A hybrid error correction method based on EEMD and ConvLSTM for offshore wind power forecasting

Efficient offshore wind power generation and seamless grid integration rely on accurate forecasting. However, the nonlinear and highly complex nature of ocean environmental conditions presents significant challenges, particularly for offshore applications. Thus, this paper introduces a novel hybrid deep learning model to address these challenges. The model employs Long Short-Term Memory (LSTM) to generate an initial offshore wind power forecast. The error signal is then extracted by subtracting the forecasted values from the actual wind power values. To capture nonlinear patterns in the forecasting error, the Ensemble Empirical Mode Decomposition (EEMD) decomposes the error signal into Intrinsic Mode Functions (IMFs) and a residual component. These components are forecasted using the Convolutional Long Short-Term Memory (ConvLSTM) model, which extracts spatial and temporal dependencies. The forecasted error components are summed to reconstruct the wind power error, which is subsequently added to the initial forecast to produce the adjusted wind power forecast. The model's performance is evaluated using an hourly wind power dataset from a Siemens SWT-3.6-120 Offshore turbine at the Amrumbank West wind farm. Comparative analysis with benchmark and hybrid models demonstrates superior accuracy, particularly for extended forecasting horizons of 2, 3, and 4 h. These findings underscore the proposed model's effectiveness in enhancing short-term offshore wind power forecasting. • A hybrid deep learning model with error correction is proposed for offshore wind power forecasting. • EEMD captures nonlinear patterns and reduces noise in offshore wind power forecasting errors. • ConvLSTM extracts spatial and temporal dependencies in offshore wind power forecasting error sequences. • The proposed model outperforms benchmarks with improved evaluation metrics across multiple temporal scales.