Predicting Internal Energy Consumption of a Wind Turbine Using Semi-Supervised Deep Learning

Most previous works on wind power generation focused on the impact of the external environment on the efficiency of energy generation, but ignored the energy consumption of internal parts of the wind turbine. Reducing internal energy consumption can not only improve the generation efficiency, but also reduce the maintenance cost of wind turbines. Therefore, this study uses deep learning to predict the energy consumption inside the wind turbine by installing dozens of sensors inside it, and finds the parts that have greater impact on energy consumption to reduce energy consumption and improve generating efficiency. Since most data on wind turbines is collected by humans currently, it is inevitable that the data will have missing or wrong. Due to the large number of parts inside the wind turbine, the collected data belongs to multi-dimension data. In order to use these data effectively, this study proposes a semi-supervised deep learning method which can correct the data to solve this problem. After all the data are corrected and the model is completely trained, this study uses the MCC method to judge the predicting results of the model. The results show that when the label data accounts for 15-20% of the total data the trained model has the best predictive ability. Therefore, this study suggests that when establishing a prediction model of internal energy consumption of wind turbine in the future, the label data should account for 15-20% of the total data. In this way, not only can train a model with considerable accuracy, but also the most economical ways to determine the amount of revision data.