A method, device, equipment and storage medium for simulating and predicting time series random signals based on KAN network

Through the KAN network-based timing random signal simulation prediction method, the random volatility sequence of the power system is extracted and inputted into the power load prediction model, which solves the problem of low accuracy of power load prediction and improves the accuracy and generalization of power load prediction performance.

CN118693823BActive Publication Date: 2024-11-26CHINA SOUTHERN POWER GRID ARTIFICIAL INTELLIGENCE TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411172275.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-26
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

In the prior art, the accuracy of power load prediction is low, which affects the safe operation and accurate planning of the power system.

Method used

The time-sequence random signal simulation prediction method based on KAN network is adopted. By obtaining the historical load timing data of the power system, the random volatility sequence is extracted as a feature and input it into the pre-trained power load prediction model to improve prediction accuracy.

Benefits of technology

By adding random volatility factors and capturing the dynamic changes in time series data, the power load prediction model adapts to real production environment noise, and improves the accuracy and generalization performance of power load prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118693823B_ABST
    Figure CN118693823B_ABST
Patent Text Reader

Abstract

The present application relates to a method, device, equipment and storage medium for simulating and predicting a time series random signal based on a KAN network. The method includes: obtaining the historical load time series data of the power system within a preset time interval before the prediction time point; inputting the historical load time series data into a pre-trained fluctuation feature extraction model to obtain the fluctuation characteristics corresponding to the historical load time series data; the fluctuation characteristics include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model using a random process to calculate the volatility of the historical load time series data; inputting the historical load time series data and the random volatility sequence into a pre-trained power load prediction model, and outputting the power load prediction data corresponding to the prediction time point; the pre-trained power load prediction model is constructed based on the KAN neural network. The use of this method can improve the accuracy of power load prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, computer-readable storage medium and computer program product for simulating and predicting a time series random signal based on a KAN network. Background Art

[0002] Time series signals are a type of data that is widely present in the real world. Modeling and prediction of time series signals also have a wide range of applications. For example, power load forecasting in the power industry is a typical time series signal prediction technology.

[0003] Power load forecasting refers to the prediction of power load after a preset time length of the power system. Among them, power load forecasting is an important basis for the safe operation and accurate planning of the power system. In related technologies, prediction models are used to predict power loads, but the accuracy of prediction models in related technologies is low, which affects the accuracy of power load forecasting.

[0004] Therefore, there is a problem of low accuracy in power load prediction in the related technology. Summary of the invention

[0005] Based on this, it is necessary to provide a KAN network-based time series random signal simulation prediction method, device, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of power load prediction in response to the above technical problems.

[0006] In a first aspect, the present application provides a time series random signal simulation prediction method based on a KAN network, comprising:

[0007] Obtaining historical load time series data of the power system within a preset time interval before the prediction time point;

[0008] Input the historical load time series data into a pre-trained fluctuation feature extraction model to obtain fluctuation features corresponding to the historical load time series data; the fluctuation features include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model by using a random process to calculate the volatility of the historical load time series data;

[0009] The historical load time series data and the random volatility sequence are input into a pre-trained power load forecasting model, and the power load forecasting data corresponding to the forecast time point is output; the pre-trained power load forecasting model is constructed based on a KAN neural network.

[0010] In one embodiment, the preset time interval includes a plurality of historical time points; the historical load time series data includes a plurality of historical load data corresponding to different historical time points;

[0011] The step of inputting the historical load time series data into a pre-trained fluctuation feature extraction model to obtain the fluctuation features corresponding to the historical load time series data includes:

[0012] Determine the historical load data in the preset time window corresponding to each target historical time point respectively, and obtain the target historical load time series data in the preset time window corresponding to each target historical time point;

[0013] The time length of the preset time interval is greater than or equal to the time length of the preset time window; the preset time window corresponding to each of the target historical time points has the corresponding target historical time point as the right endpoint; the target historical time point is a historical time point whose corresponding time length in the preset time interval is greater than or equal to the time length of the preset time window;

[0014] Performing random fitting on the target historical load time series data corresponding to each of the target historical time points to obtain the random volatility sequence;

[0015] According to the random volatility sequence, the fluctuation characteristics corresponding to the historical load time series data are obtained.

[0016] In one embodiment, performing random fitting on the target historical load time series data corresponding to each target historical time point to obtain the random volatility sequence includes:

[0017] Performing random fitting on the target historical load time series data corresponding to each of the target historical time points to obtain the random volatility corresponding to each of the target historical time points;

[0018] The stochastic volatility sequence is composed according to the stochastic volatility corresponding to each of the target historical time points.

