Wind power prediction method and system based on self-service method and extreme learning machine

Abstract

The invention provides a wind power prediction method and system based on a self-service method and an extreme learning machine, relates to the technical field of wind power prediction, and aims to improve the accuracy of wind power prediction, and the method comprises the following steps: obtaining a plurality of wind power probability prediction values based on the self-service method and the extreme learning machine; and obtaining a prediction variance based on the plurality of wind power probability prediction values and carrying out prediction interval estimation. The method has the advantage of more accurate wind power prediction.

Description

Technical Field

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

[0002] Wind power forecasting is an important means to monitor and ensure the status of wind power.

[0003] In the prior art, wind power forecasting can be performed by combining a bootstrap method with a model. The bootstrap method can flexibly approximate non-constant variance and heterogeneous noise, thereby providing reliable performance for wind power generation probability. However, due to the limitations of traditional neural networks, the computational burden of the data obtained by the bootstrap method is very high, especially for large data sets, and the prediction accuracy is also limited.

[0004] Therefore, it is necessary to improve the wind power forecasting based on the self-service method to improve the accuracy of wind power forecasting. Summary of the invention

[0005] The purpose of the present invention is to provide a wind power prediction method and system based on a self-help method and an extreme learning machine, which can improve the accuracy of wind power prediction.

[0006] The present invention is achieved through the following technical solutions:

[0007] The present invention first provides a wind power prediction method based on a self-help method and an extreme learning machine, comprising the following steps:

[0008] Obtain multiple wind power probability prediction values ​​based on the bootstrap method and extreme learning machine;

[0009] A prediction variance is obtained based on the multiple wind power probability prediction values ​​and a prediction interval is estimated.

[0010] Preferably, the method for obtaining multiple wind power probability prediction values ​​based on the self-help method and the extreme learning machine is:

[0011] Obtaining a bootstrap sequence by bootstrapping based on the training sequence;

[0012] Obtaining the wind power probability prediction value through an extreme learning machine according to the bootstrap sequence;

[0013] Repeat the above steps to obtain multiple wind power probability prediction values.

[0014] Preferably, the method for obtaining a bootstrap sequence by a bootstrap method based on a training sequence is:

[0015] The bootstrap sequence is obtained by sampling the training sequence with uniform distribution and replacement.

[0016] Preferably, the method for obtaining the prediction variance and estimating the prediction interval based on the plurality of wind power probability prediction values ​​is:

[0017] establishing a total prediction error, the total prediction error comprising a plurality of error components;

[0018] Get the variance of the total forecast error.

[0019] Preferably, the error component includes a model error ε m (x i ) and data noise error ε d (x i ):

[0020]

[0021] in, is the wind power probability prediction value, y(x i ) represents the corresponding actual value, x i is an input sequence for inputting a model and obtaining the wind power probability prediction value.

[0022] Preferably, the method for obtaining the variance of the total prediction error is:

[0023] Get the variance σ of the model error m 2 and the variance of the data noise error σ d 2 ;

[0024] Get the variance σ of the total prediction error t 2 :

[0025] σ t 2 =σ m 2 +σ d 2 .

[0026] Preferably, the variance σ of the model error m 2 The method to obtain is:

[0027]

[0028] Among them, B M is the total number of wind power probability prediction values, is the lth wind power probability prediction value, is the average of the wind power probability prediction values.

[0029] Preferably, the variance σ of the data noise error isd 2 The method to obtain is:

[0030]

[0031] in, To obtain the error between the predicted value and the actual observed value when obtaining the lth wind power probability prediction value, is the mean error value.

[0032] The present invention also provides a wind power prediction system based on a self-help method and an extreme learning machine, which is applied to any of the above wind power prediction methods based on a self-help method and an extreme learning machine, and comprises:

[0033] Bootstrap sampling module: used to obtain bootstrap sequences based on bootstrap sampling;

[0034] An extreme learning machine module, used for obtaining a plurality of wind power probability prediction values ​​based on an extreme learning machine;

[0035] The variance acquisition module is used to acquire the prediction variance and perform prediction interval estimation based on the multiple wind power probability prediction values.

