Method for dynamically predicting imbalance of power market system based on multiple artificial intelligence models
Through dynamic prediction methods based on multiple artificial intelligence models, the problem of dynamic changes in the power market system and the difficult nonlinear relationships are solved, and higher prediction accuracy and decision-making support capabilities are achieved.
Patent Information
- Application Number
- CN202411442917.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-16
- Publication Date
- 2025-02-07
AI Technical Summary
The existing technology is difficult to accurately capture the dynamic changes and complex nonlinear relationships of the power market system, and is not convenient to provide support for the stable operation and optimization decisions of the power market.
Using dynamic prediction methods based on multiple artificial intelligence models, including data collection and preprocessing, training and prediction of multiple artificial intelligence models, cross-validation and hyperparameter selection, model result fusion and other steps, we obtain the weighted average and ensemble learning methods. The final prediction result.
It improves the accuracy of prediction of power market system imbalance, can process new data in real time, dynamically adjust the prediction model, adapt to the continuous changes in power market system, and provide better decision-making basis.
Smart Images
Figure CN119398968A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power market, and in particular to a method for dynamically predicting imbalance of power market system based on multiple artificial intelligence models. Background Art
[0002] The electricity market is the process by which buyers and sellers of electricity interact to determine the price and quantity of electricity. It is mainly composed of independent system operators, power generation companies, transmission companies, distribution companies and users. With the continuous development and reform of the electricity market, the complexity and uncertainty of the power system are increasing.
[0003] The imbalance of the power market system may lead to insufficient or excessive power supply, affecting the stability and reliability of the power system. Traditional forecasting methods often find it difficult to accurately capture the dynamic changes and complex nonlinear relationships of the power market system, making it difficult to provide support for the stable operation and optimal decision-making of the power market. Summary of the invention
[0004] The purpose of the present invention is to provide a method for dynamically predicting the imbalance of the power market system based on multiple artificial intelligence models, aiming to solve the technical problem that the prediction methods in the prior art are often difficult to accurately capture the dynamic changes and complex nonlinear relationships of the power market system, and are not convenient for providing support for the stable operation and optimization decision-making of the power market.
[0005] To achieve the above object, the present invention adopts a method for dynamically predicting the imbalance of the power market system based on multiple artificial intelligence models, comprising the following steps:
[0006] Step 1: First, collect relevant data of the power market system and pre-process it;
[0007] Step 2: Use multiple artificial intelligence models to train and predict the preprocessed data, and use cross-validation to evaluate and select model hyperparameters;
[0008] Step 3: Fusion the prediction results of the model, using weighted average and ensemble learning methods to obtain the final prediction results.
[0009] Among them, in step one, the collected data include power supply and demand data, price data, weather data, and holiday data.
[0010] Among them, in step 1, the method of preprocessing the collected relevant data of the power market system includes data cleaning and normalization;
[0011] The data is cleaned to remove outliers and missing values.
[0012] Among them, the methods of removing missing values include direct deletion and interpolation;
[0013] Direct deletion method is used: if the proportion of missing values in the data set is low, the rows or columns containing missing values are directly deleted;
[0014] Interpolation method: For time series data or data with a certain order, use linear interpolation. Assume that the two data points before and after the missing value are A and B, the time points are t1 and t2, and the time point of the missing value is t; then the missing value can be calculated by the following formula:
[0015] Missing value = A + ((BA)*(t-t1)) / (t2-t1);
[0016] For numeric data, the mean of the column is used to fill the missing values; first calculate the mean of the non-missing values in the column, and then assign it to the missing values;
[0017] If there are other variables related to missing values in the data set, a regression model can be built to predict missing values.
[0018] Among them, the methods of removing outliers include statistical methods and machine learning methods;
[0019] In statistical methods:
[0020] The method of removing outliers based on the mean and standard deviation is as follows: for data that obeys the normal distribution, the mean and standard deviation of the data are calculated, and the data points that are more than a certain multiple of the standard deviation from the mean are regarded as outliers;
[0021] The box plot method is used to remove outliers: the box plot can intuitively display the distribution of data; outliers are defined as data points that are less than the lower quartile minus 1.5 times the interquartile range or greater than the upper quartile plus 1.5 times the interquartile range, and the interquartile range is equal to greater than the upper quartile minus less than the lower quartile;
[0022] In machine learning methods:
[0023] The method of removing outliers using the isolation forest algorithm is to quickly isolate outliers by randomly splitting the data space;
[0024] The way to remove outliers using the local outlier factor algorithm is as follows: The LOF algorithm detects outliers by calculating the local outlier factor of each data point.