[0019] In one embodiment, the random fitting is performed multiple times; the random fitting is performed on the target historical load time series data corresponding to each target historical time point to obtain the random volatility corresponding to each target historical time point, including:

[0020] For any target historical time point, obtain the random fitting result corresponding to the any target historical time point; the random fitting result includes the random fitting value obtained by performing the random fitting each time;

[0021] According to the randomness fitting value that meets the preset numerical threshold in the randomness fitting result, the random volatility corresponding to any target historical time point is obtained.

[0022] In one embodiment, for any target historical time point, obtaining a random fitting result corresponding to the any target historical time point includes:

[0023] In each process of performing random fitting, the volatility corresponding to any target historical time point, the long-term mean of the volatility corresponding to any target historical time point, the mean reversion speed of the volatility corresponding to any target historical time point, the volatility of the volatility corresponding to any target historical time point, and the Brownian motion corresponding to any target historical time point are obtained;

[0024] According to the volatility, the long-term mean of the volatility, the mean reversion speed of the volatility, the volatility of the volatility and the Brownian motion, a random fitting value corresponding to any target historical time point is obtained.

[0025] In one embodiment, the training process of the pre-trained power load forecasting model includes:

[0026] Acquire sample load time series data; the sample load time series data includes sample load data corresponding to each sample time point;

[0027] Obtain a sample random volatility sequence corresponding to the sample load time series data; the sample random volatility sequence includes a sample random volatility corresponding to each target sample time point; the target sample time point is a sample time point corresponding to a time length in a sample time interval, which is greater than or equal to a time length of a preset time window; the sample time interval is a time interval corresponding to the sample load time series data;

[0028] According to the sample load time series data and the sample random volatility sequence, the power load forecasting model to be trained is iteratively trained to obtain the pre-trained power load forecasting model; the power load forecasting model to be trained is constructed based on the KAN neural network.

[0029] In a second aspect, the present application also provides a time series random signal simulation prediction device based on a KAN network, comprising:

[0030] An acquisition module is used to acquire historical load time series data of the power system within a preset time interval before the prediction time point;

[0031] A feature extraction module is used to input the historical load time series data into a pre-trained fluctuation feature extraction model to obtain fluctuation features corresponding to the historical load time series data; the fluctuation features include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model by using a random process to calculate the volatility of the historical load time series data;

[0032] A prediction module is used to input the historical load time series data and the random volatility sequence into a pre-trained power load prediction model, and output the power load prediction data corresponding to the prediction time point; the pre-trained power load prediction model is constructed based on the KAN neural network.

[0033] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the above method are implemented.

[0034] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0035] In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0036] The above-mentioned KAN network-based time series random signal simulation prediction method, device, computer equipment, computer-readable storage medium and computer program product obtain the historical load time series data of the power system within a preset time interval before the prediction time point; input the historical load time series data into a pre-trained fluctuation feature extraction model to obtain the fluctuation characteristics corresponding to the historical load time series data; the fluctuation characteristics include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model by calculating the volatility of the historical load time series data using a random process; the historical load time series data and the random volatility sequence are input into a pre-trained power load prediction model, and the power load prediction data corresponding to the prediction time point is output; the pre-trained power load prediction model is constructed based on the KAN neural network.

[0037] Thus, in terms of factor feature extraction, a random process is used to simulate and model the time series signal of the historical load time series data, and the corresponding fluctuation characteristics are extracted. The fluctuation characteristics include a random volatility sequence, so as to add a one-dimensional factor "random volatility" to the time series signal. This calculation of the random volatility of the time series data can better capture the dynamic changes of the time series data. In addition to increasing the dimension of the original time series data, it can also simulate the randomness of the original time series data in small-scale cycles to a certain extent. Therefore, in the process of inputting the historical load time series data and the random volatility sequence into the pre-trained power load forecasting model and outputting the power load forecasting data corresponding to the forecast time point, the adaptability of the power load forecasting model to the noise in the real production environment is increased, the generalization performance of the power load forecasting model is improved, and the accuracy of the power load forecasting is effectively improved.

