is as follows:

[0047] Among them, residual sequence .

[0048] S29: At this point, the total electrical load... It has been decomposed into a linear superposition of four modal components, and the calculation formula is:

[0049] in, This is the original electrical load; The base load component represents a long-term, gradual trend of change. This is a temperature-sensitive component, the portion of the load directly driven by the temperature of the day or recent days; As periodic components, they reflect the patterns of human social activities, such as daily and weekly cycles; This is the residual component.

[0050] S3: Input the base charge sequence, temperature-sensitive sequence, periodic sequence, and residual sequence into the corresponding prediction model to obtain the corresponding prediction results.

[0051] In this embodiment, historical power load data and temperature data from the same period are used to perform multimodal decomposition. The four modal components are used to train prediction models. Hyperparameters such as the number of hidden layer neurons and the learning rate are optimized using an improved particle swarm optimization algorithm to obtain the prediction results for each component.

[0052] This embodiment uses a Long Short-Term Memory (LSTM) network model as the prediction model, and optimizes the hyperparameters of the LSM network model using an improved particle swarm optimization algorithm, specifically including the following steps: S31: The sequence data of the four modal components are divided into training and test sets using a sliding window strategy. The training set accounts for 70% of the dataset. Then, the training set sample data is normalized, and the normalization parameters of the training set are used to normalize the test set. Finally, the corresponding training and test sets are constructed for the base charge sequence, temperature sensitive sequence, periodic sequence and residual sequence.

[0053] S32: Train a separate LSTM model for each of the four sequence data mentioned above.

[0054] An improved particle swarm optimization algorithm is used to search and determine hyperparameters such as time_steps, number of hidden layer neurons, number of hidden layers, learning rate, number of model iterations (epochs), training batch size, and dropout probability, in order to optimize the model's prediction performance as much as possible.

[0055] The Huber loss function is chosen as the loss function during model training, as it combines the advantages of both MSE and MAE loss functions. The calculation formula is as follows:

[0056] in, y For the true value, p(x) These are the model's predicted values. This is an adjustable parameter of Huber loss, which determines whether it is more biased towards MSE or MAE.

[0057] S33: An improved particle swarm optimization algorithm used during training. The improvement lies in the particle velocity update formula, which is calculated as follows:

[0058] In this context, the position of each particle represents a set of hyperparameters for the LSTM. Let be the position vector of the i-th particle in iteration t. For the first i The velocity vector of a particle in t iterations The optimal position for the entire group. For inertial weights, As a learning factor, A random number uniformly distributed in the interval [0,1]. The loss function for the prediction model is as follows: The gradient of the loss function can be solved using the finite difference method. Taking a single hyperparameter θ as an example, the approximate formula for calculating the gradient is:

[0059] in For a very small positive number (e.g., 10) -5 For multiple hyperparameters, the above calculations can be performed separately for each dimension.

[0060] The calculation formula is:

[0061]

[0062] Where T is the maximum number of iterations; t is the current number of iterations; Let i be the optimal position for the i-th particle. This is the balance coefficient, used in the early stages of iteration. It tends to focus on individual cognition and conducts a broad, dispersed search. However, in the later stages of iteration, it tends to focus on group cognition, concentrate on the optimal region, and accelerate convergence.

[0063] The calculation formula is:

[0064]

[0065] Among them, the first item To achieve distance awareness, the gradient effect is reduced when the solution is far from the optimal solution (to avoid misleading); the gradient effect is enhanced when the solution is close to the optimal solution (for finer-grained search). (Second term) For exponential decay, κ This is the attenuation coefficient, used to prevent gradient noise from interfering with convergence later. (Third term) For directional confidence, Here, is the confidence coefficient, and sign is the sign function. It is the increment of the particle's displacement in the search space, that is, the particle's velocity. The norm is used to indicate the direction of the objective function gradient. When the gradient of the objective function is in the same direction as the particle velocity (i.e., the dot product is greater than 0), the confidence level is greater than 1; when the directions are not in the same direction, the confidence level is less than 1.

[0066] This method has the following advantages: 1. Accelerated convergence: Gradient information directly points to the descent direction of the function, which can find the local optimum more quickly.

[0067] 2. Improved accuracy: With reliable gradient information, the optimal solution can be located more accurately.

[0068] 3. Avoid premature convergence: In the early stages of iteration, the gradient term can help particles break free from the attraction of individual optima and global optima, explore new regions, and thus potentially avoid getting trapped in local optima.

[0069] S4: Superimpose the obtained prediction results to obtain the power load prediction result.