[0025] Among them, in step 2, the specific method of using multiple artificial intelligence models to train the preprocessed data is:
[0026] Data preparation;
[0027] Build the model architecture;
[0028] Determine the loss function and optimizer;
[0029] Training process.
[0030] Wherein, in step 2, the cross-validation method includes K-fold cross-validation and leave-one-out cross-validation;
[0031] The specific method of using K-fold cross validation is: randomly divide the data set into K subsets of equal size, usually called folds; for each hyperparameter combination, select one of the folds as the validation set and the remaining K-1 folds as the training set for model training and validation; repeat the above process K times so that each fold has the opportunity to be used as a validation set; finally, calculate the average of the K validation results as the performance evaluation indicator of the hyperparameter combination;
[0032] The specific method of using leave-one-out cross-validation is as follows: leave-one-out cross-validation is a special case of K-fold cross-validation, where K is equal to the number of samples in the data set; only one sample is left as the validation set each time, and the remaining samples are used as training sets for model training and validation; this process is repeated until each sample has the opportunity to be used as a validation set; finally, the average of all validation results is calculated as the performance evaluation indicator of the model.
[0033] Among them, the methods for selecting the optimal hyperparameter combination include grid search and random search;
[0034] The specific method of grid search is as follows: first determine the range and step size of the hyperparameters that need to be adjusted, combine the values of different hyperparameters to form a hyperparameter grid, and for each hyperparameter combination, use the cross-validation method to evaluate the performance of the model. Finally, select the hyperparameter combination with the best performance as the optimal solution.
[0035] The specific method when using random search is: randomly sample in the hyperparameter space, use the cross-validation method to evaluate the performance of the model for each randomly sampled hyperparameter combination, repeat this process multiple times, and finally select the hyperparameter combination with the best performance as the optimal solution.
[0036] Among them, in step 3, when the weighted average method is adopted: the basis for determining the weight is the model performance and the historical prediction accuracy. The model performance is determined by evaluating the performance of each model on the validation set, and the historical prediction accuracy is calculated based on the prediction records of historical data to calculate the prediction accuracy of the model in the past period of time;
[0037] The method of calculating weights is as follows: first calculate the performance index value or historical prediction accuracy of each model, and then normalize these values so that the sum of all weights is 1;
[0038] The process of weighted averaging is as follows: for each time point or sample to be predicted, the prediction results of each model are obtained respectively.
[0039] Among them, in step three, the ensemble learning methods used include random forest ensemble and Adaboost ensemble;
[0040] When using random forest ensemble, the specific method is as follows: for the preprocessed data, randomly select a part of samples and features to build a decision tree; repeat this process multiple times to build multiple different decision trees;
[0041] The prediction method is as follows: for new input data, prediction is made through each decision tree, and then the prediction results of all decision trees are counted; if it is a classification problem, the final category is determined by majority voting; if it is a regression problem, the average of the prediction results of all decision trees is calculated as the final prediction value;
[0042] When using the Adaboost ensemble method, the specific method is as follows: first, give each sample the same initial weight;
[0043] Then, the weak learner is trained iteratively, and in each round of iteration, a weak learner is trained using the current sample weights, and then the sample weights are adjusted based on the prediction results of the weak learner;
[0044] Finally, the weak learners are combined: after multiple rounds of iterations, multiple weak learners are obtained, and these weak learners are linearly combined to obtain the final strong learner; the weight of the combination is determined according to the performance of each weak learner, and the weak learner with better performance is given a greater weight.
[0045] The present invention provides a method for dynamically predicting the imbalance of the power market system based on multiple artificial intelligence models. By integrating the advantages of multiple artificial intelligence models, it can more comprehensively capture the characteristics and laws of the power market system, thereby improving the accuracy of the prediction. It can process new data in real time, dynamically adjust the prediction model, and adapt to the continuous changes in the power market system. The accurate prediction results can provide a decision-making basis for participants in the power market, such as power market transactions, production plans of power generation enterprises, and scheduling strategies of power supply enterprises. In this way, the technical problem that the prediction methods in the prior art are often difficult to accurately capture the dynamic changes and complex nonlinear relationships of the power market system, and are not convenient for providing support for the stable operation and optimization decision-making of the power market is solved. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0047] Figure 1 It is a flow chart of the steps of the method for dynamically predicting the imbalance of the power market system based on multiple artificial intelligence models of the present invention. DETAILED DESCRIPTION
[0048] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.
