Wind power probability prediction method based on extreme learning machine

Abstract

The invention relates to the technical field of wind power prediction. The invention provides a wind power probability prediction method based on an extreme learning machine, and the method comprises the steps: obtaining wind power historical data, and carrying out the preprocessing of the wind power historical data, and obtaining a preprocessing data set; performing multiple times of random sampling on the preprocessed data set to obtain a sample set, and dividing the sample set into a training set and a verification set; building an initial prediction model through an extreme learning machine based on the sample set; training the initial prediction model through the training set, integrating wind power probability prediction results of the initial prediction model in the training process, and calculating a prediction mean value and a prediction interval of each time point to obtain a prediction model; adjusting and optimizing the prediction interval through the verification set to obtain a prediction optimization model; and predicting the target object through the prediction optimization model, and outputting a final wind power probability prediction result. The problem that an existing traditional neural network is uncertain in wind power generation prediction is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power prediction, and in particular to a wind power probability prediction method based on an extreme learning machine. Background Art

[0002] As an important renewable energy source, the development and utilization of wind power is of great significance to promoting the optimization of energy structure and environmental protection. However, the intermittent and unstable nature of wind power generation poses a challenge to the stable operation of the power grid. Therefore, accurate prediction of wind power generation has become a key link in improving the efficiency of wind energy utilization and ensuring the stable operation of the power grid.

[0003] In the field of wind power generation forecasting, traditional neural network methods have been widely used. However, this method has significant uncertainty in the forecasting process, which mainly comes from two aspects: model uncertainty and data uncertainty.

[0004] Model uncertainty is mainly determined by the structure and parameters of the neural network model. Since the training process of the neural network is prone to fall into local minima, and the randomly generated input weights may also cause the model to be unstable. In addition, even if the model can reach the global minimum, its structural misspecification will still introduce non-negligible uncertainty in the prediction results. More importantly, based on limited sample training data, the neural network cannot guarantee consistent generalization performance in the unforeseeable future, especially in complex and changeable scenarios such as wind power generation, where it is almost impossible to obtain perfect information, which further exacerbates the uncertainty of the model.

[0005] Data uncertainty is also an important factor that affects the prediction results of neural networks. If the data itself has noise or random characteristics, it will become very difficult to model it deterministically. Wind power data often show a high degree of chaos, and this data noise will have a significant impact on the results during the prediction process. Especially when dealing with non-stationary time series, the impact of data noise is particularly prominent. Summary of the invention

[0006] The purpose of the present invention is to provide a wind power probability prediction method based on extreme learning machine, aiming to solve the uncertainty problem existing in the existing traditional neural network in wind power generation prediction.

[0007] The present invention is achieved through the following technical solutions:

[0008] A wind power probability prediction method based on extreme learning machine comprises the following steps:

[0009] Acquire historical wind power data, and preprocess the historical wind power data to obtain a preprocessed data set;

[0010] Perform random sampling on the preprocessed data set several times to obtain a sample set, and divide the sample set into a training set and a validation set;

[0011] Based on the sample set, an initial prediction model is constructed through an extreme learning machine;

[0012] The initial prediction model is trained through the training set, the wind power probability prediction results of the initial prediction model during the training process are integrated, and the prediction mean and prediction interval at each time point are calculated to obtain the prediction model;

[0013] The prediction interval is adjusted and optimized through the validation set so that the coverage and confidence level of the prediction interval meet the requirements and a prediction optimization model is obtained;

[0014] The target object is predicted through the prediction optimization model, and the final wind power probability prediction result is output.

[0015] Optionally, the specific process of acquiring the historical wind power data and preprocessing the historical wind power data to obtain the preprocessed data set is:

[0016] Collect historical wind power data from wind power generation systems or wind power generation related databases, the data including environmental factors affecting wind power output and corresponding wind power output power time series; clean the collected historical wind power data to remove outliers, missing values ​​or invalid data, and normalize or standardize the data to obtain a preprocessed data set.

[0017] Optionally, the specific process of performing random sampling on the preprocessed data set several times to obtain the sample set is:

[0018] The Bootstrap method is used to perform random sampling with replacement on the preprocessed data set. The random sampling is repeated several times to generate several different sample subsets. The probability of each data point being selected is set to be equal in each sampling, and several sample subsets are combined to generate a sample set.