[0036] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0037] The present invention extracts samples through a self-help method so that the prediction method of the present invention can be well adapted to complex data;

[0038] The present invention combines the self-help method with the extreme learning machine, which is more suitable for real-time data processing, and has faster response and speed processing, which helps to obtain prediction results more timely and efficiently;

[0039] The present invention improves the instability and inaccuracy of prediction results through error variance, improves the defect that predictions under the self-help method are sensitive to extreme or outliers, and also improves the inaccuracy caused by the uncertainty of the prediction model itself, further improving the accuracy and reliability of predictions;

[0040] The present invention has reasonable design, high calculation efficiency, and high cost performance of computing power, and is convenient for coping with the probability prediction of wind power under various characteristic value collection conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic flow chart of a wind power prediction method based on a self-help method and an extreme learning machine provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0043] Example 1

[0044] The present invention first provides a wind power prediction method based on a self-help method and an extreme learning machine, comprising the following steps:

[0045] Obtain multiple wind power probability prediction values ​​based on the bootstrap method and extreme learning machine;

[0046] A prediction variance is obtained based on the multiple wind power probability prediction values ​​and a prediction interval is estimated.

[0047] This embodiment takes into account the heteroscedasticity of the data sequence collected from wind power generation to construct a prediction confidence interval, thereby further improving the accuracy of the prediction result.

[0048] The prediction model adopted in this embodiment is the extreme learning machine, which is a learning algorithm used to train a single hidden layer feedforward neural network (SLFN). The related learning process is quite different from the learning process of the traditional neural network learning algorithm. It randomly selects the input weights of the hidden layer neurons and determines the output weights through simple matrix calculation analysis. Therefore, it has a faster learning speed than the most commonly used BP learning algorithm. In addition, it also exhibits good generalization ability, which is better than the traditional neural network.

[0049] In this embodiment, the method for obtaining multiple wind power probability prediction values ​​based on the self-help method and the extreme learning machine is:

[0050] Obtaining a bootstrap sequence by bootstrapping based on the training sequence;

[0051] Obtaining the wind power probability prediction value through an extreme learning machine according to the bootstrap sequence;

[0052] Repeat the above steps to obtain multiple wind power probability prediction values.

[0053] Furthermore, the method for obtaining a bootstrap sequence by bootstrap based on a training sequence is:

[0054] The bootstrap sequence is obtained by sampling with replacement of uniform distribution based on the training sequence. That is to say, when obtaining the bootstrap sequence by bootstrap method based on the training sequence, an original training sequence is first obtained, and then the uniformly distributed probability is sampled therefrom, and the sampled data is put into the training set, and the data put into the training set will not be removed from the original training sequence. After the sampling is repeated to form a complete training set, the data that has not been sampled at all is put into the test set, and finally the training data of the complete prediction model is constructed.

[0055] As a preferred solution of this embodiment, a method for obtaining a prediction variance and estimating a prediction interval based on a plurality of wind power probability prediction values ​​is as follows:

[0056] establishing a total prediction error, the total prediction error comprising a plurality of error components;

[0057] Get the variance of the total forecast error.

[0058] Furthermore, the error component includes the model error ε m (x i ) and data noise error ε d (x i ):

[0059]

[0060] in, is the wind power probability prediction value, y(x i ) represents the corresponding actual value, x i is an input sequence for inputting a model and obtaining the wind power probability prediction value.

[0061] This construction method comes from model uncertainty and data uncertainty. Among them, errors in model structure and parameters explain the uncertainty of neural network prediction, which may be caused by local minima during training, randomly generated input weights, etc. In addition, even if the global minimum can be reached, the incorrect specification of the model structure will introduce non-negligible uncertainty in the prediction results. Model uncertainty also comes from another fact, that is, training based on finite samples can never guarantee that the neural network has consistent generalization performance in the unforeseen future. Especially in wind power generation prediction research, it is impossible to find perfect information to reduce the uncertainty of prediction. These factors are collectively referred to as model uncertainty. In addition, data noise can also lead to prediction uncertainty. If the data exhibits random characteristics, it is difficult to model it in a deterministic way. Especially when dealing with non-stationary time series, data noise has a significant impact on the prediction results. In the study, wind power generation data is highly chaotic. Determining the variance of data noise is crucial to constructing the prediction interval.

[0062] On this basis, the method for obtaining the variance of the total prediction error is:

[0063] Get the variance σ of the model error m 2 and the variance of the data noise error σ d 2 ;

[0064] Get the variance σ of the total prediction error t 2 :

[0065] σ t 2 =σ m 2 +σ d 2 .

[0066] Specifically, the variance σ of the model error m 2 The method to obtain is:

[0067]

[0068] Among them, B M is the total number of wind power probability prediction values, is the lth wind power probability prediction value, is the average of the wind power probability prediction values.

[0069] On the other hand, the variance σ of the data noise error d 2 The method to obtain is:

[0070]

[0071] in, To obtain the error between the predicted value and the actual observed value when obtaining the lth wind power probability prediction value, is the mean error value.