[0049] See also Figure 1 , Figure 1 It is a flow chart of the steps of the method for dynamically predicting the imbalance of the power market system based on multiple artificial intelligence models of the present invention.
[0050] The present invention provides a method for dynamically predicting power market system imbalance based on multiple artificial intelligence models, comprising the following steps:
[0051] S101, first collect relevant data of the power market system;
[0052] For this specific implementation, relevant data are collected from the database, monitoring system and other channels of the power market to ensure the integrity and accuracy of the data. The time span of the collected data should be long enough to cover different market conditions and seasonal changes;
[0053] S102, preprocessing the collected data related to the power market system;
[0054] According to this specific implementation method, the method of preprocessing the collected relevant data of the power market system includes data cleaning and normalization, normalizing the data and mapping the data to the [0,1] interval to improve the training efficiency and accuracy of the model;
[0055] The data is cleaned to remove outliers and missing values.
[0056] Ways to remove missing values include direct deletion and interpolation;
[0057] Direct deletion method: If the proportion of missing values in the data set is low, directly delete the rows or columns containing missing values. When deleting rows, it is necessary to evaluate the impact of deletion on the overall data distribution and model training.
[0058] Interpolation method: For time series data or data with a certain order, use linear interpolation. Assume that the two data points before and after the missing value are A and B, the time points are t1 and t2, and the time point of the missing value is t; then the missing value can be calculated by the following formula:
[0059] Missing value = A + ((BA)*(t-t1)) / (t2-t1);
[0060] For numeric data, the mean of the column is used to fill the missing values; first calculate the mean of the non-missing values in the column, and then assign it to the missing values;
[0061] If there are other variables related to missing values in the data set, a regression model can be built to predict missing values.
[0062] Ways to remove outliers include statistical methods and machine learning methods;
[0063] In statistical methods:
[0064] The method of removing outliers based on the mean and standard deviation is as follows: for data that obeys the normal distribution, the mean and standard deviation of the data are calculated, and the data points that are more than a certain multiple of the standard deviation from the mean are regarded as outliers;
[0065] The box plot method is used to remove outliers: the box plot can intuitively display the distribution of data; outliers are defined as data points that are less than the lower quartile minus 1.5 times the interquartile range or greater than the upper quartile plus 1.5 times the interquartile range, and the interquartile range is equal to greater than the upper quartile minus less than the lower quartile;
[0066] In machine learning methods:
[0067] The method of removing outliers using the isolation forest algorithm is to quickly isolate outliers by randomly splitting the data space;
[0068] The way to remove outliers using the local outlier factor algorithm is as follows: The LOF algorithm detects outliers by calculating the local outlier factor of each data point.
[0069] Through normalization, data with different features can have the same scale, thereby improving the training efficiency and accuracy of the model. Because many machine learning algorithms are sensitive to the scale of data, if the scale of the data is very different, it may cause the algorithm to prefer certain features and ignore other features during the training process. At the same time, normalized data also helps to speed up the convergence of the algorithm.
[0070] S103, build CNN, RNN, LSTM, LightGBM and other models respectively, and use the preprocessed data for training;
[0071] For this specific implementation, the CNN training method is as follows: the pre-processed power market data is organized into a format suitable for CNN input. For example, if the data is time series data, it can be converted into a two-dimensional image format, where the horizontal axis represents the time dimension and the vertical axis represents different features. Or the data is organized into a three-dimensional tensor format, which is suitable for processing situations with time series and multiple features;
[0072] Design a CNN architecture suitable for power market data. It usually includes convolutional layers, pooling layers, fully connected layers, etc. The convolutional layer is used to extract local features in the data, the pooling layer is used to reduce the data dimension and reduce the amount of calculation, and the fully connected layer is used for the final classification or regression prediction.
[0073] For the prediction problem of power market system imbalance, mean square error (MSE) can be selected as the loss function to measure the difference between the predicted value and the true value. The optimizer can choose the common Adam optimizer, which can adaptively adjust the learning rate and speed up the convergence of the model.
[0074] Divide the preprocessed dataset into training set, validation set and test set. Use the training set to train the CNN model and minimize the loss function by continuously adjusting the weights and biases of the model. During the training process, the validation set can be used regularly to evaluate the performance of the model to prevent overfitting. When the performance of the model on the validation set no longer improves, training can be stopped.
[0075] The training method of RNN is as follows: First, data preparation is performed. Since RNN is suitable for processing time series data, the pre-processed power market data can be directly input into the model in chronological order. The data of each time step can include power supply and demand values, prices, weather and other characteristics.
