A Four-Stage Hybrid Short-term Wind Direction Forecasting Method

Abstract

The invention discloses a four-stage mixed short-term wind direction prediction method, which belongs to the field of wind direction prediction. The method first sets the sampling time, collects the original data of the wind direction, and establishes the input and output data matrix of the wind direction; The eigenvectors with low correlation degree are retained, and the dimension L of the input matrix is ​​determined; the variational mode decomposition VMD method is used to perform K-order decomposition of the L-dimensional feature sequence, and the characteristic information of the wind direction input is mined; the DBN network is used Build a wind direction deep learning prediction model, input the decomposed K×L dimensional wind direction input subsequence to obtain the wind direction prediction value; correct the prediction error of the DBN wind direction prediction model through the BP neural network, and improve the wind direction prediction accuracy; finally, the corrected wind direction The prediction model accuracy was verified. The invention has great significance for wind direction prediction in high-precision and complex areas.

Description

technical field [0001] The invention relates to the technical field of wind direction prediction, in particular to a four-stage mixed short-term wind direction prediction method. Background technique [0002] With the rapid growth of new energy, wind power as a green energy has developed rapidly in the past century. With the continuous growth of wind power access, the impact of intermittent and random fluctuations of wind power generation on the power grid is becoming more and more obvious. The instability, intermittent and randomness of wind direction increase the difficulty of wind direction modeling and prediction. Accurate wind direction prediction can improve the power generation efficiency of wind turbines, increase wind power generation capacity and profits, and provide the necessary basis for grid scheduling and unit combination operations. [0003] The main methods of wind direction prediction at this stage include classical statistical methods and data-driven met...

AI Technical Summary

Problems solved by technology

The accuracy of these models is very dependent on the artificial setting of parameters, such as the number of layers and nodes of the feed-forward neural network, the kernel function and penalty factor of the support vector machine, etc. Improper parameter setting may lead to insufficient accuracy or overfitting

Therefore, the data-driven approach requires a large number of repeated experiments to determine the data parameter settings

Images