[0038] In addition, the pre-trained power load forecasting model is constructed based on the KAN neural network. In this way, the KAN network is used to replace the traditional neural network. Instead of using a fixed neuron activation function, a learnable and iterative function is used to replace the fixed neuron function to fit the time series data and its random volatility, and provide a certain degree of interpretability. Therefore, this application improves the generalization ability and accuracy of the prediction of power load time series data under real conditions from the two directions of factor composition and model optimization, and effectively improves the accuracy of power load prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A schematic diagram of a flow chart of a time series random signal simulation prediction method based on a KAN network in one embodiment;

[0041] Figure 2 A schematic diagram of a flow chart of a step of extracting fluctuation characteristics corresponding to historical load time series data in an embodiment;

[0042] Figure 3 A schematic diagram of a flow chart of a time series random signal simulation prediction method based on a KAN network in another embodiment;

[0043] Figure 4 It is a structural block diagram of a time series random signal simulation prediction device based on a KAN network in one embodiment;

[0044] FIG. 5 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0046] In one embodiment, Figure 1 As shown, a time series random signal simulation prediction method based on a KAN network is provided. This embodiment uses the method applied to a computer device as an example. In this embodiment, the method includes the following steps:

[0047] Step S110, obtaining historical load time series data of the power system within a preset time interval before the prediction time point.

[0048] The prediction time point may be a time point at which power load prediction is required.

[0049] The preset time interval includes multiple historical time points before the predicted time point.

[0050] The historical load time series data within the preset time interval is composed of the historical load data corresponding to each historical time point.

[0051] For example, taking the prediction time point as April 2 of a certain year, the computer device can obtain the historical load data of the power system from March 1 to April 1 of a certain year to obtain the historical load time series data.

[0052] Step S120, inputting the historical load time series data into a pre-trained fluctuation feature extraction model to obtain the fluctuation features corresponding to the historical load time series data.

[0053] Among them, the volatility characteristics include random volatility sequences.

[0054] The random volatility sequence is obtained by using a pre-trained volatility feature extraction model to calculate the volatility of historical load time series data using a random process.

[0055] In a specific implementation, a computer device can input historical load time series data into a pre-trained fluctuation feature extraction model, and use the pre-trained fluctuation feature extraction model to calculate the volatility of the historical load time series data using a random process to obtain the fluctuation characteristics corresponding to the historical load time series data, which may include a random volatility sequence.

[0056] Among them, the pre-trained fluctuation feature extraction model can be a Heston model.

[0057] Step S130, inputting the historical load time series data and the random volatility sequence into the pre-trained power load forecasting model, and outputting the power load forecasting data corresponding to the forecast time point.

[0058] Among them, the pre-trained power load forecasting model is constructed based on the KAN (Kolmogorov-Arnold Networks) neural network.

[0059] In a specific implementation, the computer device can input the historical load time series data and the random volatility sequence into a pre-trained power load forecasting model, fit the historical load time series data and the random volatility sequence through the pre-trained power load forecasting model, and output the power load forecasting data corresponding to the forecast time point.

[0060] Among them, the power load forecast data can be a power load forecast value.

[0061] In the above-mentioned KAN network-based time series random signal simulation prediction method, the historical load time series data of the power system within a preset time interval before the prediction time point is obtained; the historical load time series data is input into a pre-trained fluctuation feature extraction model to obtain the fluctuation characteristics corresponding to the historical load time series data; the fluctuation characteristics include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model by using a random process to calculate the volatility of the historical load time series data; the historical load time series data and the random volatility sequence are input into a pre-trained power load prediction model, and the power load prediction data corresponding to the prediction time point is output; the pre-trained power load prediction model is constructed based on the KAN neural network.

[0062] Thus, in terms of factor feature extraction, a random process is used to simulate and model the time series signal of the historical load time series data, and the corresponding fluctuation characteristics are extracted. The fluctuation characteristics include a random volatility sequence, so as to add a one-dimensional factor "random volatility" to the time series signal. This calculation of the random volatility of the time series data can better capture the dynamic changes of the time series data. In addition to increasing the dimension of the original time series data, it can also simulate the randomness of the original time series data in small-scale cycles to a certain extent. Therefore, in the process of inputting the historical load time series data and the random volatility sequence into the pre-trained power load forecasting model and outputting the power load forecasting data corresponding to the forecast time point, the adaptability of the power load forecasting model to the noise in the real production environment is increased, the generalization performance of the power load forecasting model is improved, and the accuracy of the power load forecasting is effectively improved.

[0063] In addition, the pre-trained power load forecasting model is constructed based on the KAN neural network. In this way, the KAN network is used to replace the traditional neural network. Instead of using a fixed neuron activation function, a learnable and iterative function is used to replace the fixed neuron function to fit the time series data and its random volatility, and provide a certain degree of interpretability. Therefore, this application improves the generalization ability and accuracy of the prediction of power load time series data under real conditions from the two directions of factor composition and model optimization, and effectively improves the accuracy of power load prediction.

