Electricity price prediction method, system and device based on learning architecture and iTransform and medium
By introducing the maximum information coefficient screening feature and the TimeKAN frequency decomposition learning architecture, combined with the iTransformer model, the problems of insufficient frequency component modeling and inflexible feature selection in electricity price forecasting are solved, achieving high-precision and efficient electricity price forecasting.
Patent Information
- Application Number
- CN202510915511.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-03
- Publication Date
- 2025-10-17
AI Technical Summary
When processing electricity price series, existing electricity price forecasting methods have insufficient frequency component modeling capabilities, and the feature selection mechanism lacks nonlinear expression capabilities. The model has weak reasoning efficiency and interpretability in high-frequency real-time scenarios, and it is difficult to effectively integrate local and global frequency features to improve prediction stability and accuracy.
The maximum information coefficient is used to screen feature variables, and the TimeKAN frequency decomposition learning architecture is constructed. Combined with the iTransformer model, the electricity price time series is modeled, and electricity prices are predicted through iterative training and evaluation.
It improves the accuracy and stability of electricity price forecasting, enhances interpretability and real-time performance, reduces redundant feature interference, and improves the model's generalization ability and deployment efficiency.
Smart Images
Figure CN120805043A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computer platform load balancing, and in particular to a method and system for predicting electricity prices based on a learning architecture and iTransformer. Background Art
[0002] With the in-depth development of smart grids and power market mechanisms, electricity price forecasting has gradually become a crucial component of power system dispatch optimization and market operation. Accurate electricity price forecasting not only helps improve grid operating efficiency and supply-demand matching capabilities, but also has important implications for demand response, power trading strategy formulation, and distributed energy grid connection and scheduling. In recent years, with the introduction of deep learning technology, electricity price forecasting methods have evolved from early statistical models such as ARIMA to time series forecasting models based on structures such as LSTM, GRU, and Transformer. These methods have significantly improved the ability to model nonlinear characteristics and time series dependencies, laying the technical foundation for improving electricity price forecasting accuracy.
[0003] Although existing deep learning models such as LSTM, GRU, and Transformer have achieved certain results in modeling complex time series, they still face significant challenges in electricity price forecasting. First, electricity price series naturally have high-frequency perturbations and multi-scale fluctuations, and the frequency components are mixed. Traditional models have difficulty in effectively separating and modeling these components, resulting in limited prediction accuracy. Second, existing methods often lack flexibility when dealing with multi-source heterogeneous input features. In particular, most feature selection methods rely on linear correlation indicators, ignoring the possible strong nonlinear correlation between input variables and target variables, thereby introducing redundant features and reducing modeling efficiency. Thirdly, although the Transformer model has strong time series modeling capabilities, it still has problems such as low reasoning efficiency and poor model interpretability after deep stacking, which limits its adaptability to high-frequency rapid prediction in the real-time power market. In contrast, the present invention effectively extracts the most relevant features to the target by introducing a nonlinear feature screening mechanism based on the maximum information coefficient (MIC); and uses the TimeKAN frequency decomposition learning architecture to deeply model the global and local frequency features. Combined with the structurally optimized iTransformer model, it improves the prediction accuracy while enhancing the interpretability and real-time performance, which is a technological breakthrough that is difficult to achieve with existing technologies. Summary of the Invention
[0004] In view of the above-mentioned problems, the present invention is proposed.
[0005] Therefore, the present application solves the technical problems of the existing electricity price prediction method, including insufficient frequency component modeling capability, lack of nonlinear expression capability in feature selection mechanism, weak inference efficiency and interpretability of the model in high-frequency real-time scenarios, and how to fuse local and global frequency features to improve prediction stability and accuracy.
[0006] To solve the above technical problems, the present application provides the following technical solutions: a learning architecture and iTransformer-based electricity price prediction method, including preprocessing the original electricity price dataset and screening feature variables using the maximum information coefficient; constructing a TimeKAN frequency decomposition learning architecture and modeling the time series to obtain an iTransformer model; based on the iTransformer model, iteratively training the TimeKAN embedded sequence and outputting the electricity price prediction result and performing prediction evaluation.