[0019] Optionally, the specific process of constructing the initial prediction model based on the sample set by using an extreme learning machine is:

[0020] Set up the network structure of the extreme learning machine, including setting the number of neurons in the input layer, hidden layer, and output layer;

[0021] Based on the sample set obtained by random sampling, the input weights of the hidden layer neurons are randomly assigned, and the input weights of the hidden layer neurons are kept unchanged during the training process;

[0022] Using the input data and corresponding output data of the sample set, calculate the output matrix of the hidden layer;

[0023] By solving the linear equations, the weights of the output layer are determined to obtain the initial prediction model.

[0024] Optionally, the specific process of training the initial prediction model through the training set is:

[0025] Input the input data in the training set into the initial prediction model, calculate through the hidden layer of the extreme learning machine, and obtain the output of the hidden layer;

[0026] Calculate the weight adjustment of the output layer based on the output of the hidden layer and the actual output data in the training set;

[0027] Based on the weight adjustment amount, the output layer weights of the initial prediction model are updated;

[0028] Repeat the above training steps until the preset training conditions are reached.

[0029] Optionally, the specific process of integrating the wind power probability prediction results of the initial prediction model during the training process and calculating the prediction mean and prediction interval at each time point to obtain the prediction model is as follows:

[0030] Calculate the mean of several prediction results at each time point in the training set as the prediction mean of the corresponding time point;

[0031] Based on several forecast results, the forecast variance at each time point is calculated, and the forecast interval at each time point is determined by statistical methods according to the preset confidence level;

[0032] The prediction results and prediction intervals at each time point are integrated to obtain a prediction model that covers the prediction performance of the entire training set.

[0033] Optionally, the specific process of calculating the prediction variance at each time point and determining the prediction interval according to a preset confidence level includes:

[0034] Calculate the variance of several prediction results at each time point as the uncertainty variance of the prediction model;

[0035] Based on the prediction model, the error between the predicted value and the true observed value is calculated to obtain the variance of the data noise;

[0036] Add the uncertainty variance of the prediction model and the variance of the data noise to get the total prediction variance;

[0037] Based on the preset confidence level, find the critical value of the standard normal distribution and calculate the prediction interval for each time point based on the total prediction variance.

[0038] Optionally, the prediction interval is adjusted and optimized through the validation set so that the coverage and confidence level of the prediction interval meet the requirements, and the specific process of obtaining the prediction optimization model is:

[0039] Input the input data in the validation set into the prediction model to obtain the prediction mean and prediction interval for each time point in the validation set;

[0040] Evaluate the coverage of the prediction interval corresponding to the validation set on the validation set. If the coverage is lower than the preset confidence level requirement, adjust the parameters of the prediction model or retrain the prediction model.

[0041] Evaluate the width of the prediction interval based on the preset confidence level requirements;

[0042] Repeat the above steps until the coverage and confidence level of the prediction interval meet the preset requirements and obtain the prediction optimization model.

[0043] Optionally, the specific process of predicting the target object through the prediction optimization model and outputting the final wind power probability prediction result is:

[0044] The wind power data to be predicted is input into the prediction optimization model, and the predicted value of wind power is calculated according to the input wind power data;

[0045] The prediction optimization model determines the prediction interval of the target object at each time point based on the prediction results integrated during the training process, the calculated prediction variance, and the preset confidence level;

[0046] The wind power forecast value at each time point and its corresponding forecast interval are output through the forecast optimization model as the final wind power probability forecast result.