[0064] In one embodiment, the preset time interval includes multiple historical time points; the historical load time series data includes multiple historical load data corresponding to different historical time points. Figure 2 As shown, step S120, inputting the historical load time series data into the pre-trained fluctuation feature extraction model to obtain the fluctuation feature corresponding to the historical load time series data, includes the following steps:

[0065] Step S210, respectively determine the historical load data in the preset time window corresponding to each target historical time point, and obtain the target historical load time series data in the preset time window corresponding to each target historical time point.

[0066] The time length of the preset time interval is greater than or equal to the time length of the preset time window.

[0067] Among them, the preset time window corresponding to each target historical time point has the corresponding target historical time point as the right endpoint.

[0068] The target historical time point is a historical time point whose corresponding time length in the preset time interval is greater than or equal to the time length of the preset time window.

[0069] The target historical time point is the time length corresponding to the preset time interval, which can be used to characterize the order of the target historical time point in the preset time interval. It can be understood that the historical load time series data within the preset time interval is obtained by sorting the historical load data from early to late according to the corresponding historical time points.

[0070] For example, if the time length of the preset time window is 30 time units, then the time length of the preset time interval is at least 30 time units. For example, if the time length of the preset time window is 30 days, if each day corresponds to a load data, then the historical load time series data must contain at least 30 historical load data. If the prediction time point is April 2 of a certain year, it is assumed that the historical load time series data in the preset time interval includes the historical load data from March 1 to April 1 of a certain year. Since the time length corresponding to March 30 in the preset time interval is 30, which is equal to the time length of the preset time window, March 30 can be used as the target historical time point. Correspondingly, the time length corresponding to March 31 in the preset time interval is 31, which is greater than the time length of the preset time window, so March 31 can be used as the target historical time point; the time length corresponding to April 1 in the preset time interval is 32, which is greater than the time length of the preset time window, so April 1 can be used as the target historical time point.

[0071] Correspondingly, the preset time window corresponding to March 30 is March 1 to March 30; the preset time window corresponding to March 31 is March 2 to March 31; and the preset time window corresponding to April 1 is March 3 to April 1.

[0072] Correspondingly, the target historical load time series data within the preset time window corresponding to March 30 includes the historical load data from March 1 to March 30; the target historical load time series data within the preset time window corresponding to March 31 includes the historical load data from March 2 to March 31; the target historical load time series data within the preset time window corresponding to April 1 includes the historical load data from March 3 to April 1.

[0073] Step S220, performing random fitting on the target historical load time series data corresponding to each target historical time point to obtain a random volatility sequence.

[0074] In the specific implementation, the pre-trained fluctuation feature extraction model can perform random fitting on the target historical load time series data corresponding to each target historical time point to obtain a random volatility sequence.

[0075] Step S230, obtaining the fluctuation characteristics corresponding to the historical load time series data according to the random volatility sequence.

[0076] In the specific implementation, the pre-trained fluctuation feature extraction model can obtain the fluctuation features corresponding to the historical load time series data according to the random volatility sequence.

[0077] In some embodiments, the pre-trained fluctuation feature extraction model randomly fits the target historical load time series data corresponding to each target historical time point to obtain a random volatility sequence, including: the pre-trained fluctuation feature extraction model randomly fits the target historical load time series data corresponding to each target historical time point to obtain the random volatility corresponding to each target historical time point; the pre-trained fluctuation feature extraction model composes a random volatility sequence according to the random volatility corresponding to each target historical time point.

[0078] In some embodiments, the number of fittings of random fitting is multiple times; random fitting is performed on the target historical load time series data corresponding to each target historical time point to obtain the random volatility corresponding to each target historical time point, including: for any target historical time point, obtaining the random fitting result corresponding to any target historical time point; the random fitting result includes the random fitting value obtained each time the random fitting is performed; according to the random fitting value in the random fitting result that meets the preset numerical threshold, the random volatility corresponding to any target historical time point is obtained.

[0079] Among them, the number of fitting times of random fitting is multiple times. In the process of performing random fitting on the target historical load time series data corresponding to each target historical time point by the pre-trained fluctuation feature extraction model to obtain the random volatility corresponding to each target historical time point, for any target historical time point, the pre-trained fluctuation feature extraction model can obtain the random fitting result corresponding to the any target historical time point, and the random fitting result includes the random fitting value obtained by each random fitting; in this way, the pre-trained fluctuation feature extraction model can obtain the random volatility corresponding to any target historical time point according to the random fitting value that meets the preset numerical threshold in the random fitting result, and the random volatility is the random volatility of the preset time window corresponding to the any target historical time point.

