Comparison and Evaluation of Multiple Neural Network Models in Wind Power Generation Forecasting

This paper constructs the prediction models of three neural networks: full connected, Recurrent Neural Networks (RNN) and Long Short-Term Memory network (LSTM). The actual wind power output is taken as the experimental data for prediction and analysis. Through multi-angle quantitative comparison, it is found that the three methods can achieve an accuracy rate of more than 95% and a qualification rate (relative error less than 15%) of more than 98%. The training rounds of the fully connected neural network are much higher than those of the other two methods. The LSTM has a strong single training ability. The researchers can select the appropriate super parameters on a small number of data sets to obtain an ideal learning effect quickly.