[0007] As a preferred scheme of the learning architecture and iTransformer-based electricity price prediction method of the present application, wherein: the preprocessing includes obtaining each feature data related to the electricity price, forming an electricity price dataset, and filling in the missing values of the data in the electricity price dataset using linear interpolation, and checking for outliers using the IQR boxplot method; the outlier detection includes calculating the first quartile Q1 and the third quartile Q3 of the variable, and calculating the interquartile range IQR, defining outliers, and filling in the identified outliers using linear interpolation; the calculation of the interquartile range IQR is represented as: the definition of outliers includes: if the calculated IQR value sample is less than Q1-1.5*IQR or greater than Q3+1.5*IQR, it is defined as an outlier; after processing, the remaining samples are normalized using the maximum and minimum normalization method, and are uniformly scaled to the [0, 1] interval.
[0008] As a preferred scheme of the learning architecture and iTransformer-based electricity price prediction method of the present application, wherein: the screening of feature variables includes calculating the maximum information coefficient value MIC between the electricity price feature and the electricity price for each input electricity price feature, setting a minimum MIC threshold, and filtering target variables based on the minimum MIC threshold to obtain a feature set that meets the threshold.
[0009] As a preferred scheme of the learning architecture-based and iTransformer electricity price prediction method provided by the application, the TimeKAN frequency decomposition learning architecture comprises: mean removal and variance scaling are respectively performed on each time series according to a channel; a kernel function mapping layer based on a Chebyshev polynomial is constructed, high-order nonlinear mode representations are obtained by performing arccos+cos transformation on input data and using polynomial weighted summation; the entire time series is mapped by using a ChebyKANLayer module to capture long-distance global time correlation; a one-dimensional convolution module is used to model a local segment of the time series; an activation function, batch normalization and Dropout are set to perform iterative convergence and generalization.
[0010] As a preferred scheme of the learning architecture-based and iTransformer electricity price prediction method provided by the application, the modeling of the time series comprises: element-wise addition fusion of the global time correlation and the model constructed by the local segment is performed to form a multi-frequency joint representation, the fused representation contains global and local time dependence features, and the final multi-order KAN output is formed as an embedding input of sequence modeling for use by a downstream iTransformer model.
[0011] As a preferred scheme of the learning architecture-based and iTransformer electricity price prediction method provided by the application, the iterative training of the TimeKAN embedding sequence comprises: an iTransformer model is established, time stamp information and position embedding are added to the TimeKAN architecture input sequence to form a time sequence embedding layer, input sequence length, prediction sequence length, label sequence length and batch size are set, and the model is optimized by using an Adam optimizer; the iTransformer model is iteratively trained by using a training set, mean square error is used as a loss function, and the model parameters are optimized by using a back propagation algorithm.
[0012] As a preferred scheme of the learning architecture-based and iTransformer electricity price prediction method provided by the application, the prediction evaluation comprises: after each training cycle ends, the performance of the iTransformer model on unseen data is evaluated by using a validation set, a validation loss is calculated, and the training process is monitored by using an Early Stopping strategy; if the performance of the validation set no longer improves, the training is stopped, and the model parameters with the optimal validation loss are saved as the final model; the prediction result and the actual value are de-normalized to restore the original scale, and the prediction performance of the model is evaluated by using a prediction evaluation index; the prediction evaluation index comprises mean absolute percentage error, mean absolute error and mean square error.
[0013] Another object of the present application is to provide a learning architecture and iTransformer electricity price prediction system, which can solve the technical problems of unclear frequency mixed modeling, low feature utilization rate and insufficient reasoning efficiency in current electricity price sequence prediction technology by introducing a feature screening mechanism based on the maximum information coefficient, constructing a TimeKAN frequency learning architecture and fusing an optimized iTransformer model.
[0014] As a preferred scheme of the learning architecture and iTransformer electricity price prediction system, the scheme comprises a feature screening module, a model construction module and a prediction evaluation module; the feature screening module is used for preprocessing an original electricity price data set and screening feature variables by using the maximum information coefficient; the model construction module is used for constructing a TimeKAN frequency decomposition learning architecture and modeling a time sequence to obtain an iTransformer model; and the prediction evaluation module is used for iteratively training a TimeKAN embedded sequence based on the iTransformer model and outputting an electricity price prediction result and performing prediction evaluation.