[0080] In some embodiments, the random volatility corresponding to any target historical time point is obtained based on the random fitting value that meets the preset numerical threshold in the random fitting result, including: when there are multiple random fitting values ​​that meet the preset numerical threshold in the random fitting result, each random fitting value that meets the preset numerical threshold is used as a candidate random fitting value; the candidate random fitting value that is most recently fitted is selected from each candidate random fitting value as the target random fitting value; and the random volatility corresponding to any target historical time point is obtained based on the target random fitting value.

[0081] In a specific implementation, in the process of obtaining the random volatility corresponding to any target historical time point based on the random fitting value that meets the preset numerical threshold in the random fitting result, the pre-trained fluctuation feature extraction model, when there are multiple random fitting values ​​that meet the preset numerical threshold in the random fitting result, the pre-trained fluctuation feature extraction model can use each random fitting value that meets the preset numerical threshold as a candidate random fitting value; select the most recently fitted candidate random fitting value from each candidate random fitting value as the target random fitting value; and obtain the random volatility corresponding to any target historical time point based on the target random fitting value.

[0082] In this way, based on the same method, the stochastic volatility corresponding to each target historical time point can be obtained.

[0083] In some embodiments, for any target historical time point, a randomness fitting result corresponding to any target historical time point is obtained, including: in each process of randomness fitting, the volatility corresponding to any target historical time point, the long-term mean of the volatility corresponding to any target historical time point, the mean reversion speed of the volatility corresponding to any target historical time point and the volatility of the volatility corresponding to any target historical time point, as well as the Brownian motion corresponding to any target historical time point are obtained; the randomness fitting value corresponding to any target historical time point is obtained according to the volatility, the long-term mean of the volatility, the mean reversion speed of the volatility, the volatility of the volatility and the Brownian motion.

[0084] In a specific implementation, in the process of obtaining the randomness fitting result corresponding to any target historical time point, in each process of randomness fitting, the volatility corresponding to any target historical time point, the long-term mean of the volatility corresponding to any target historical time point, the mean reversion speed of the volatility corresponding to any target historical time point, the volatility of the volatility corresponding to any target historical time point, and the Brownian motion corresponding to any target historical time point can be obtained. Therefore, the randomness fitting value corresponding to any target historical time point can be obtained based on the volatility corresponding to any target historical time point, the long-term mean of the volatility corresponding to any target historical time point, the mean reversion speed of the volatility corresponding to any target historical time point, the volatility of the volatility corresponding to any target historical time point, and the Brownian motion corresponding to any target historical time point, so that the random process is more stable.

[0085] Taking the Heston model as an example, the volatility dynamic process of the target historical load time series data corresponding to each target historical time point is as follows during the random fitting of the target historical load time series data:

[0086]

[0087] in, is the volatility corresponding to the target historical time point t; is the long-term average volatility corresponding to the target historical time point t (calculated based on the volatility corresponding to each historical time point within the preset time window corresponding to the target historical time point t), is the mean reversion speed of the volatility corresponding to the target historical time point t (calculated based on the volatility corresponding to each historical time point within the preset time window corresponding to the target historical time point t), is the volatility of the volatility corresponding to the target historical time point t (i.e., the volatility of the volatility sequence corresponding to the target historical time point t is calculated again), is the Brownian motion corresponding to the target historical time point, is a standard Brownian motion used to simulate randomness.

[0088] In this way, the randomness fitting value obtained by each randomness fitting can be calculated according to the calculation formula of the volatility dynamic process.

[0089] In practical applications, 1,000 random fittings can be performed for each target historical time point. Among the random fitting results corresponding to each target historical time point, the last random fitting value greater than 0 is taken as the random volatility of the preset time window corresponding to the target historical time point.

[0090] In this way, when doing data preprocessing, the calculation of the random volatility of the time series data is added, which can better capture the dynamic changes of the original time series data. In addition to increasing the dimension of the original time series data, it can also simulate the randomness of the original time series data in small-scale cycles to a certain extent, increasing the adaptability of the power load forecasting model to the noise in the real production environment, thereby increasing the generalization performance of the power load forecasting model.

[0091] In one embodiment, the training process of the pre-trained power load prediction model includes: obtaining sample load time series data; the sample load time series data includes sample load data corresponding to each sample time point; obtaining a sample random volatility sequence corresponding to the sample load time series data; iteratively training the power load prediction model to be trained according to the sample load time series data and the sample random volatility sequence to obtain a pre-trained power load prediction model; the power load prediction model to be trained is constructed based on the KAN neural network.