[0047] Optionally, the training of the initial prediction model through the training set, integrating the wind power probability prediction results of the initial prediction model during the training process, and calculating the prediction mean and prediction interval at each time point specifically includes the steps of:

[0048] The prediction error is divided into two components, namely model error and data noise error. The expression of the prediction error is shown in the following formula (1):

[0049]

[0050] in, represents the prediction error, t i represents the true value at time point i, represents the predicted mean at time point i, represents the model error, ε(x i ) represents the data noise error at time point i;

[0051] Assuming that the model uncertainty error and the data noise error are statistically independent, the variance of the prediction error is obtained as shown in the following formula (2):

[0052]

[0053] in, represents the variance of the prediction error at time point i, represents the variance of the model error at time point i, represents the variance of the data noise at time point i;

[0054] Find the critical value z of the standard normal distribution based on the preset confidence level 100(1-α)%. 1-α / 2 , and using the critical value z 1-α / 2 and the variance of the prediction error The prediction interval for each time point is calculated. The expressions of the confidence upper bound and confidence lower bound of the prediction interval are shown in the following equations (3) and (4):

[0055]

[0056] in, represents the upper confidence bound, represents the confidence lower bound;

[0057] B M Round of Bootstrap sampling, get B M Different training subsets;

[0058] For each training subset, an initial prediction model is constructed and trained by the extreme learning machine to obtain B M prediction models and their corresponding prediction results;

[0059] Calculate B at each time point M The mean of the prediction results is taken as the prediction mean at the corresponding time point. The expression of the prediction mean is shown in the following formula (5):

[0060]

[0061] in, represents the prediction result of the lth prediction model at time point i;

[0062] According to B M The model variance at each time point is calculated based on the prediction results. The model variance reflects the discrete degree of the prediction results of the prediction model on different training subsets. The expression of the model variance is shown in the following formula (6):

[0063]

[0064] B N Round of Bootstrap sampling, get B N Different validation subsets to assess the uncertainty of data noise;

[0065] Calculate the error between the predicted value and the true observed value of each validation subset and get B N Error results;

[0066] According to B N The error results are used to calculate the mean value and variance of the data noise. The variance of the data noise reflects the degree of dispersion of the error between the true observation value and the predicted value. The expressions of the mean value and variance of the data noise are shown in the following equations (7) and (8):

[0067]

[0068] in, or represents the estimate of the mean of the data noise at time point i, represents the error between the predicted value and the true value of the l-th validation subset at time point i, represents the estimate of the data noise variance at time point i;

[0069] The total prediction variance is the sum of the model variance and the data noise variance, as shown in the following formula (9):

[0070]

[0071] in, represents the total prediction variance at time point i.

[0072] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0073] Improve prediction accuracy: By using extreme learning machines (ELM) to build prediction models, ELM has faster training speed and better generalization ability than traditional neural networks, which helps to reduce the uncertainty caused by model structure and parameter errors and improve the accuracy of wind power prediction. At the same time, by integrating the prediction results, calculating the prediction mean and prediction interval at each time point, the stability and reliability of the prediction are further improved.

[0074] Reducing the impact of data uncertainty: Before building the prediction model, the historical wind power data was preprocessed, and multiple sample sets were generated through random sampling for training. This helps to reduce the impact of data noise on the prediction results and improve the model's adaptability to data changes. In particular, when processing non-stationary time series data, it can significantly improve the model's prediction performance.

[0075] Optimizing the prediction interval: By adjusting and optimizing the prediction interval through the validation set, it can be ensured that the coverage and confidence level of the prediction interval meet the actual application requirements, which helps to better evaluate the uncertainty of the prediction results in practical applications and provide more reliable reference information for decision makers.

[0076] Improve prediction efficiency: The extreme learning machine has the characteristics of fast learning and efficient calculation, which makes the present invention more efficient when processing large-scale wind power data; at the same time, by integrating prediction results and calculating prediction intervals, it can improve the degree of automation of the prediction process while ensuring prediction accuracy and reduce the cost of manual intervention.

[0077] Enhance model robustness: Through multiple random sampling and training, it can better adapt to different data distributions and characteristics, enhance the robustness of the model, help deal with various complex situations in practical applications, and improve the practicality and reliability of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 Schematic diagram of the flow of a wind power probability prediction method based on an extreme learning machine according to an embodiment of the present invention;

[0079] Figure 2 The present invention is a schematic diagram of a prediction model construction process of a wind power probability prediction method based on an extreme learning machine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0080] The following is a specific implementation method in conjunction with the drawings.

