A method, system, device, and medium for predicting extreme values ​​of charging power in a charging field.

Abstract

This invention relates to a method, system, device, and medium for predicting extreme values ​​of charging power in charging stations, aiming to solve the problems of low prediction accuracy and weak model generalization ability in existing technologies. The method includes: acquiring the charging power sequence within the current sampling period, extracting the feature sequence using a KANLayer, and then inputting it into an LSTMLayer to predict the charging power for the next sampling period. The KANLayer enhances the temporal features of the input data through Fourier transform, and the filtering module automatically adjusts the threshold using a pre-trained MLP model, improving the accuracy and efficiency of feature extraction. This invention, by combining an improved KAN model with LSTM, enhances the prediction model's sensitivity to extreme values ​​and its generalization ability, making it suitable for intelligent power management in new energy charging stations.

Description

Technical Field

[0001] This invention relates to intelligent power management of new energy charging stations, and in particular to a method, system, device and medium for predicting extreme values ​​of charging power in charging stations. Background Technology

[0002] In current power prediction technologies, traditional prediction models often struggle to accurately capture extreme values ​​of charging power when processing complex charging power time series data. While LSTM (Long Short-Term Memory) models have strong generalization capabilities, they are highly sensitive to data quality, and outliers can significantly impact training results, necessitating the use of new methods to improve data quality. Furthermore, existing technologies rely on simple feature extraction methods when processing charging power sequences, failing to fully uncover complex features and effectively handle high-frequency noise, periodic signals, and nonlinear relationships, resulting in insufficient feature extraction. Traditional machine learning models may face low training efficiency when processing large-scale time series data. Using KAN (Kolmogorov-Arnold Networks) models to improve feature extraction and data quality is a feasible solution, but KAN models, using B-spline basis function fitting, are poorly sensitive to extreme values, requiring new methods to enhance their sensitivity to these extremes. Additionally, these models may struggle to generalize to new datasets or scenarios, limiting their applicability. Therefore, a new model and method with strong generalization ability is needed to accurately predict the extreme values ​​of charging power in charging fields. Summary of the Invention

[0003] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a method, system, device and medium for predicting extreme values ​​of charging power in a charging field, in order to solve the problems of low prediction accuracy and weak generalization ability of prediction models in the prior art.

[0004] To achieve the above and other related objectives, this invention provides a method, system, device, and medium for predicting extreme values ​​of charging power in charging stations, applied to an intelligent power management system for new energy charging stations. The method includes: acquiring the current charging power sequence within the current sampling period; inputting the current charging power sequence into the KANLayer of a charging station charging power extreme value prediction model, and extracting a feature sequence based on Fourier transform and B-spline basis functions; inputting the feature sequence into the LSTMLayer of the charging station charging power extreme value prediction model to obtain the predicted charging power for the next sampling period, wherein the peak and valley values ​​in the predicted charging power are the extreme values ​​of charging power for the next sampling period.

[0005] In one embodiment of the present invention, the step of obtaining the current charging power sequence within the current sampling period includes: sampling the charging power within the current sampling period at a preset time interval and saving the sampling time; and combining all the charging power according to the sampling time to generate the current charging power sequence.

[0006] In one embodiment of the present invention, the step of combining all the charging power to generate a current charging power sequence according to the sampling time includes: combining all the charging power to generate an initial charging power sequence according to the sampling time; normalizing the initial charging power sequence; and obtaining the current charging power sequence through a preset sliding window.

[0007] In one embodiment of the present invention, the current charging power sequence is input into the KANLayer of the charging power extreme value prediction model of the charging field, and a feature sequence is extracted according to the Fourier transform and B-spline basis function. The method includes: inputting the current charging power sequence into the Fourier transform module of the KANLayer to obtain a time-domain enhanced sequence; and inputting the time-domain enhanced sequence into the B-spline basis function of the KANLayer for smoothing to obtain the feature sequence.

[0008] In one embodiment of the present invention, the step of inputting the current charging power sequence into the Fourier transform module of the KANLayer to obtain a time-domain enhanced sequence includes: inputting the current charging power sequence into the Fourier transform module of the KANLayer to obtain a frequency domain amplitude spectrum through Fourier transform; inputting the frequency domain amplitude spectrum into the filtering module of the KANLayer to filter and obtain a characteristic frequency domain amplitude spectrum; and performing an inverse Fourier transform on the characteristic frequency domain amplitude spectrum through the Fourier transform module to obtain a time-domain enhanced sequence.