[0092] The sample random volatility sequence includes the sample random volatility corresponding to each target sample time point.

[0093] The target sample time point is a sample time point whose corresponding time length in the sample time interval is greater than or equal to the time length of the preset time window.

[0094] The sample time interval is the time interval corresponding to the sample load time series data.

[0095] Among them, the torch deep learning framework can be used to build a KAN model to build a power load forecasting model to be trained. The activation function of the KAN model is on the "edge" of the neural network, not on the node, and these activation functions are not fixed but can be learned. The specific method is to understand the training process from raw data to labels as a multivariate function, and then split the function into the accumulation of multiple single-variable functions, and define two functions of curve->coefficient and coefficient->curve to model the single-variable function, and iterate the coefficient through the feedforward network to achieve the purpose of activation function learning, so that the result of the accumulation of multiple single-variable functions is close to the function from input to output.

[0096] Due to the characteristic of the KAN model that it trains "edges" rather than "nodes", the parameter scale of the model can be greatly reduced and the training speed of the model can be improved without affecting the accuracy. Correspondingly, if the same parameter scale as the traditional neural network model is maintained, the accuracy of the model will be improved and the training speed will still be faster.

[0097] In this way, the computer device can obtain the sample load time series data. Based on the same method, the computer device can obtain the random volatility corresponding to each target sample time point, and obtain the sample random volatility corresponding to each target sample time point to form a sample random volatility sequence as the sample random volatility sequence corresponding to the sample load time series data. The target sample time point is the sample time point corresponding to the time length in the sample time interval, which is greater than or equal to the time length of the preset time window.

[0098] Thus, the computer device can use the sample load time series data and the sample random volatility sequence to iteratively train the power load forecasting model to be trained, and obtain the pre-trained power load forecasting model. In the training process, the target sample load time series data (including the sample load data in the preset time window corresponding to the target sample time point t') and the corresponding sample random volatility in the preset time window corresponding to the target sample time point t' are used as input data x, which are input into the power load forecasting model to be trained, and the sample load forecasting data corresponding to the next target sample time point t'+1 of t' is obtained, and the root mean square error between the sample load data corresponding to t'+1 and the corresponding sample load forecasting data is calculated to obtain the loss function value, so as to iteratively train the power load forecasting model to be trained.

[0099] After the training is completed, the model weights of the pre-trained power load forecasting model are saved as a pt format file of the torch framework for loading during inference. In this way, when using the pre-trained power load forecasting model for inference prediction, it is necessary to load the pt format weight file of the model, input the historical load time series data within the preset time interval before the prediction time point to the pre-trained power load forecasting model, and the model outputs the power load forecasting data corresponding to the next time point at the right end point of the preset time interval; thus, based on the power load forecasting data corresponding to the next time point, the power load forecasting data corresponding to the next time point can be predicted, until the next time point is the prediction time point.

[0100] In another embodiment, Figure 3 As shown, a time series random signal simulation prediction method based on a KAN network is provided, and the method is applied to a computer device as an example for explanation, including the following steps:

[0101] Step S302, obtaining historical load time series data of the power system within a preset time interval before the prediction time point.

[0102] Step S304, respectively determine the historical load data in the preset time window corresponding to each target historical time point, and obtain the target historical load time series data in the preset time window corresponding to each target historical time point.

[0103] Step S306, performing random fitting on the target historical load time series data corresponding to each target historical time point to obtain the random volatility corresponding to each target historical time point.

[0104] Step S308, composing a random volatility sequence according to the random volatility corresponding to each target historical time point.

[0105] Step S310, inputting historical load time series data and random volatility sequence into a pre-trained power load forecasting model, and outputting power load forecasting data corresponding to the forecast time point.

[0106] It should be noted that the specific limitations of the above steps can refer to the specific limitations of a time series random signal simulation prediction method based on a KAN network mentioned above.

[0107] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0108] Based on the same inventive concept, the embodiment of the present application also provides a KAN network-based temporal random signal simulation prediction device for implementing the above-mentioned KAN network-based temporal random signal simulation prediction method. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the KAN network-based temporal random signal simulation prediction device provided below can refer to the above limitations on the KAN network-based temporal random signal simulation prediction method, which will not be repeated here.