[0081] Example 1

[0082] Reference Figure 1 , a wind power probability prediction method based on extreme learning machine, comprising the steps of:

[0083] Step 1: Obtain historical wind power data and preprocess the historical wind power data to obtain a preprocessed data set. In this embodiment, the specific process is:

[0084] Collect historical wind power data from wind power generation systems or wind power generation related databases. The data includes environmental factors that affect wind power output and the corresponding wind power output power time series. Environmental factors that affect wind power output include wind speed, wind direction, temperature, humidity and other environmental factors; clean the collected historical wind power data, remove outliers, missing values ​​or invalid data, and normalize or standardize the data to obtain a preprocessed data set. Outliers may be caused by equipment failure, data recording errors, etc., which will have a negative impact on model training. Statistical methods (such as the 3σ principle) or machine learning algorithms (such as isolation forests) can be used to detect outliers. For missing data points, interpolation methods (such as linear interpolation, polynomial interpolation, etc.) can be used or estimates can be made based on the trends of adjacent data points. If there are too many missing values, you may need to consider discarding the data for that time period. Remove data that is illogical or obviously wrong, such as negative wind speed values ​​or wind power output that exceeds the rated power of the equipment. Scale the data to a specific range (such as between 0 and 1) to eliminate the dimensional differences between different features. Select an appropriate normalization or standardization method based on the characteristics of the data and the requirements of the model.

[0085] Step 2: Perform random sampling on the preprocessed data set several times to obtain a sample set, and divide the sample set into a training set and a validation set.

[0086] In this embodiment, the specific process of performing random sampling on the preprocessed data set several times to obtain the sample set is as follows:

[0087] The Bootstrap method is used to perform random sampling with replacement on the preprocessed data set. The random sampling is repeated several times to generate several different sample subsets. The probability of each data point being selected is set to be equal in each sampling, and several sample subsets are combined to generate a sample set.

[0088] In this embodiment, the number of samplings required is determined according to the size of the preprocessed data set and the complexity of the model; the size of each sample subset is determined, and the size of each sample subset depends on the amount of data required for the training model and the limitation of computing resources. For the preprocessed data set, the Bootstrap method is used to repeat several random samplings, and each sampling generates a different sample subset. In each sampling process, the probability of each data point being selected is equal, which ensures the randomness and representativeness of the sampling. The training set and the validation set are randomly divided from the sample set; the training set is used to train the extreme learning machine model, that is, to build and adjust the parameters of the model; the validation set is used to evaluate the performance of the model, including prediction accuracy, coverage of the prediction interval, and confidence level; the division ratio can be adjusted according to actual needs. Usually, the training set will account for a larger proportion and the validation set will account for a smaller proportion. When dividing the training set and the validation set, ensure that the data in the two sets have similar distribution characteristics.

[0089] Step 3: Based on the sample set, build the initial prediction model through the extreme learning machine.

[0090] In this embodiment, the specific process is:

[0091] Set the network structure of the extreme learning machine, including setting the number of neurons in the input layer, hidden layer and output layer; according to the characteristics of the wind power prediction problem, the number of neurons in the input layer should be equal to the number of input features, which usually include environmental factors such as wind speed, wind direction, temperature, humidity, and possible time-related features (such as timestamp, seasonal information, etc.); the number of neurons in the hidden layer is a hyperparameter, which is determined by experiments or cross-validation so that the model can capture the nonlinear relationship of the data without overfitting the training data; the number of neurons in the output layer is equal to the number of prediction targets.

[0092] Based on the sample set obtained by random sampling, the input weights of the hidden layer neurons are randomly assigned, and the input weights of the hidden layer neurons are kept unchanged during the training process; the weight assignment can adopt a random distribution method such as uniform distribution or normal distribution.

[0093] Calculate the output matrix of the hidden layer using the input data and corresponding output data of the sample set. Select a suitable activation function, such as Sigmoid function, ReLU function, Tanh function, etc., for the neurons of the hidden layer; for each input sample (i.e. each data point) in the sample set, perform matrix multiplication with the input weight of the hidden layer; add the result to the bias of the hidden layer; perform nonlinear transformation on the result of the previous step through the activation function to obtain the output of the hidden layer; repeat the above calculation process for all input samples in the sample set, and finally obtain the output matrix of the hidden layer of the entire sample set.