[0076] Choose a suitable RNN architecture, such as simple RNN, long short-term memory network (LSTM) or gated recurrent unit (GRU). These models can capture long-term dependencies in time series and are very helpful for predicting system imbalances in the power market.
[0077] Then select MSE as the loss function, and the optimizer can be Adam, etc.;
[0078] The training process is to divide the data set into training set, validation set and test set. Use the training set to train the RNN model and update the model parameters through the back propagation algorithm. At each time step, the model receives the input data of the current time step and the hidden state of the previous time step, and outputs the predicted value of the current time step. By continuously adjusting the model parameters, the difference between the predicted value and the true value is minimized. During the training process, the validation set is used to evaluate the model performance to prevent overfitting.
[0079] The training method of LSTM is as follows: First, data preparation is performed. Similar to RNN, LSTM is also suitable for processing time series data, and the data preparation method is the same;
[0080] Then, the model architecture is constructed. LSTM consists of multiple memory cells, each of which includes a forget gate, an input gate, and an output gate. These gating mechanisms can effectively control the flow of information and capture long-term dependencies in time series.
[0081] Then select MSE as the loss function, and the optimizer can be Adam, etc.;
[0082] The specific training process is similar to that of RNN, but LSTM can better handle long-term dependency problems, and may require more time and computing resources during the training process. The training set is also used for training, and the model parameters are updated through the back-propagation algorithm. The performance is evaluated using the validation set to prevent overfitting.
[0083] The training method of LightGBM is as follows: the preprocessed electricity market data is organized into a format suitable for LightGBM input. The data can usually be represented in a tabular form, with each row representing a sample and each column representing a feature.
[0084] LightGBM is a gradient boosting framework based on decision trees, which is efficient and accurate. Before training, you need to set some parameters, such as the number of trees, the depth of the trees, the learning rate, etc.
[0085] LightGBM usually uses mean square error or other regression loss functions as the objective function. It uses the gradient boosting algorithm for optimization, and reduces the loss function by continuously adding new decision trees;
[0086] The dataset is divided into a training set and a test set. The LightGBM model is trained using the training set, and the objective function is optimized by iteratively building a decision tree. In each iteration, the model calculates the gradient of the loss function based on the current prediction result and the true value, and then builds a new decision tree based on the gradient information to update the model's prediction result. The training process continues until the preset stop condition is reached, such as reaching a certain number of iterations or the loss function no longer decreases.
[0087] S104, using a cross-validation method to validate the trained model and select the optimal hyperparameter combination;
[0088] For this specific implementation, the cross-validation method includes K-fold cross-validation and leave-one-out cross-validation; the specific method of using K-fold cross-validation is: randomly divide the data set into K subsets of equal size, usually called folds; for each hyperparameter combination, select one of the folds as the validation set in turn, and the remaining K-1 folds as the training set to perform model training and validation; repeat the above process K times so that each fold has the opportunity to be used as a validation set; finally, calculate the average value of the K validation results as the performance evaluation indicator of the hyperparameter combination;
[0089] The specific method of using leave-one-out cross-validation is as follows: leave-one-out cross-validation is a special case of K-fold cross-validation, where K is equal to the number of samples in the data set; only one sample is left as the validation set each time, and the remaining samples are used as training sets for model training and validation; this process is repeated until each sample has the opportunity to be used as a validation set; finally, the average of all validation results is calculated as the performance evaluation indicator of the model.
[0090] Methods for selecting the optimal hyperparameter combination include grid search and random search;
[0091] The specific method of grid search is as follows: first determine the range and step size of the hyperparameters that need to be adjusted, combine the values of different hyperparameters to form a hyperparameter grid, and for each hyperparameter combination, use the cross-validation method to evaluate the performance of the model. Finally, select the hyperparameter combination with the best performance as the optimal solution.
[0092] The specific method when using random search is: randomly sample in the hyperparameter space, use the cross-validation method to evaluate the performance of the model for each randomly sampled hyperparameter combination, repeat this process multiple times, and finally select the hyperparameter combination with the best performance as the optimal solution.
[0093] S105, using the trained model to predict new data to obtain the prediction results of each model;
[0094] S106. The prediction results of the model are integrated, and the weighted average and ensemble learning methods are used to obtain the final prediction results.