[0109] In an exemplary embodiment, Figure 4 As shown, a time series random signal simulation prediction device based on a KAN network is provided, comprising: an acquisition module 410, a feature extraction module 420 and a prediction module 430, wherein:

[0110] The acquisition module 410 is used to acquire the historical load time series data of the power system within a preset time interval before the prediction time point.

[0111] The feature extraction module 420 is used to input the historical load time series data into a pre-trained fluctuation feature extraction model to obtain fluctuation features corresponding to the historical load time series data; the fluctuation features include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model by using a random process to calculate the volatility of the historical load time series data.

[0112] The prediction module 430 is used to input the historical load time series data and the random volatility sequence into a pre-trained power load prediction model, and output the power load prediction data corresponding to the prediction time point; the pre-trained power load prediction model is constructed based on the KAN neural network.

[0113] In one embodiment, the preset time interval includes multiple historical time points; the historical load time series data includes multiple historical load data corresponding to different historical time points; the feature extraction module 420 is specifically used to determine the historical load data in the preset time window corresponding to each target historical time point, and obtain the target historical load time series data in the preset time window corresponding to each target historical time point; the time length of the preset time interval is greater than or equal to the time length of the preset time window; the preset time window corresponding to each target historical time point has the corresponding target historical time point as the right endpoint; the target historical time point is a historical time point whose corresponding time length in the preset time interval is greater than or equal to the time length of the preset time window; the target historical time point is a historical time point corresponding to each target historical time point and has a time length greater than or equal to the time length of the preset time window; random fitting is performed on the target historical load time series data corresponding to each target historical time point to obtain the random volatility sequence; according to the random volatility sequence, the fluctuation characteristics corresponding to the historical load time series data are obtained.

[0114] In one of the embodiments, the feature extraction module 420 is specifically used to perform random fitting on the target historical load time series data corresponding to each of the target historical time points to obtain the random volatility corresponding to each of the target historical time points; and to form the random volatility sequence based on the random volatility corresponding to each of the target historical time points.

[0115] In one embodiment, the stochastic fitting is performed multiple times; the feature extraction module 420 is specifically used to obtain, for any target historical time point, a stochastic fitting result corresponding to the any target historical time point; the stochastic fitting result includes a stochastic fitting value obtained each time the stochastic fitting is performed; and the stochastic volatility corresponding to the any target historical time point is obtained according to the stochastic fitting value in the stochastic fitting result that meets a preset value threshold.

[0116] In one embodiment, the feature extraction module 420 is specifically used to obtain the volatility corresponding to any target historical time point, the long-term mean of the volatility corresponding to any target historical time point, the mean reversion speed of the volatility corresponding to any target historical time point, the volatility of the volatility corresponding to any target historical time point, and the Brownian motion corresponding to any target historical time point during each random fitting process; and obtain the random fitting value corresponding to any target historical time point according to the volatility, the long-term mean of the volatility, the mean reversion speed of the volatility, the volatility of the volatility and the Brownian motion.

[0117] In one of the embodiments, the device also includes: a training module, used to obtain sample load time series data; the sample load time series data includes sample load data corresponding to each sample time point; obtain a sample random volatility sequence corresponding to the sample load time series data; the sample random volatility sequence includes a sample random volatility corresponding to each target sample time point; the target sample time point is a sample time point corresponding to a time length in a sample time interval, which is greater than or equal to the time length of a preset time window; the sample time interval is a time interval corresponding to the sample load time series data; according to the sample load time series data and the sample random volatility sequence, the power load forecasting model to be trained is iteratively trained to obtain the pre-trained power load forecasting model; the power load forecasting model to be trained is constructed based on a KAN neural network.

[0118] Each module in the above-mentioned KAN network-based time series random signal simulation prediction device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0119] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG5. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store model parameter data of a pre-trained power load prediction model. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a time series random signal simulation prediction method based on a KAN network is implemented.

[0120] Those skilled in the art will understand that the structure shown in FIG. 5 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0121] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0122] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0123] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0124] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0125] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0126] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0127] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A time series random signal simulation prediction method based on KAN network, characterized in that: The method comprises: Obtaining historical load time series data of the power system within a preset time interval before the prediction time point; The historical load time series data is input into a pre-trained fluctuation feature extraction model to obtain fluctuation features corresponding to the historical load time series data; the fluctuation features include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model when performing volatility calculation on the historical load time series data using a random process; the random volatility sequence is composed of random volatility corresponding to each target historical time point; the random volatility corresponding to each target historical time point is obtained by performing random fitting on the target historical load time series data within a preset time window corresponding to each target historical time point; the target historical time point is a historical time point whose corresponding time length in the preset time interval is greater than or equal to the time length of the preset time window; The historical load time series data and the random volatility sequence are input into a pre-trained power load forecasting model, and the power load forecasting data corresponding to the forecast time point is output; the pre-trained power load forecasting model is constructed based on a KAN neural network.