[0094] By solving the linear equations, the weights of the output layer are determined to obtain the initial prediction model. Given the output matrix of the hidden layer and the target output matrix of the training set, the weights of the output layer can be solved by the least squares solution to determine the relevant parameters of the initial prediction model and obtain the initial prediction model.

[0095] Step 4: Train the initial prediction model through the training set, integrate the wind power probability prediction results of the initial prediction model during the training process, and calculate the prediction mean and prediction interval at each time point to obtain the prediction model.

[0096] In this embodiment, the specific process of training the initial prediction model through the training set is:

[0097] Input the input data in the training set into the initial prediction model, calculate through the hidden layer of the extreme learning machine to obtain the output of the hidden layer; calculate the weight adjustment of the output layer based on the output of the hidden layer and the actual output data in the training set; based on the weight adjustment, update the output layer weight of the initial prediction model; repeat the above training steps until the preset training conditions are met.

[0098] In this embodiment, the wind power probability prediction results of the initial prediction model during the training process are integrated, and the prediction mean and prediction interval at each time point are calculated. The specific process of obtaining the prediction model is as follows:

[0099] Calculate the mean of several prediction results at each time point in the training set as the prediction mean of the corresponding time point; calculate the prediction variance of each time point based on several prediction results, and determine the prediction interval of each time point by statistical methods according to the preset confidence level; integrate the prediction results and prediction intervals of each time point to obtain a prediction model that covers the prediction performance of the entire training set.

[0100] In this embodiment, the specific process of calculating the prediction variance at each time point and determining the prediction interval according to the preset confidence level includes:

[0101] Calculate the variance of several prediction results at each time point as the uncertainty variance of the prediction model; based on the prediction model, calculate the error between the predicted value and the true observed value to obtain the variance of the data noise; add the uncertainty variance of the prediction model and the variance of the data noise to obtain the total prediction variance; according to the preset confidence level, find the critical value of the standard normal distribution, and calculate the prediction interval for each time point based on the total prediction variance. Integrate the predicted mean and prediction interval at each time point to obtain a prediction model that covers the prediction performance of the entire training set. This integrated prediction model not only includes the predicted value at each time point, but also the corresponding prediction interval, so that it can more comprehensively reflect the uncertainty of wind power.

[0102] Step 5: Adjust and optimize the prediction interval through the validation set so that the coverage and confidence level of the prediction interval meet the requirements and obtain the prediction optimization model.

[0103] In this embodiment, the specific process is:

[0104] Input the input data in the validation set into the prediction model to obtain the prediction mean and prediction interval for each time point in the validation set. Input the input data in the validation set into the constructed prediction model, including environmental factors such as wind speed, wind direction, temperature, humidity, and possible time-related characteristics; the prediction model calculates and outputs the wind power prediction mean and prediction interval for each time point in the validation set based on these input data.

[0105] Evaluate the coverage of the prediction interval corresponding to the validation set on the validation set. If the coverage is lower than the preset confidence level requirement, adjust the parameters of the prediction model or retrain the prediction model. Calculate whether the actual wind power value at each time point in the validation set falls within the prediction interval to calculate the coverage of the prediction interval; if the coverage is lower than the preset confidence level requirement (for example, 90% or 95% confidence level), it indicates that the prediction interval may be too narrow or inaccurate and needs to be adjusted.

[0106] Based on the preset confidence level requirements, evaluate the width of the prediction interval. If the coverage of the prediction interval does not meet the requirements, consider adjusting the parameters of the prediction model, such as the number of hidden layer neurons, the choice of activation function, the initialization method of the output layer weights, etc.; after adjusting the parameters, retrain the prediction model and use the validation set for verification again to evaluate the effect of the adjustment. In addition to the coverage, the width of the prediction interval also needs to be evaluated. A prediction interval that is too wide may mean that the model has high uncertainty, while a prediction interval that is too narrow may not fully cover the actual wind power fluctuations; based on the preset confidence level requirements, evaluate whether the width of the prediction interval is reasonable. If the prediction interval is too wide or too narrow, consider adjusting the model training process or data preprocessing method.