[0072] This embodiment uses the probability prediction method to avoid the above two uncertainties and quantify the prediction interval with relevant confidence. That is, a prediction interval can be constructed based on the total error variance and the actual prediction value, thereby improving the accuracy of wind power probability prediction. For example, when obtaining multiple wind power probability prediction values, the average or median of the prediction values ​​is obtained, and then the upper and lower floating to construct a confidence interval.

[0073] In addition, the specific predicted value can also be directly corrected. For example, in the specific implementation, the total error variance σ is obtained. t2 , and multiple wind power probability prediction values The final corrected forecast data can also be obtained by assigning different weights to each wind power probability forecast value, for example:

[0074] Based on the total error variance σ t 2 Obtain the error evaluation parameters of each wind power probability prediction value respectively:

[0075]

[0076] in, is the error evaluation parameter of the lth wind power probability prediction value;

[0077] Get the comprehensive proportion of error evaluation parameters of the lth wind power probability prediction value τ l :

[0078]

[0079] Finally, the weight π of the lth wind power probability prediction value is obtained l :

[0080]

[0081] Finally, the predicted value y is corrected according to the weight:

[0082]

[0083] Example 2

[0084] This embodiment provides a wind power prediction system based on a self-help method and an extreme learning machine, which is applied to any of the above wind power prediction methods based on a self-help method and an extreme learning machine, including:

[0085] Bootstrap sampling module: used to obtain bootstrap sequences based on bootstrap sampling;

[0086] An extreme learning machine module, used for obtaining a plurality of wind power probability prediction values ​​based on an extreme learning machine;

[0087] The variance acquisition module is used to acquire the prediction variance and perform prediction interval estimation based on the multiple wind power probability prediction values.

[0088] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A wind power prediction method based on a bootstrap method and an extreme learning machine, characterized in that: The following steps are involved: Obtain multiple wind power probability prediction values ​​based on the bootstrap method and extreme learning machine; A prediction variance is obtained based on the multiple wind power probability prediction values ​​and a prediction interval is estimated.

2. The wind power prediction method based on the self-help method and extreme learning machine according to claim 1 is characterized in that: The method for obtaining multiple wind power probability prediction values ​​based on the self-help method and the extreme learning machine is: Obtaining a bootstrap sequence by bootstrapping based on the training sequence; Obtaining the wind power probability prediction value through an extreme learning machine according to the bootstrap sequence; Repeat the above steps to obtain multiple wind power probability prediction values.

3. The wind power prediction method based on the self-help method and extreme learning machine according to claim 2 is characterized in that: The method for obtaining a bootstrap sequence by bootstrap based on a training sequence is: The bootstrap sequence is obtained by sampling the training sequence with uniform distribution and replacement.

4. The wind power prediction method based on the self-help method and extreme learning machine according to claim 1 is characterized in that: The method for obtaining the prediction variance and estimating the prediction interval based on the multiple wind power probability prediction values ​​is as follows: establishing a total prediction error, the total prediction error comprising a plurality of error components; Get the variance of the total forecast error.

5. The wind power prediction method based on the self-help method and extreme learning machine according to claim 4 is characterized in that: The error components include the model error ε m (x i ) and data noise error ε d (x i ): in, is the wind power probability prediction value, y(x i ) represents the corresponding actual value, x i is an input sequence for inputting a model and obtaining the wind power probability prediction value.

6. The wind power prediction method based on the self-help method and extreme learning machine according to claim 5 is characterized in that: The method for obtaining the variance of the total prediction error is: Get the variance σ of the model error m 2 and the variance of the data noise error σ d 2 ; Get the variance σ of the total prediction error t 2 : s t 2 =s m 2 +s d 2 。 7. The wind power prediction method based on the self-help method and extreme learning machine according to claim 6 is characterized in that: The variance of the model error σ m 2 The method to obtain is: Among them, B M is the total number of wind power probability prediction values, is the lth wind power probability prediction value, is the average of the wind power probability prediction values.

8. The wind power prediction method based on the self-help method and extreme learning machine according to claim 7 is characterized in that: The variance of the data noise error σ d 2 The method to obtain is: in, To obtain the error between the predicted value and the actual observed value when obtaining the lth wind power probability prediction value, is the mean error value.

9. A wind power prediction system based on a self-help method and an extreme learning machine, applied to the wind power prediction method based on a self-help method and an extreme learning machine as claimed in any one of claims 1 to 8, characterized in that: include: Bootstrap sampling module: used to obtain bootstrap sequences based on bootstrap sampling; An extreme learning machine module, used for obtaining a plurality of wind power probability prediction values ​​based on an extreme learning machine; The variance acquisition module is used to acquire the prediction variance and perform prediction interval estimation based on the multiple wind power probability prediction values.

Images