[0095] For this specific implementation, when the weighted average method is adopted: the basis for determining the weight is the model performance and the historical prediction accuracy. The model performance is determined by evaluating the performance of each model on the validation set, and the historical prediction accuracy is calculated based on the prediction records of historical data to calculate the prediction accuracy of the model in the past period of time;
[0096] The method of calculating weights is as follows: first calculate the performance index value or historical prediction accuracy of each model, and then normalize these values so that the sum of all weights is 1;
[0097] The process of weighted averaging is as follows: for each time point or sample to be predicted, the prediction results of each model are obtained respectively.
[0098] The ensemble learning methods used include random forest ensemble and Adaboost ensemble;
[0099] When using random forest ensemble, the specific method is as follows: for the preprocessed data, randomly select a part of samples and features to build a decision tree; repeat this process multiple times to build multiple different decision trees;
[0100] The prediction method is as follows: for new input data, prediction is made through each decision tree, and then the prediction results of all decision trees are counted; if it is a classification problem, the final category is determined by majority voting; if it is a regression problem, the average of the prediction results of all decision trees is calculated as the final prediction value;
[0101] When using the Adaboost ensemble method, the specific method is as follows: first, give each sample the same initial weight;
[0102] Then, the weak learner is trained iteratively, and in each round of iteration, a weak learner is trained using the current sample weights, and then the sample weights are adjusted based on the prediction results of the weak learner;
[0103] Finally, the weak learners are combined: after multiple rounds of iterations, multiple weak learners are obtained, and these weak learners are linearly combined to obtain the final strong learner; the weight of the combination is determined according to the performance of each weak learner, and the weak learner with better performance is given a greater weight.
[0104] Through the two schemes of weighted average and ensemble learning, the prediction results of multiple models can be effectively integrated to improve the accuracy and stability of the prediction.
[0105] S107. Use indicators such as mean square error (MSE) and mean absolute error (MAE) to evaluate the prediction results, and further optimize and improve the model based on the evaluation results.
[0106] Example: Assume that the electricity supply and demand data, price data, weather data and holiday data of a certain region in the past year are collected;
[0107] After preprocessing, CNN, RNN and LSTM models are used for training and prediction respectively;
[0108] In the prediction stage, the prediction results of the three models are: the CNN model predicts that the power imbalance in the next week will be 100MW, the RNN model predicts 120MW, and the LSTM model predicts 110MW. According to the historical accuracy of the three models, the weights are determined to be 0.4, 0.3, and 0.3 respectively. The final fusion prediction result is: 0.4×100+0.3×120+0.3×110=109MW.
[0109] The method of dynamically predicting the imbalance of the power market system based on multiple artificial intelligence models of the present invention can more comprehensively capture the characteristics and laws of the power market system by integrating the advantages of multiple artificial intelligence models, thereby improving the accuracy of the prediction, and can process new data in real time, dynamically adjust the prediction model, and adapt to the continuous changes in the power market system. The accurate prediction results can provide a basis for decision-making for participants in the power market, such as power market transactions, production plans of power generation enterprises, and scheduling strategies of power supply enterprises. In this way, the technical problem that the prediction methods in the prior art are often difficult to accurately capture the dynamic changes and complex nonlinear relationships of the power market system, and are not convenient for providing support for the stable operation and optimization decision-making of the power market is solved.
[0110] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.
Claims
1. A method for dynamically predicting power market system imbalance based on multiple artificial intelligence models, characterized in that: The following steps are involved: Step 1: First, collect relevant data of the power market system and pre-process it; Step 2: Use multiple artificial intelligence models to train and predict the preprocessed data, and use cross-validation to evaluate and select model hyperparameters; Step 3: Fusion the prediction results of the model, using weighted average and ensemble learning methods to obtain the final prediction results.
2. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models according to claim 1, characterized in that: In step one, the collected data include power supply and demand data, price data, weather data, and holiday data.
3. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 2, characterized in that: In step 1, the collected data related to the power market system are preprocessed by means of data cleaning and normalization; The data is cleaned to remove outliers and missing values.
4. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 3, characterized in that: Ways to remove missing values include direct deletion and interpolation; Direct deletion method: If the proportion of missing values in the data set is low, directly delete the rows or columns containing missing values. When deleting rows, it is necessary to evaluate the impact of deletion on the overall data distribution and model training. Interpolation method: For time series data or data with a certain order, use linear interpolation. Assume that the two data points before and after the missing value are A and B, the time points are t1 and t2, and the time point of the missing value is t; then the missing value can be calculated by the following formula: Missing value = A + ((BA)*(t-t1)) / (t2-t1); For numeric data, the mean of the column is used to fill the missing values; first calculate the mean of the non-missing values in the column, and then assign it to the missing values; If there are other variables related to missing values in the data set, a regression model can be built to predict missing values.
5. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 4, characterized in that: Ways to remove outliers include statistical methods and machine learning methods; In statistical methods: The method of removing outliers based on the mean and standard deviation is as follows: for data that obeys the normal distribution, the mean and standard deviation of the data are calculated, and the data points that are more than a certain multiple of the standard deviation from the mean are regarded as outliers; The box plot method is used to remove outliers: the box plot can intuitively display the distribution of data; outliers are defined as data points that are less than the lower quartile minus 1.5 times the interquartile range or greater than the upper quartile plus 1.5 times the interquartile range, and the interquartile range is equal to greater than the upper quartile minus less than the lower quartile; In machine learning methods: The method of removing outliers using the isolation forest algorithm is: quickly isolating outliers by randomly splitting the data space; The way to remove outliers using the local outlier factor algorithm is as follows: The LOF algorithm detects outliers by calculating the local outlier factor of each data point.
6. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 5, characterized in that: In step 2, the specific method of using multiple artificial intelligence models to train the preprocessed data is as follows: Data preparation; Build the model architecture; Determine the loss function and optimizer; Training process.
7. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 6, characterized in that: In step 2, the cross-validation method includes K-fold cross-validation and leave-one-out cross-validation; The specific method of using K-fold cross validation is as follows: randomly divide the data set into K subsets of equal size, called folds; for each hyperparameter combination, select one of the folds as the validation set and the remaining K-1 folds as the training set for model training and validation; repeat the above process K times so that each fold has the opportunity to be used as a validation set; finally, calculate the average of the K validation results as the performance evaluation indicator of the hyperparameter combination; The specific method of using leave-one-out cross validation is as follows: leave-one-out cross validation is a special case of K-fold cross validation, where K is equal to the number of samples in the data set; each time only one sample is left as the validation set, and the remaining samples are used as the training set for model training and validation; This process is repeated until each sample has a chance to serve as a validation set; finally, the average of all validation results is calculated as the performance evaluation indicator of the model.
8. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 7, characterized in that: Methods for selecting the optimal hyperparameter combination include grid search and random search; The specific method of grid search is as follows: first determine the range and step size of the hyperparameters that need to be adjusted, combine the values of different hyperparameters to form a hyperparameter grid, and for each hyperparameter combination, use the cross-validation method to evaluate the performance of the model. Finally, select the hyperparameter combination with the best performance as the optimal solution. The specific method when using random search is: randomly sample in the hyperparameter space, use the cross-validation method to evaluate the performance of the model for each randomly sampled hyperparameter combination, repeat this process multiple times, and finally select the hyperparameter combination with the best performance as the optimal solution.
9. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 8, characterized in that: In step 3, when the weighted average method is used: the basis for determining the weight is model performance and historical prediction accuracy. Model performance is determined by evaluating the performance of each model on the validation set, and historical prediction accuracy is calculated based on the prediction records of historical data to calculate the prediction accuracy of the model in the past period of time; The method of calculating weights is as follows: first calculate the performance index value or historical prediction accuracy of each model, and then normalize these values so that the sum of all weights is 1; The process of weighted averaging is as follows: for each time point or sample to be predicted, the prediction results of each model are obtained respectively.
10. The method for dynamically predicting power market system imbalance based on multiple artificial intelligence models as claimed in claim 9, characterized in that: In step three, the ensemble learning methods used include random forest ensemble and Adaboost ensemble; When using random forest ensemble, the specific method is as follows: for the preprocessed data, randomly select a part of samples and features to build a decision tree; repeat this process multiple times to build multiple different decision trees; The prediction method is as follows: for new input data, prediction is made through each decision tree, and then the prediction results of all decision trees are counted; if it is a classification problem, the final category is determined by majority voting; if it is a regression problem, the average of the prediction results of all decision trees is calculated as the final prediction value; When using the Adaboost ensemble method, the specific method is as follows: first, give each sample the same initial weight; Then, the weak learner is trained iteratively, and in each round of iteration, a weak learner is trained using the current sample weights, and then the sample weights are adjusted based on the prediction results of the weak learner; Finally, the weak learners are combined: after multiple rounds of iterations, multiple weak learners are obtained, and these weak learners are linearly combined to obtain the final strong learner; the weight of the combination is determined according to the performance of each weak learner, and the weak learner with better performance is given a greater weight.