[0107] Repeat the above steps until the coverage and confidence level of the prediction interval meet the preset requirements and obtain the prediction optimization model. The steps include inputting the validation set data, evaluating the coverage and width of the prediction interval, adjusting the prediction model parameters, etc., until the coverage and confidence level of the prediction interval meet the preset requirements. During the iteration process, the performance of the prediction model after each adjustment can be recorded for comparison and analysis. When the coverage and confidence level of the prediction interval meet the preset requirements, the prediction model is considered to have been optimized and can be used as the final prediction optimization model.

[0108] Step 6: Predict the target object through the prediction optimization model and output the final wind power probability prediction result.

[0109] In this embodiment, the specific process is:

[0110] The wind power data to be predicted is input into the prediction optimization model, and the predicted value of wind power is calculated based on the input wind power data; the prediction optimization model determines the prediction interval of the target object at each time point based on the prediction results integrated in the training process and the calculated prediction variance, as well as the preset confidence level; the prediction optimization model outputs the wind power prediction value and its corresponding prediction interval at each time point as the final wind power probability prediction result.

[0111] In this embodiment, the prediction results can be verified using actual observation data to evaluate the performance of the prediction model, which can be achieved by comparing indicators such as the difference between the predicted value and the actual observed value, the coverage of the prediction interval, and the confidence level. According to the evaluation results, if it is found that the prediction model has problems such as the prediction interval is too wide or too narrow, the coverage is insufficient, etc., the model can be further optimized and adjusted to improve the accuracy and reliability of the prediction. The final wind power probability prediction results are applied to wind farm operations, power market transactions, energy planning and other fields to provide a scientific basis for decision makers. According to the prediction results, formulate corresponding wind farm scheduling plans, power market trading strategies, energy reserve plans, etc. to deal with the uncertainty of wind power.

[0112] Example 2

[0113] Based on Example 1, refer to Figure 2 In this embodiment, the initial prediction model is trained through the training set, the wind power probability prediction results of the initial prediction model are integrated during the training process, and the prediction mean and prediction interval at each time point are calculated, specifically including the steps of:

[0114] The prediction error is divided into two components, namely model error and data noise error. The expression of the prediction error is shown in the following formula (1):

[0115]

[0116] in, represents the prediction error, t i represents the true value at time point i, represents the predicted mean at time point i, represents the model error, ε(x i ) represents the data noise error at time point i;

[0117] Assuming that the model uncertainty error and the data noise error are statistically independent, the variance of the prediction error is obtained as shown in the following formula (2):

[0118]

[0119] in, represents the variance of the prediction error at time point i, represents the variance of the model error at time point i, represents the variance of the data noise at time point i;

[0120] Find the critical value z of the standard normal distribution based on the preset confidence level 100(1-α)%. 1-α / 2 , and using the critical value z 1-α / 2 and the variance of the prediction error The prediction interval for each time point is calculated. The expressions of the confidence upper bound and confidence lower bound of the prediction interval are shown in the following equations (3) and (4):

[0121]

[0122] in, represents the upper confidence bound, represents the confidence lower bound;

[0123] B M Round of Bootstrap sampling, get B M Different training subsets;

[0124] For each training subset, an initial prediction model is constructed and trained by the extreme learning machine to obtain B M prediction models and their corresponding prediction results;

[0125] Calculate B at each time point M The mean of the prediction results is taken as the prediction mean at the corresponding time point. The expression of the prediction mean is shown in the following formula (5):

[0126]

[0127] in, represents the prediction result of the lth prediction model at time point i;

[0128] According to B M The model variance at each time point is calculated based on the prediction results. The model variance reflects the discrete degree of the prediction results of the prediction model on different training subsets. The expression of the model variance is shown in the following formula (6):

[0129]

[0130] B N Round of Bootstrap sampling, get B N Different validation subsets to assess the uncertainty of data noise;

[0131] Calculate the error between the predicted value and the true observed value of each validation subset and get B N Error results;

[0132] According to B N The error results are used to calculate the mean value and variance of the data noise. The variance of the data noise reflects the degree of dispersion of the error between the true observation value and the predicted value. The expressions of the mean value and variance of the data noise are shown in the following equations (7) and (8):

[0133]

[0134] in, or represents the estimate of the mean of the data noise at time point i, represents the error between the predicted value and the true value of the l-th validation subset at time point i, represents the estimate of the data noise variance at time point i;

[0135] The total prediction variance is the sum of the model variance and the data noise variance, as shown in the following formula (9):

[0136]

[0137] in, Represents the total prediction variance at time point i. By substituting the total prediction variance into the calculation formula of the prediction interval, the prediction interval at each time point is obtained, thereby obtaining a more accurate and reliable wind power probability prediction result.

