Causal convolutional gated recurrent unit network with multiple decomposition methods for short-term wind speed forecasting

摘要

Wind speed exhibits different and complex fluctuation characteristics, which makes it challenging for wind speed forecasting. Decomposition methods have been widely and successfully applied in wind speed forecasting, for they could extract the fluctuation patterns by decomposing wind speed into sub-signals. However, the sub-signals are always modeled and forecasted separately, which neglects the intercorrelations of the sub-signals. Capturing the intercorrelations helps to obtain more effective features and further improve the forecasting performance. To address this issue, we propose a new hybrid model by combining a causal convolutional network (CCN), a gated recurrent unit (GRU) network, and multiple decomposition methods. In the proposed model, multiple decomposition methods are adopted to decompose the original wind speed into diversified sub-signals, CCN is applied to extract more effective features from the decomposed sub-signals, and GRU is employed to identify the temporal dependencies between the extracted features and future wind speed. Four wind speed datasets collected in different seasons are introduced for experimental analysis. The experimental results demonstrate that: (1) the proposed model outperforms the benchmark models consistently in terms of forecasting accuracy and stability; (2) the forecasting performance of the proposed model could be significantly improved by using multiple decomposition methods; (3) CCN and GRU adopted in the proposed model are both effective for improving the forecasting performance.