2. The method according to claim 1, characterized in that The preset time interval includes a plurality of historical time points; the historical load time series data includes a plurality of historical load data corresponding to different historical time points; The step of inputting the historical load time series data into a pre-trained fluctuation feature extraction model to obtain the fluctuation features corresponding to the historical load time series data includes: Determine the historical load data in the preset time window corresponding to each target historical time point respectively, and obtain the target historical load time series data in the preset time window corresponding to each target historical time point; The time length of the preset time interval is greater than or equal to the time length of the preset time window; the preset time window corresponding to each of the target historical time points has the corresponding target historical time point as the right endpoint; the target historical time point is a historical time point whose corresponding time length in the preset time interval is greater than or equal to the time length of the preset time window; Performing random fitting on the target historical load time series data corresponding to each of the target historical time points to obtain the random volatility sequence; According to the random volatility sequence, the fluctuation characteristics corresponding to the historical load time series data are obtained.

3. The method according to claim 2, characterized in that The step of performing random fitting on the target historical load time series data corresponding to each target historical time point to obtain the random volatility sequence includes: Performing random fitting on the target historical load time series data corresponding to each of the target historical time points to obtain the random volatility corresponding to each of the target historical time points; The stochastic volatility sequence is composed according to the stochastic volatility corresponding to each of the target historical time points.

4. The method according to claim 3, characterized in that The number of fitting times of the random fitting is multiple times; the random fitting of the target historical load time series data corresponding to each target historical time point is performed to obtain the random volatility corresponding to each target historical time point, including: For any target historical time point, obtain the random fitting result corresponding to the any target historical time point; the random fitting result includes the random fitting value obtained by performing the random fitting each time; According to the randomness fitting value that meets the preset numerical threshold in the randomness fitting result, the random volatility corresponding to any target historical time point is obtained.

5. The method according to claim 4, characterized in that The obtaining, for any target historical time point, a random fitting result corresponding to the any target historical time point includes: In each process of performing random fitting, the volatility corresponding to any target historical time point, the long-term mean of the volatility corresponding to any target historical time point, the mean reversion speed of the volatility corresponding to any target historical time point, the volatility of the volatility corresponding to any target historical time point, and the Brownian motion corresponding to any target historical time point are obtained; According to the volatility, the long-term mean of the volatility, the mean reversion speed of the volatility, the volatility of the volatility and the Brownian motion, a random fitting value corresponding to any target historical time point is obtained.

6. The method according to any one of claims 1 to 5, characterized in that: The training process of the pre-trained power load forecasting model includes: Acquire sample load time series data; the sample load time series data includes sample load data corresponding to each sample time point; Obtain a sample random volatility sequence corresponding to the sample load time series data; the sample random volatility sequence includes a sample random volatility corresponding to each target sample time point; the target sample time point is a sample time point corresponding to a time length in a sample time interval, which is greater than or equal to a time length of a preset time window; the sample time interval is a time interval corresponding to the sample load time series data; According to the sample load time series data and the sample random volatility sequence, the power load forecasting model to be trained is iteratively trained to obtain the pre-trained power load forecasting model; the power load forecasting model to be trained is constructed based on the KAN neural network.

7. A time series random signal simulation prediction device based on KAN network, characterized in that: The device comprises: An acquisition module is used to acquire historical load time series data of the power system within a preset time interval before the prediction time point; A feature extraction module is used to input the historical load time series data into a pre-trained fluctuation feature extraction model to obtain fluctuation features corresponding to the historical load time series data; the fluctuation features include a random volatility sequence; the random volatility sequence is obtained by the pre-trained fluctuation feature extraction model when performing volatility calculation on the historical load time series data using a random process; the random volatility sequence is composed of random volatility corresponding to each target historical time point; the random volatility corresponding to each target historical time point is obtained by performing random fitting on the target historical load time series data within a preset time window corresponding to each target historical time point; the target historical time point is a historical time point whose corresponding time length in the preset time interval is greater than or equal to the time length of the preset time window; A prediction module is used to input the historical load time series data and the random volatility sequence into a pre-trained power load prediction model, and output the power load prediction data corresponding to the prediction time point; the pre-trained power load prediction model is constructed based on the KAN neural network.

8. 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 method according to any one of claims 1 to 6 are implemented.

9. 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 method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.