To address the challenges posed by the strong randomness and difficulty in accurately predicting wave power generation

this paper proposes a hybrid prediction method based on ensemble empirical mode decomposition (EEMD)

temporal convolutional network (TCN)

and bidirectional long short-term memory network (BiLSTM). The EEMD method is applied to decompose the original wave power time series into intrinsic mode functions (IMFs) across different time scales. Then

the primary environmental features are extracted using the least absolute shrinkage and selection operator algorithm. The IMFs are grouped and reconstructed into high-frequency

mid-frequency

and low-frequency components based on their sample entropy values. These reconstructed components

along with the extracted features

are input into the TCN-BiLSTM prediction model. The predictions of each component are summed to obtain the wave height forecasting results

which are subsequently fed into a direct-drive wave power generation system model to predict wave power output. Simulation results demonstrate that the proposed method achieves higher prediction accuracy and exhibits superior generalization capabilities compared to traditional models. It also performs exceptionally well in forecasting tasks across different time horizons.