[0015] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps of the learning architecture and iTransformer electricity price prediction method.
[0016] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the learning architecture and iTransformer electricity price prediction method.
[0017] The learning architecture and iTransformer electricity price prediction method provided by the present application realizes non-parametric and non-linear correlation analysis by introducing the maximum information coefficient as a feature selection indicator, effectively filters low-value features, reduces redundant information interference, and improves model generalization ability and training efficiency; by constructing a TimeKAN frequency decomposition learning architecture, fusing a cascaded frequency decomposition, multi-order KAN learning and frequency mixing module, and deeply modeling global and local frequency dependent features in the electricity price time sequence, the model's expression power in processing high-frequency disturbance and multi-scale fluctuation electricity price data is enhanced; by combining the structure-optimized iTransformer model, the prediction accuracy is guaranteed, the improved attention mechanism and efficient coding structure reduce the computational complexity and reasoning time of the model, improve the practicability and deployment efficiency of the model, and at the same time, the multi-frequency embedded representation output by TimeKAN improves the explainability of the entire system to the modeling result, facilitating the operation personnel to understand the prediction basis. BRIEF DESCRIPTION OF DRAWINGS
[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. 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 creative effort on the basis of these drawings.
[0019] Figure 1 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application.
[0020] Figure 2 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application.
[0021] Figure 3 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application.
[0022] Figure 4 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application.
[0023] Figure 5 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application.
[0024] Figure 6 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application. DETAILED DESCRIPTION
[0025] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort should be within the scope of protection of the present application.
[0026] Embodiment 1, refer to Figure 1 The present application provides a learning architecture and iTransformer electricity price prediction method based on an embodiment of the present application, comprising:
[0027] S1: preprocessing the original electricity price dataset and screening the characteristic variables by using the maximum information coefficient.
[0028] Further, the preprocessing includes obtaining each characteristic data related to the electricity price, forming the electricity price dataset, and filling the missing values in the electricity price dataset by using the linear interpolation method and checking the outliers by using the IQR boxplot method.
[0029] The outlier detection includes calculating the first quartile Q1 and the third quartile Q3 of the variable and calculating the interquartile range IQR, defining the outliers, and filling the identified outliers by using the linear interpolation method.
[0030] The interquartile range IQR is calculated and represented as:
[0031] IQR = Q3 - Q1
[0032] The definition of outliers includes that if the calculated IQR value sample is less than Q1 - 1.5*IQR or greater than Q3 + 1.5*IQR, it is defined as an outlier.
[0033] After the processing is completed, the retained samples are subjected to the maximum and minimum normalization processing, and are uniformly scaled to the [0, 1] interval.
[0034] The data is normalized to the interval to eliminate the adverse effects of different dimensions on model training, a time window is defined, and the training set, the test set and the validation set are divided in the ratio of 6:2:2. For a certain characteristic variable x, the maximum and minimum normalization formula is represented as:
[0035]
[0036] wherein x min represents the minimum value of the characteristic x in all samples, x max represents the maximum value of the characteristic x in all samples, and x` represents the normalized value.
[0037] It should be noted that the screening of the characteristic variables includes calculating the maximum information coefficient value MIC between the electricity price characteristic and the electricity price for each input electricity price characteristic, setting a minimum MIC threshold, and filtering the target variable based on the minimum MIC threshold to obtain a characteristic set that meets the threshold.
[0038] It should be noted that by analyzing the nonlinear correlation between the input characteristics and the target variable, the limitation of linear correlation screening is broken through, a non-parametric and nonlinear feature screening mechanism is introduced, and the features that contribute most to the prediction are automatically screened to improve the model performance and reduce the interference of redundant features.
[0039] S2: Constructing TimeKAN frequency decomposition learning architecture and modeling time series to obtain an iTransformer model.
[0040] Further, constructing the TimeKAN frequency decomposition learning architecture includes performing mean removal and variance scaling on each time series according to the channel; constructing a kernel function mapping layer based on a Chebyshev polynomial, obtaining a representation of high-order nonlinear mode by performing arccos+cos transformation on the input data and using polynomial weighted summation; using a ChebyKANLayer module to map the entire time series, capturing long-distance global time correlation; using a one-dimensional convolution module to model the local segment of the time series; setting the activation function, batch normalization and Dropout, and performing iterative convergence and generalization.
