Wind Power Forecast Based on Dimensionality Reduction Using Ridge Regression

In order to reduce the randomness and fluctuation of wind power, numerical weather prediction is usually used to forecast power generation. Due to the redundancy of numerical weather prediction factors, dimensionality reduction is carried out to improve the effectiveness of data processing and reduce the complexity of algorithm. In this paper, the multicollinearity diagnosis is carried out to show the existence of multicollinearity among numerical weather prediction variables. The ridge regression method is adopted to screen variables, so as to reduce the dimension of numerical weather prediction data, which are trained later by back propagation neural network or convolutional neural network to fit corresponding power data. In comparison with results using training data without preprocessing, dimensionality reduction can improve predicting effect in most cases and reduce the error caused by improper selection of neural network.