[0070] After training the LSTM model for each modal component, it is used to predict the four modal components to be predicted. The results are then inversely normalized to obtain the basis charge sequence. Temperature-sensitive sequence Periodic sequences and residual sequence The predicted values ​​of the four modal components are linearly added together to obtain the final power load forecast result, as shown in the formula:

[0071] The multimodal decomposition method proposed in this embodiment decomposes the power load sequence into a physically meaningful base load component, a temperature-sensitive component, a periodic component, and a residual component. Then, a separate model is trained for each modal component to learn the temporal characteristics of that component. Finally, the trained model is used to predict each modal component, and the prediction results of all components are superimposed to obtain an accurate prediction result of the power load.

[0072] This embodiment employs a targeted extraction method to sequentially extract each component from the power load sequence, integrating physical mechanisms with data-driven approaches. This avoids the mode aliasing problem present in traditional decomposition methods and achieves physical-meaning-based mode decomposition of the power load sequence.

[0073] In this embodiment, separate LSTM models are trained for each modal component with obvious temporal characteristics. The hyperparameter selection of the LSTM model is optimized by using an improved particle swarm optimization algorithm. Compared with directly predicting power load data containing complex nonlinear relationships, this can effectively reduce prediction errors and improve prediction accuracy.

[0074] In this embodiment, separate prediction models are trained for each modal component with obvious time-series characteristics. The hyperparameter selection of the prediction model is optimized by using an improved particle swarm optimization algorithm. Compared with directly predicting power load data containing complex nonlinear relationships, this can effectively reduce prediction errors and improve prediction accuracy.

[0075] This embodiment decomposes the power load sequence into modal components with clear physical meanings, and then predicts each component separately. This method enables individual prediction of factors affecting power load, has physical verifiability, avoids the complete black box nature of deep learning methods, and enhances the interpretability and credibility of the method.

[0076] Example 2 The purpose of this embodiment is to provide a power load forecasting system based on multimodal decomposition, including: The acquisition module is configured to acquire power load data and corresponding temperature data. The multimodal decomposition module is configured to perform multimodal decomposition on the power load data and the corresponding temperature data to obtain a base load sequence, a temperature-sensitive sequence, a periodic sequence, and a residual sequence. The preliminary prediction module is configured to input the base charge sequence, the temperature-sensitive sequence, the periodic sequence, and the residual sequence into the corresponding prediction models to obtain the corresponding prediction results. The prediction module is configured to superimpose the various prediction results to obtain the power load prediction result.

[0077] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0078] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0079] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0080] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0081] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0082] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0083] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0084] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0085] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0086] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0087] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for power load forecasting based on multi-modal decomposition, characterized in that, include: Acquire power load data and corresponding temperature data; The power load data and corresponding temperature data are subjected to multimodal decomposition to obtain the base load sequence, temperature-sensitive sequence, periodic sequence, and residual sequence, specifically: Based on the aforementioned power load data, an adaptive bandwidth Gaussian filter is used to extract the base load sequence. The calculation formula is as follows: Among them, the base charge sequence , For load The base charge component; The power load value after boundary processing of its index, i.e., the subscript; The standard deviation of the Gaussian kernel; Basic bandwidth; The curvature coefficient; The standard deviation of the global load; It is the first Electricity load at a specific point in time; It is the first Electricity load at a specific point in time; The sampling interval; q The integer index offset represents the positional offset of each data point within the sliding window relative to the center point during Gaussian filtering, used to traverse all sampling points in the neighborhood of the center point; This is a boundary handling function that performs mirror-symmetric processing on indices that exceed the boundary. Based on the temperature data, a temperature-sensitive sequence is extracted by minimizing an objective function that considers both the modal characteristics and the temperature response. in, For electrical load Temperature-sensitive components, temperature-sensitive sequences , For the optimal balance parameters of grid search, For smoothing terms, The formula for calculating the fit term is: in, For smoothing constraint coefficients; These are the regression coefficients; For electrical load Temperature-sensitive components; For the first Temperature data at various time points; For the first Temperature data at 1 time point; N is the total number of sampling time points; The first residual is obtained from the extracted base charge sequence, and the calculation formula is as follows: The second residual is obtained based on the first residual and the temperature-sensitive sequence, and the calculation formula is as follows: Perform a Fourier transform on the second residual to obtain the amplitude of the frequency components; select the k most significant frequency components by amplitude value sorting, where k is a positive integer, and the selection of k is based on the minimum number of frequency components that satisfy the requirement of having an energy share of 90%, calculated using the following formula: in, Amplitude values ​​arranged in descending order , Let m be the amplitude of the m-th frequency component. It is the set of positive integers. This is used as an iterative variable to test different component quantities until the energy percentage condition is met; By reconstructing the signal using significant frequency components, a periodic sequence is obtained, specifically: in, The number of significant frequency components, The physical frequency corresponding to the j-th significant component. The phase angle, For electrical load Periodic components, periodic sequences ; The residual sequence is obtained based on the periodic sequence, specifically as follows: Among them, residual sequence ; The base charge sequence, the temperature-sensitive sequence, the periodic sequence, and the residual sequence are respectively input into the corresponding prediction models to obtain the corresponding prediction results; The power load forecast is obtained by superimposing the various forecast results.