[0041] It should be noted that modeling the time series includes element-wise addition fusion of the global time correlation and the model constructed by the local segment to form a multi-frequency joint representation, and the fused representation contains global and local time dependence features to form a final multi-order KAN output as an embedding input for the downstream iTransformer model.
[0042] S3: Iteratively training the TimeKAN embedding sequence based on the iTransformer model and outputting the electricity price prediction result and performing prediction evaluation.
[0043] Further, iteratively training the TimeKAN embedding sequence includes establishing an iTransformer model, adding timestamp information and position embedding to the TimeKAN architecture input sequence to form a time series embedding layer, setting the input sequence length, the prediction sequence length, the label sequence length, and the batch size, and optimizing the model through the Adam optimizer; iteratively training the iTransformer model through the training set, using mean square error as the loss function, and optimizing the model parameters through the back propagation algorithm.
[0044] It should be noted that the prediction evaluation includes evaluating the performance of the iTransformer model on unseen data through the validation set after each training period, calculating the validation loss, and monitoring the training process through the Early Stopping strategy; if the performance of the validation set no longer improves, the training is stopped, and the model parameters with the optimal validation loss are saved as the final model; the prediction result and the actual value are denormalized to restore the original scale, and the prediction performance of the model is evaluated through the prediction evaluation index.
[0045] The prediction evaluation index includes the mean absolute percentage error, the mean absolute error, and the mean square error.
[0046] The mean absolute percentage error is expressed as:
[0047]
[0048] where N represents the total number of samples, y i represents the true value of the i-th sample, represents the predicted value, represents the average value of the true value.
[0049] The mean absolute error is expressed as:
[0050]
[0051] The mean square error is expressed as:
[0052]
[0053] Example 2, reference Figures 2-5 An iTransformer electricity price prediction method based on a learning architecture is provided as an embodiment of the present application. In order to verify the beneficial effects of the present application, economic benefit calculation and simulation experiments are used for scientific demonstration.
[0054] In this embodiment, the actual operation data of the Australian electricity market is used to construct an electricity price prediction model. The data set covers multi-dimensional features, including historical electricity prices, power load records, meteorological parameters, and date type information. The data time span is half a year, the sampling frequency is 30 minutes, and the dynamic change process of the electricity price is recorded completely. In order to verify the effectiveness of the electricity price prediction method proposed in the present application, comparative experiments with various models are carried out. The results show that the TimeKAN-iTransformer model proposed in this paper performs best in the electricity price prediction task, has the optimal error control and higher prediction accuracy, and the results are shown in Tables 1, Figures 2-5 and 2,
[0055] Example 3, as the second embodiment of the present application, is different from the first two embodiments in that:
[0056] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the technical solutions that essentially contribute to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.
[0057] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logic functions, and can be specifically embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, a system including a processor, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or in conjunction with these instructions execution systems, apparatuses, or devices. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport programs for use by an instruction execution system, apparatus, or device, or in conjunction with these instruction execution systems, apparatuses, or devices.
[0058] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by editing, interpreting, or otherwise processing, if necessary, in other suitable ways to be electronically obtained, and then stored in the computer memory.
[0059] It should be understood that various parts of the present application can be realized in hardware, software, firmware, or a combination thereof. In the above-described embodiments, a plurality of steps or methods can be realized with software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized in hardware, and as in another embodiment, it can be realized with any one or a combination of the following technologies known in the art: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.
[0060] Embodiment 4, with reference to Figure 6 As a third embodiment of the present application, the embodiment provides a system based on a learning architecture and an iTransformer electricity price prediction method, comprising a feature screening module, a model construction module, and a prediction evaluation module; the feature screening module is used for preprocessing the original electricity price data set and screening feature variables by using the maximum information coefficient; the model construction module is used for constructing a TimeKAN frequency decomposition learning architecture and modeling a time series to obtain an iTransformer model; the prediction evaluation module is used for iteratively training a TimeKAN embedded sequence based on the iTransformer model and outputting an electricity price prediction result and performing prediction evaluation.
[0061] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.