[0009] In one embodiment of the present invention, the filtering module employs a pre-trained MLP (Multilayer Perceptron) model, automatically adjusts the threshold according to the frequency domain amplitude spectrum, and filters the feature frequency domain amplitude spectrum based on the threshold.

[0010] In one embodiment of the present invention, the extreme value prediction model of the charging power of the charging field is obtained through pre-training, which uses mean square error as the loss function and Adam (Adaptive Moment Estimation) as the optimizer.

[0011] In one embodiment of the present invention, a charging power extreme value prediction system for a charging field is also provided. The system includes: a data acquisition unit for acquiring the current charging power sequence within the current sampling period; a feature sequence extraction unit for inputting the current charging power sequence into the KANLayer of the charging power extreme value prediction model for the charging field, and extracting feature sequences based on Fourier transform and B-spline basis functions; and a prediction unit for inputting the feature sequences into the LSTMLayer of the charging power extreme value prediction model for the charging field, and obtaining the predicted charging power for the next sampling period, wherein the peak and valley values ​​in the predicted charging power are the charging power extreme values ​​for the next sampling period.

[0012] In one embodiment of the present invention, an electronic device is also provided, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the electronic device enables the charging power extreme value prediction method for charging fields as described above.

[0013] In one embodiment of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a computer processor, causes the computer to perform the charging power extreme value prediction method described above.

[0014] As described above, the charging power extreme value prediction method, system, device, and medium of the present invention have the following beneficial effects: the charging power extreme value prediction model of the charging site, by combining KANLayer and LSTMLayer, using mean square error as the loss function and adopting the Adam optimizer training strategy, further improves the training efficiency and generalization ability of the model; the addition of Fourier transform processing to the KAN model achieves high-precision feature extraction. This technology can not only effectively suppress noise and capture periodicity and long-term dependencies in the data, but also improve the accuracy and efficiency of feature extraction through an adaptive feature selection mechanism, providing an efficient, accurate, and practical systematic solution for the power management of new energy charging sites, and significantly improving the scientificity and reliability of power grid dispatching and charging facility planning. Attached Figure Description

[0015] Figure 1 A flowchart illustrating a method for predicting extreme values ​​of charging power in a charging field, provided in an embodiment of the present invention;

[0016] Figure 2 The diagram shown is a structural block diagram of a charging power extreme value prediction system for a charging field according to an embodiment of the present invention.

[0017] Figure 3 The diagram shown is a structural schematic of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0018] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0019] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0020] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0021] In new energy power systems, power output is influenced by various factors, exhibiting high uncertainty and nonlinearity. However, existing power system prediction technologies primarily focus on overall forecasting, neglecting the accuracy of predictions for extreme power values. Yet, extreme value prediction is crucial for charging stations. Appropriately configuring circuit facilities based on peak power levels at charging stations can effectively ensure equipment and grid safety, and optimize grid connection and charging efficiency. Among current power prediction technologies, while the LSTM model has strong generalization ability, it requires high data quality; outliers can significantly impact model training results, necessitating new methods to improve data quality. Using the KAN model to improve data quality is a feasible solution, but the KAN model's B-spline basis function fitting method has poor sensitivity to extreme values, requiring new methods to enhance its sensitivity to these values.

[0022] To address the above issues, this invention provides a method for predicting extreme values ​​of charging power in a charging field. By adding a Fourier transform module to the KAN model, the method filters out frequency domain components of greater interest in the frequency domain of extreme charging power values. The hidden layers of the charging power extreme value prediction model used in this method include a KANLayer and an LSTMLayer with added Fourier transform modules. The KANLayer uses the KAN model as the first hidden layer of the charging power extreme value prediction model, and the LSTMLayer uses the LSTM model as the second hidden layer. Combining the KAN model and the LSTM model can improve the prediction accuracy of extreme values ​​without reducing the original performance.

