High Accuracy Short-term Wind Speed Prediction Methods based on LSTM

Abstract

Recently, with the growing problem of global warming and the depletion of fossil resources, government's attention has been focused on renewable energy sources. Wind power, a major renewable energy source, is underutilized in Japan due to its unstable output. So high accurate short-term wind speed prediction methods are the key to the promotion and popularization of wind power generation. Improving the accuracy of wind speed prediction can effectively improve the stability and economy of wind turbines. In this paper, the authors propose a wind speed prediction method by using neural network called LSTM (Long Short-Term Memory) and carry out the accuracy verification by case study. In addition, the authors propose two improved prediction methods to solve the problems of insufficient learning data, poor prediction accuracy due to the wide range of data dispersion, and poor prediction accuracy at the peak. The effectiveness of the improved methods is verified by case study. And the accuracy of all the methods was verified by using meteorological data from Akita and Hokkaido in Japan.