Claims
1. A method for predicting electricity prices based on a learning architecture and iTransformer, characterized in that: include: The original electricity price dataset is preprocessed and the characteristic variables are screened using the maximum information coefficient; Build the TimeKAN frequency decomposition learning architecture and model the time series to obtain the iTransformer model; The TimeKAN embedding sequence is iteratively trained based on the iTransformer model and the electricity price prediction results are output and the prediction evaluation is performed.
2. The method for predicting electricity prices based on a learning architecture and iTransformer according to claim 1, wherein: The preprocessing includes obtaining various characteristic data related to electricity prices to form an electricity price data set, and filling missing values in the data in the electricity price data set using linear interpolation and checking for outliers using IQR box plot method; The outlier detection includes calculating the first quartile Q1 and the third quartile Q3 of the variable, and calculating the interquartile range IQR, defining outliers, and filling the identified outliers using linear interpolation; The definition of an outlier includes: if the calculated IQR value sample is less than Q1-1.5*IQR or greater than Q3+1.5*IQR, it is defined as an outlier; After the processing is completed, the retained samples are normalized to the maximum and minimum values and uniformly scaled to the interval [0,1].
3. The method for predicting electricity prices based on a learning architecture and iTransformer according to claim 2, wherein: The screening of characteristic variables includes, for each input electricity price feature, calculating the maximum information coefficient value MIC between the electricity price feature and the electricity price, setting a minimum MIC threshold, and filtering the target variable based on the minimum MIC threshold to obtain a feature set that meets the threshold.
4. The method for predicting electricity prices based on a learning architecture and iTransformer according to claim 3, wherein: The construction of the TimeKAN frequency decomposition learning architecture includes performing mean removal and variance scaling on each time series according to the channel; Construct a kernel function mapping layer based on Chebyshev polynomials, perform arccos+cosine transformation on the input data and use polynomial weighted summation to obtain the representation of high-order nonlinear patterns; Use the ChebyKANLayer module to map the entire time series and capture long-range global temporal correlations; Use one-dimensional convolution modules to model local segments of time series; Set activation function, batch normalization, and Dropout for iterative convergence and generalization.
5. The method for predicting electricity prices based on a learning architecture and iTransformer according to claim 4, characterized in that: The time series modeling includes element-by-element additive fusion of the global temporal correlation and the local segment-based models to form a multi-frequency joint representation. The fused representation contains global and local temporal dependency features, forming a final multi-order KAN output that serves as the embedding input for sequence modeling and is used by the downstream iTransformer model.
6. The method for predicting electricity prices based on a learning architecture and iTransformer according to claim 5, characterized in that: The iterative training of the TimeKAN embedding sequence includes establishing an iTransformer model, adding timestamp information and position embedding to the TimeKAN architecture input sequence to form a time series embedding layer, setting the input sequence length, prediction sequence length, label sequence length, and batch size, and optimizing the model using the Adam optimizer; The iTransformer model is iteratively trained using the training set, using mean square error as the loss function, and the model parameters are optimized through the back propagation algorithm.
7. The method for predicting electricity prices based on a learning architecture and iTransformer according to claim 6, wherein: The prediction evaluation includes evaluating the performance of the iTransformer model on unseen data using a validation set after each training cycle, calculating the validation loss, and monitoring the training process using an Early Stopping strategy; If the performance of the validation set no longer improves, stop training and save the model parameters with the best validation loss as the final model; Denormalize the predicted results and actual values to restore them to their original scale, and evaluate the prediction performance of the model using the prediction evaluation index; The prediction evaluation indicators include mean absolute percentage error, mean absolute error and mean square error.
8. A system using the method for electricity price prediction based on a learning architecture and iTransformer according to any one of claims 1 to 7, characterized in that: Including feature screening module, model building module, and prediction evaluation module; The feature screening module is used to pre-process the original electricity price data set and screen the feature variables using the maximum information coefficient; The model building module is used to build the TimeKAN frequency decomposition learning architecture and model the time series to obtain the iTransformer model; The prediction and evaluation module is used to iteratively train the TimeKAN embedding sequence based on the iTransformer model, output the electricity price prediction results, and perform prediction evaluation.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the electricity price prediction method based on the learning architecture and iTransformer are implemented as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the electricity price prediction method based on a learning architecture and iTransformer are implemented as described in any one of claims 1 to 7.