[0023] Please see Figure 1 The flowchart illustrates a method for predicting extreme values ​​of charging power in a charging field according to an exemplary embodiment of this application, including the following steps:

[0024] S1, obtain the current charging power sequence within the current sampling period.

[0025] The real-time charging power values ​​and collection time of the charging site are recorded within a preset sampling period as timestamps. The collected charging power values ​​are grouped into a data list to obtain the charging power sequence. When obtaining the current charging power sequence within the sampling period, the collection process needs to be carried out under the actual operating conditions of the charging site.

[0026] Specifically, in one embodiment of the present invention, obtaining the current charging power sequence within the current sampling period includes:

[0027] The charging power within the current sampling time period is sampled according to a preset time interval, and the sampling time is saved;

[0028] According to the sampling time, all the charging powers are combined to generate the current charging power sequence.

[0029] In a preferred embodiment of the present invention, the sampling time is set to one day and the time interval is set to 15 minutes, then there are 96 charging powers.

[0030] Specifically, in one embodiment of the present invention, the step of combining all the charging powers to generate the current charging power sequence according to the sampling time includes:

[0031] According to the sampling time, all the charging powers are combined to generate an initial charging power sequence;

[0032] The initial charging power sequence is normalized, and the current charging power sequence is obtained through a preset sliding window.

[0033] Based on the order of sampling time, the collected charging power is sorted to obtain the initial charging power sequence.

[0034] Furthermore, the initial charging power sequence is preprocessed, including but not limited to data cleaning and standardization, to eliminate the influence of noise and outliers.

[0035] The preprocessed initial charging power sequence is then normalized to convert the collected charging power into standardized data. In a preferred embodiment, the Min-max method is used for normalization. The Min-max normalization method is a linear transformation that does not change the distribution shape of the data; it only adjusts the scale of the data. It is computationally efficient and fast. The normalization formula is as follows: Where x min x is the minimum value in this feature column. max The maximum value in this feature column, x norm This is the normalized value.

[0036] To ensure the consistency of data size input to the charging power extreme value prediction model, a sliding window is set to process the normalized initial charging power sequence to obtain a current charging power sequence of fixed size. In a preferred embodiment of the present invention, since the sampling time is one day and the sampling interval is 15 minutes, the size of the sliding window is set to 96.

[0037] S2, input the current charging power sequence into the KANLayer of the charging field charging power extreme value prediction model, and extract the feature sequence according to the Fourier transform and B-spline basis function.

[0038] KANLayer is the first hidden layer of the charging power extreme value prediction model for the charging field. It includes Fourier transform frequency domain filtering and B-spline basis function processing. The input is the current charging power sequence, and the output is the extracted feature sequence.

[0039] Specifically, in one embodiment of the present invention, the current charging power sequence is input into the KANLayer of the charging power extreme value prediction model of the charging field, and the feature sequence is extracted according to the Fourier transform and B-spline basis function, including:

[0040] The current charging power sequence is input into the Fourier transform module of the KANLayer to obtain the time-domain enhanced sequence, specifically including:

[0041] First, the current charging power sequence is input into the Fourier transform module of the KANLayer, and the frequency domain amplitude spectrum is obtained through Fourier transform.

[0042] The current charging power sequence is input into the Fourier transform module of the KANLayer charging power extreme value prediction model. The Fourier transform module converts the current charging power sequence in the time domain into the frequency domain amplitude spectrum information based on the Fast Fourier Transform (FFT). The FFT formula is as follows: X[k] is the frequency domain representation, x[n] is the time domain sequence data, N is the length of the input discrete sequence, k is the frequency index, and n represents the discrete time index of the time domain signal.

[0043] Next, the frequency domain amplitude spectrum is input into the filtering module of the KANLayer to obtain the characteristic frequency domain amplitude spectrum.

[0044] In a preferred embodiment of the present invention, the filtering module employs a pre-trained MLP model, automatically adjusts the threshold based on the frequency domain amplitude spectrum, and filters the feature frequency domain amplitude spectrum according to the threshold. By automatically adjusting the threshold based on the frequency domain amplitude spectrum and retaining frequency domain components above the threshold, and then filtering the feature frequency domain amplitude spectrum based on the threshold, this step can more effectively filter feature data components with extreme values ​​in the input data.

