Wind Power Prediction Model Based on Power Division Process and Deep Learning

In actual wind farms, the working state of wind turbines is greatly affected by environmental factors, which is prone to violent fluctuations of wind power. Especially, the wind turbines have uphill working state. Therefore, it is difficult to ensure good prediction accuracy by direct use of machine learning for power prediction. In this paper, a hybrid CNNs-LSTM power prediction model (CNNs-LSTM) based on power feature division is proposed. The model combines the advantages of power fluctuation, CNNs and LSTM to effectively reduce the error distribution. First, according to the ac-tual power fluctuation characteristics of the wind farm, the power fluctuation threshold is designed to divide the different power fluctuation processes. Secondly, CNNs were used to extract the potential relationship between wind speed, wind direction, temperature and humidity to fully explore the spatio-temporal characteristics of input data. Finally, different experimental plat-form prediction models are established. The results show that the prediction model has good prediction accuracy.