Swarm intelligence-based Multi-Layer Kernel Meta Extreme Learning Machine for tidal current to power prediction

Tidal energy, with its predictable and consistent nature, offers a scalable ocean renewable resource that can diversify the energy generation mix for countries with suitable coastal conditions. Accurate tidal current-to-power forecasting is essential to optimize power system management, improve grid stability, and inform the design of power processing and storage units. This study proposes a novel hybrid model integrating Swarm Decomposition with a Multi-Layer Kernel Meta Extreme Learning Machine to forecast non-stationary tidal currents. The Swarm Decomposition isolates key oscillatory components, reducing noise and improving feature extraction, while the kernel-based architecture enhances generalization and scalability by minimizing the need for extensive parameter tuning, resulting in higher forecasting accuracy and computational efficiency. The model is validated on two real-world tidal current datasets from distinct locations, incorporating seasonal variations, and compared against well-established extreme learning machines and deep learning models. A sensitivity analysis of signal decomposition parameters demonstrated their impact on decomposition quality and computational cost. The proposed model outperformed superior performance on both tidal datasets, achieving a 5-fold reduction in mean squared error and increased R 2 from 0.9653 to 0.9933. These findings highlight the model’s robustness and adaptability to diverse tidal conditions, making it a reliable tool for tidal power forecasting. • A novel hybrid machine learning model optimizes tidal to power forecasting. • SWD enhances sub-component extraction with richer time-frequency features. • The Kernel-based machine learning model improves computational efficiency. • Superior tidal forecasting accuracy up to 5-fold is achieved. • The sensitivity analysis validates the optimal selection of the SWD parameters.