[0045] Finally, the characteristic frequency domain amplitude spectrum is subjected to inverse Fourier transform through the Fourier transform module to obtain the time-domain enhanced sequence.

[0046] The selected characteristic frequency domain amplitude spectra are transformed back to the time domain using an inverse Fourier transform to obtain a time-domain enhanced sequence; wherein, the formula for the inverse fast Fourier transform is: x[n] is the reconstructed time-domain sequence, X[k] is the frequency-domain representation, N is the length of the input discrete sequence, k is the frequency index, and n represents the discrete-time index of the time-domain signal.

[0047] The time-domain enhanced sequence is input into the B-spline basis function of the KANLayer for smoothing to obtain the feature sequence.

[0048] The B-spline basis function is: When k = 0, the formula for the zero-order B-spline basis function is:

[0049] When k>0, the higher-order B-spline basis functions are defined recursively:

[0050]

[0051] Where t is the parameter, i is the index of the basis function, k is the order of the B-spline, and N is the number of base functions. i,k (t) represents the B-spline basis function, and N is the k-th order B-spline basis function. i,k (t) is composed of two k-1 order B-spline basis functions N. i,k-1 (t) and N i+1,k-1The linear combination of (t). The calculation process of B-spline basis functions mainly involves fitting and approximating the data using B-spline functions. However, data with periodic characteristics has a complex periodicity in the time domain, which is difficult to accurately describe using simple polynomials. Using spline functions may result in inaccurate fitting of these periodicities and fail to effectively extract the periodic patterns in the data. However, power data usually contains multiple periodic components of different frequencies. Combining Fourier transform processing can effectively solve the shortcomings of B-spline basis functions in processing periodic data.

[0052] S3. Input the feature sequence into the LSTMLayer of the charging power extreme value prediction model of the charging field to obtain the predicted charging power for the next sampling period, wherein the peak and valley values ​​in the predicted charging power are the extreme values ​​of the charging power for the next sampling period.

[0053] The model used by the charging power extreme value prediction model or LSTMLayer can be any deep learning model within a sequence-to-sequence (Seq2Seq) framework. The Seq2Seq framework comprises an encoder and a decoder, both of which can be constructed from neural networks. These neural networks include, but are not limited to, Bi-directional Gate Recurrent Units (Bi-GRUs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Bi-directional LSTM models. For example, when the charging power extreme value prediction model is a Bi-directional LSTM model, the feature sequence is input into the encoder of the LSTMLayer model to extract the time dependency information and generate multiple hidden state vectors. Each hidden state vector is then input into the decoder. Based on the current hidden state vector and the output of the previous time step, each time step is processed recursively to progressively generate the charging power sequence for the next sampling period.

[0054] It should be noted that the charging power extreme value prediction model for charging sites is trained. The dataset used to train the model can be obtained through open-source channels or collected independently. Specifically, the charging power extreme value prediction model used in the method is pre-trained.

[0055] Specifically, in one embodiment of the present invention, the training method for the extreme value prediction model of charging power in a charging field includes:

[0056] A charging power sequence dataset is obtained, which includes charging power sequences collected at multiple consecutive collection times; wherein the charging power sequence at the current collection time is used as sample data, and the charging power sequence at the next collection time is used as the corresponding label data.

[0057] Preferably, a day is used as a collection time, and data is collected once every 15 minutes to obtain a daily charging power sequence as sample data.

[0058] All sample data are preprocessed to obtain preprocessed data, and the preprocessed data is then divided into training and test sets according to the proportions.

[0059] In a preferred embodiment of the present invention, 90% of all preprocessed data is used as the training set, and the remaining 10% is used as the test set.

[0060] The training set is input into the charging power extreme value prediction model of the charging site to be trained to obtain the prediction result. Based on the prediction result and the corresponding label, the loss function is calculated. The parameters of the charging power extreme value prediction model of the charging site are adjusted according to the loss function to obtain the trained charging power extreme value prediction model.

[0061] In a preferred embodiment of the present invention, mean squared error is used as the loss function. The mean squared error is calculated based on the prediction results and corresponding labels, and the model is trained using the adaptive moment estimator (Adam) based on the mean squared error.