Claims

1. A wind power probability prediction method based on extreme learning machine, characterized in that: Includes steps: Acquire historical wind power data, and preprocess the historical wind power data to obtain a preprocessed data set; Perform random sampling on the preprocessed data set several times to obtain a sample set, and divide the sample set into a training set and a validation set; Based on the sample set, an initial prediction model is constructed through an extreme learning machine; The initial prediction model is trained through the training set, the wind power probability prediction results of the initial prediction model during the training process are integrated, and the prediction mean and prediction interval at each time point are calculated to obtain the prediction model; The prediction interval is adjusted and optimized through the validation set so that the coverage and confidence level of the prediction interval meet the requirements and a prediction optimization model is obtained; The target object is predicted through the prediction optimization model, and the final wind power probability prediction result is output.

2. The wind power probability prediction method based on extreme learning machine according to claim 1, characterized in that: The specific process of obtaining the historical wind power data and preprocessing the historical wind power data to obtain the preprocessed data set is as follows: Collect historical wind power data from wind power generation systems or wind power generation related databases, the data including environmental factors affecting wind power output and corresponding wind power output power time series; clean the collected historical wind power data to remove outliers, missing values ​​or invalid data, and normalize or standardize the data to obtain a preprocessed data set.

3. The wind power probability prediction method based on extreme learning machine according to claim 1, characterized in that: The specific process of performing random sampling on the preprocessed data set several times to obtain the sample set is as follows: The Bootstrap method is used to perform random sampling with replacement on the preprocessed data set. The random sampling is repeated several times to generate several different sample subsets. The probability of each data point being selected is set to be equal in each sampling, and several sample subsets are combined to generate a sample set.

4. The wind power probability prediction method based on extreme learning machine according to claim 3, characterized in that: The specific process of constructing the initial prediction model based on the sample set by the extreme learning machine is as follows: Set up the network structure of the extreme learning machine, including setting the number of neurons in the input layer, hidden layer, and output layer; Based on the sample set obtained by random sampling, the input weights of the hidden layer neurons are randomly assigned, and the input weights of the hidden layer neurons are kept unchanged during the training process; Using the input data and corresponding output data of the sample set, calculate the output matrix of the hidden layer; By solving the linear equations, the weights of the output layer are determined to obtain the initial prediction model.

5. The wind power probability prediction method based on extreme learning machine according to claim 4, characterized in that: The specific process of training the initial prediction model through the training set is as follows: Input the input data in the training set into the initial prediction model, calculate through the hidden layer of the extreme learning machine, and obtain the output of the hidden layer; Calculate the weight adjustment of the output layer based on the output of the hidden layer and the actual output data in the training set; Based on the weight adjustment amount, the output layer weights of the initial prediction model are updated; Repeat the above training steps until the preset training conditions are reached.

6. The wind power probability prediction method based on extreme learning machine according to claim 1, characterized in that: The specific process of integrating the wind power probability prediction results of the initial prediction model during the training process and calculating the prediction mean and prediction interval at each time point to obtain the prediction model is as follows: Calculate the mean of several prediction results at each time point in the training set as the prediction mean of the corresponding time point; Based on several forecast results, the forecast variance at each time point is calculated, and the forecast interval at each time point is determined by statistical methods according to the preset confidence level; The prediction results and prediction intervals at each time point are integrated to obtain a prediction model that covers the prediction performance of the entire training set.