2. The power load forecasting method based on multimodal decomposition as described in claim 1, characterized in that, An improved particle swarm optimization (PSO) algorithm is used to train the prediction model. Specifically, the improved PSO algorithm is as follows: The position of each particle represents a set of hyperparameters for the prediction model; For the first i Individual particles The velocity vector of the next iteration For the first i The velocity vector of a particle in t iterations For inertial weights, As a learning factor, A random number uniformly distributed in the interval [0,1]. The optimal position for the entire group. Represents the norm, Let be the position vector of the i-th particle in iteration t. The gradient of the objective function. For balance coefficient, Let be the optimal position for the i-th particle, T be the maximum number of iterations, and t be the current number of iterations. Here, is the confidence coefficient, and sign is the sign function; It is the displacement increment of the particle in the search space; κ This is the attenuation coefficient.

3. The power load forecasting method based on multimodal decomposition as described in any one of claims 1-2, characterized in that, The prediction model is trained using the Huber loss function as the loss function.

4. The power load forecasting method based on multimodal decomposition as described in claim 1, characterized in that, The prediction results output by each prediction model are back-normalized to obtain the corresponding prediction values. The prediction values ​​are then linearly added together to obtain the power load prediction result.

5. A power load forecasting system based on multimodal decomposition, characterized in that, include: The acquisition module is configured to acquire power load data and corresponding temperature data. The multimodal decomposition module is configured to perform multimodal decomposition on the power load data and the corresponding temperature data to obtain a base load sequence, a temperature-sensitive sequence, a periodic sequence, and a residual sequence, specifically: Based on the aforementioned power load data, an adaptive bandwidth Gaussian filter is used to extract the base load sequence. The calculation formula is as follows: Among them, the base charge sequence , For load The base charge component; The power load value after boundary processing of its index, i.e., the subscript; The standard deviation of the Gaussian kernel; Basic bandwidth; The curvature coefficient; The standard deviation of the global load; It is the first Electricity load at a specific point in time; It is the first Electricity load at a specific point in time; The sampling interval; q The integer index offset represents the positional offset of each data point within the sliding window relative to the center point during Gaussian filtering, used to traverse all sampling points in the neighborhood of the center point; This is a boundary handling function that performs mirror-symmetric processing on indices that exceed the boundary. Based on the temperature data, a temperature-sensitive sequence is extracted by minimizing an objective function that considers both the modal characteristics and the temperature response. in, For electrical load Temperature-sensitive components, temperature-sensitive sequences , For the optimal balance parameters of grid search, For smoothing terms, The formula for calculating the fit term is: in, For smoothing constraint coefficients; These are the regression coefficients; For electrical load Temperature-sensitive components; For the first Temperature data at various time points; For the first Temperature data at 1 time point; N is the total number of sampling time points; The first residual is obtained from the extracted base charge sequence, and the calculation formula is as follows: The second residual is obtained based on the first residual and the temperature-sensitive sequence, and the calculation formula is as follows: Perform a Fourier transform on the second residual to obtain the amplitude of the frequency components; select the k most significant frequency components by amplitude value sorting, where k is a positive integer, and the selection of k is based on the minimum number of frequency components that satisfy the requirement of having an energy share of 90%, calculated using the following formula: in, Amplitude values ​​arranged in descending order , Let m be the amplitude of the m-th frequency component. It is the set of positive integers. This is used as an iterative variable to test different component quantities until the energy percentage condition is met; By reconstructing the signal using significant frequency components, a periodic sequence is obtained, specifically: in, The number of significant frequency components, The physical frequency corresponding to the j-th significant component. The phase angle, For electrical load Periodic components, periodic sequences ; The residual sequence is obtained based on the periodic sequence, specifically as follows: Among them, residual sequence ; The preliminary prediction module is configured to input the base charge sequence, the temperature-sensitive sequence, the periodic sequence, and the residual sequence into the corresponding prediction models to obtain the corresponding prediction results. The prediction module is configured to superimpose the various prediction results to obtain the power load prediction result.

6. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the power load forecasting method based on multimodal decomposition as described in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the power load forecasting method based on multimodal decomposition as described in any one of claims 1-4.

8. A computer program product, characterized in that, The method includes a computer program that, when executed by a processor, implements the power load forecasting method based on multimodal decomposition as described in any one of claims 1-4.

Images