[0062] The test set is input into the trained charging power extreme value prediction model of the charging site to obtain the corresponding prediction results. The prediction results are compared with the corresponding labels. When the comparison results meet the preset conditions, the charging power extreme value prediction model of the charging site is trained well; otherwise, sample data is added and the trained charging power extreme value prediction model of the charging site is trained again.

[0063] Please see Figure 2 The charging power extreme value prediction system 200 for a charging site includes: a data acquisition unit 210, a feature sequence extraction unit 220, and a prediction unit 230. The data acquisition unit 210 acquires the current charging power sequence within the current sampling period. The feature sequence extraction unit 220 inputs the current charging power sequence into the KANLayer of the charging site charging power extreme value prediction model and extracts the feature sequence based on Fourier transform and B-spline basis functions. The prediction unit 230 inputs the feature sequence into the LSTMLayer of the charging site charging power extreme value prediction model to obtain the predicted charging power for the next sampling period, wherein the peak and valley values ​​in the predicted charging power are the extreme values ​​of the charging power for the next sampling period.

[0064] Specific limitations regarding the extreme value prediction system for charging power in charging stations can be found in the limitations on the method for predicting extreme values ​​of charging power in charging stations mentioned above, and will not be repeated here. Each module in the aforementioned extreme value prediction system for charging power in charging stations can be implemented entirely or partially through software, hardware, or a combination thereof.

[0065] The above modules can be embedded in the processor of the computer device in hardware format or independent of it, or they can be stored in the memory of the computer device in software format so that the processor can call the corresponding operations of the above modules.

[0066] It should be noted that, in order to highlight the innovative aspects of this invention, this embodiment does not include modules that are not closely related to solving the technical problems proposed by this invention, but this does not mean that there are no other modules in this embodiment.

[0067] Please see Figure 3 The electronic device 3 may include a memory 32, a processor 31 and a bus, and may also include a computer program stored in the memory 32 and executable on the processor 31, such as a Fourier transform program for charging power data.

[0068] The memory 32 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 32 can be an internal storage unit of the electronic device 3, such as a portable hard drive. In other embodiments, the memory 32 can be an external storage device of the electronic device 3, such as a plug-in portable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. Furthermore, the memory 32 can include both internal and external storage units of the electronic device 3. The memory 32 can be used not only to store application software and various types of data installed on the electronic device 3, such as optimization parameters of a charging power extreme value prediction model, but also to temporarily store data that has been output or will be output.

[0069] In some embodiments, the processor 31 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 31 is the control unit of the electronic device 3, connecting various components of the entire electronic device 3 via various interfaces and lines. It executes programs or modules stored in the memory 32 (e.g., correction functions for extreme power prediction models in charging fields) and calls data stored in the memory 32 to perform various functions and process data in the electronic device 3.

[0070] The processor 31 executes the operating system of the electronic device 3 and various installed applications. The processor 31 executes the applications to implement the steps in the above-described method for predicting extreme values ​​of charging power in a charging field.

[0071] For example, the computer program may be divided into one or more modules, which are stored in the memory 32 and executed by the processor 31 to complete this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device 3.

[0072] The integrated unit implemented as a software functional module can be stored in a computer-readable storage medium, which can be non-volatile or volatile. The software functional module, stored in the storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute some functions of the charging power extreme value prediction method for charging fields described in the various embodiments of this application.

[0073] In summary, the charging power extreme value prediction method, system, device, and medium disclosed in this invention, by introducing an improved Fourier transform into the KAN Layer, can better capture the periodic and nonlinear characteristics of new energy power data, especially the extreme value characteristics, thus improving the ability to predict extreme values. Combining the improved KAN model with LSTM improves both the accuracy of extreme values ​​in power prediction and the overall prediction capability. It can better handle different types of new energy power data, exhibiting more stable and excellent performance in different datasets and scenarios, reducing the risk of overfitting, and improving the model's generalization ability. Therefore, this invention effectively overcomes the various shortcomings of existing technologies and has high industrial application value.

