Multistep forecasting for diurnal wind speed based on hybrid deep learning model with improved singular spectrum decomposition

Abstract

The actual wind speed data suffers from the intermittent and fluctuating property, which implies that it is very difficult to forecast wind speed with high accuracy by applying single or shallow models. Hence, with the purpose of improving the forecasting accuracy and obtain better forecasting results, in this paper, a novel hybrid deep learning model is proposed for multistep forecasting of diurnal wind speed, which is intuitively abbreviated as ISSD-LSTM-GOASVM. Under this formulated model, the improved singular spectrum decomposition (ISSD) is firstly presented to decompose successively the raw wind speed data into several sub-series from high-frequency to low-frequency, which can solve the shortcoming of artificial experience selection of embedding dimension in the original SSD. Then, an effective deep learning model named long short-term memory (LSTM) is adopted to predict the low-frequency sub-series. Meanwhile, deep belief network (DBN) architecture with three hidden layers is constructed to predict the high-frequency sub-series, where hyper-parameters of DBN are determined automatically by grasshopper optimization algorithm (GOA). Finally, the forecasting results of all sub-series are combined to accomplish the final wind speed forecasting. Practical wind speed data from three regions are exploited to evaluate the performance of the proposed model. Case study results indicate that the proposed model for diurnal wind speed has a superior forecasting capability. Moreover, the proposed hybrid model is very competitive compared to the state-of-the-art single model and other hybrid models involved in this paper.