7. The wind power probability prediction method based on extreme learning machine according to claim 6, characterized in that: The specific process of calculating the prediction variance at each time point and determining the prediction interval according to the preset confidence level includes: Calculate the variance of several prediction results at each time point as the uncertainty variance of the prediction model; Based on the prediction model, the error between the predicted value and the true observed value is calculated to obtain the variance of the data noise; Add the uncertainty variance of the prediction model and the variance of the data noise to get the total prediction variance; Based on the preset confidence level, find the critical value of the standard normal distribution and calculate the prediction interval for each time point based on the total prediction variance.

8. The wind power probability prediction method based on extreme learning machine according to claim 1, characterized in that: The specific process of adjusting and optimizing the prediction interval through the validation set so that the coverage and confidence level of the prediction interval meet the requirements and obtaining the prediction optimization model is as follows: Input the input data in the validation set into the prediction model to obtain the prediction mean and prediction interval for each time point in the validation set; Evaluate the coverage of the prediction interval corresponding to the validation set on the validation set. If the coverage is lower than the preset confidence level requirement, adjust the parameters of the prediction model or retrain the prediction model. Evaluate the width of the prediction interval based on the preset confidence level requirements; Repeat the above steps until the coverage and confidence level of the prediction interval meet the preset requirements and obtain the prediction optimization model.

9. The wind power probability prediction method based on extreme learning machine according to claim 1, characterized in that: The specific process of predicting the target object through the prediction optimization model and outputting the final wind power probability prediction result is as follows: The wind power data to be predicted is input into the prediction optimization model, and the predicted value of wind power is calculated according to the input wind power data; The prediction optimization model determines the prediction interval of the target object at each time point based on the prediction results integrated during the training process, the calculated prediction variance, and the preset confidence level; The prediction optimization model outputs the wind power prediction value at each time point and its corresponding prediction interval as the final wind power probability prediction result.

10. The wind power probability prediction method based on extreme learning machine according to claim 1, characterized in that: The initial prediction model is trained by the training set, the wind power probability prediction results of the initial prediction model are integrated during the training process, and the prediction mean and prediction interval at each time point are calculated, specifically including the steps of: The prediction error is divided into two components, namely model error and data noise error. The expression of the prediction error is shown in the following formula (1): in, represents the prediction error, t i represents the true value at time point i, represents the predicted mean at time point i, represents the model error, ε(x i ) represents the data noise error at time point i; Assuming that the model uncertainty error and the data noise error are statistically independent, the variance of the prediction error is obtained as shown in the following formula (2): in, represents the variance of the prediction error at time point i, represents the variance of the model error at time point i, represents the variance of the data noise at time point i; Find the critical value z of the standard normal distribution based on the preset confidence level 100(1-α)%. 1-α / 2 , and using the critical value z 1-α / 2 and the variance of the prediction error The prediction interval for each time point is calculated. The expressions of the confidence upper bound and confidence lower bound of the prediction interval are shown in the following equations (3) and (4): in, represents the upper confidence bound, represents the confidence lower bound; B M Round of Bootstrap sampling, get B M Different training subsets; For each training subset, an initial prediction model is constructed and trained by the extreme learning machine to obtain B M prediction models and their corresponding prediction results; Calculate B at each time point M The mean of the prediction results is taken as the prediction mean at the corresponding time point. The expression of the prediction mean is shown in the following formula (5): in, represents the prediction result of the lth prediction model at time point i; According to B M The model variance at each time point is calculated based on the prediction results. The model variance reflects the discrete degree of the prediction results of the prediction model on different training subsets. The expression of the model variance is shown in the following formula (6): B N Round of Bootstrap sampling, get B N Different validation subsets to assess the uncertainty of data noise; Calculate the error between the predicted value and the true observed value of each validation subset and get B N Error results; According to B N The error results are used to calculate the mean value and variance of the data noise. The variance of the data noise reflects the degree of dispersion of the error between the true observation value and the predicted value. The expressions of the mean value and variance of the data noise are shown in the following equations (7) and (8): in, or represents the estimate of the mean of the data noise at time point i, represents the error between the predicted value and the true value of the l-th validation subset at time point i, represents the estimate of the data noise variance at time point i; The total prediction variance is the sum of the model variance and the data noise variance, as shown in the following formula (9): in, represents the total prediction variance at time point i.

Images