[0074] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for predicting extreme values ​​of charging power in a charging field, characterized in that, The method for predicting extreme charging power values ​​in new energy charging stations includes: Obtain the current charging power sequence within the current sampling period; The current charging power sequence is input into the KANLayer of the charging power extreme value prediction model of the charging field, and the feature sequence is extracted according to the Fourier transform and B-spline basis function. The feature sequence is input into the LSTMLayer of the charging power extreme value prediction model of the charging field to obtain the predicted charging power for the next sampling period. The peak and valley values ​​in the predicted charging power are the extreme values ​​of the charging power for the next sampling period. The step of inputting the current charging power sequence into the KANLayer of the charging power extreme value prediction model of the charging field, and extracting the feature sequence based on Fourier transform and B-spline basis functions, includes: The current charging power sequence is input into the Fourier transform module of the KANLayer to obtain the time-domain enhanced sequence; The time-domain enhanced sequence is input into the B-spline basis function of the KANLayer for smoothing to obtain the feature sequence; The step of inputting the current charging power sequence into the Fourier transform module of the KANLayer to obtain the time-domain enhanced sequence includes: The current charging power sequence is input into the Fourier transform module of the KANLayer, and the frequency domain amplitude spectrum is obtained through Fourier transform. The frequency domain amplitude spectrum is input into the filtering module of the KANLayer to obtain the characteristic frequency domain amplitude spectrum; the characteristic frequency domain amplitude spectrum is then subjected to inverse Fourier transform through the Fourier transform module to obtain the time domain enhancement sequence.

2. The method for predicting extreme values ​​of charging power in a charging field according to claim 1, characterized in that, The step of obtaining the current charging power sequence within the current sampling period includes: The charging power within the current sampling time period is sampled according to a preset time interval, and the sampling time is saved; According to the sampling time, all the charging powers are combined to generate the current charging power sequence.

3. The method for predicting extreme values ​​of charging power in a charging field according to claim 2, characterized in that, The step of combining all the charging powers to generate the current charging power sequence according to the sampling time includes: According to the sampling time, all the charging powers are combined to generate an initial charging power sequence; The initial charging power sequence is normalized, and the current charging power sequence is obtained through a preset sliding window.

4. The method for predicting extreme values ​​of charging power in a charging field according to claim 1, characterized in that, The filtering module uses a pre-trained MLP model to automatically adjust the threshold based on the frequency domain amplitude spectrum and filters the feature frequency domain amplitude spectrum according to the threshold.

5. The method for predicting extreme values ​​of charging power in a charging field according to claim 1, characterized in that, The extreme value prediction model for charging power in the charging field is obtained through pre-training, which uses mean square error as the loss function and Adam as the optimizer.

6. A system for predicting extreme values ​​of charging power in a charging field, characterized in that, The system includes: The data acquisition unit is used to acquire the current charging power sequence within the current sampling period; The feature sequence extraction unit is used to input the current charging power sequence into the KANLayer of the charging field charging power extreme value prediction model, and extract the feature sequence according to the Fourier transform and B-spline basis function. The prediction unit is used to input the feature sequence into the LSTMLayer of the charging power extreme value prediction model of the charging field to obtain the predicted charging power for the next sampling period, wherein the peak and valley values ​​in the predicted charging power are the extreme values ​​of the charging power for the next sampling period. The step of inputting the current charging power sequence into the KANLayer of the charging power extreme value prediction model of the charging field, and extracting the feature sequence based on Fourier transform and B-spline basis functions, includes: The current charging power sequence is input into the Fourier transform module of the KANLayer to obtain the time-domain enhanced sequence; The time-domain enhanced sequence is input into the B-spline basis function of the KANLayer for smoothing to obtain the feature sequence; The step of inputting the current charging power sequence into the Fourier transform module of the KANLayer to obtain the time-domain enhanced sequence includes: The current charging power sequence is input into the Fourier transform module of the KANLayer, and the frequency domain amplitude spectrum is obtained through Fourier transform. The frequency domain amplitude spectrum is input into the filtering module of the KANLayer to obtain the characteristic frequency domain amplitude spectrum; the characteristic frequency domain amplitude spectrum is then subjected to inverse Fourier transform through the Fourier transform module to obtain the time domain enhancement sequence.

7. An electronic device, characterized in that: The electronic device includes: One or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, cause the electronic device to implement the method for predicting extreme values ​​of charging power in a charging field as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the extreme value prediction method for charging power in the charging field as described in any one of claims 1 to 5.

Images