An ocean tidal energy point-interval forecasting system based on enhanced auxiliary feature, mode decomposition combined with compressive sensing and attention interaction

Promoting the transformation of energy structure and developing clean energy technology has become a common consensus. As a clean energy, the accurate forecasting of tidal energy is of great significance to control and decision-making. To solve the high-dimensional problem of tidal energy influencing factor and extract the spatial feature between tidal energy and influencing factor, an enhanced auxiliary forecasting method combing nonlinear dimension reduction and spatial distance, named EAF (Enhanced auxiliary feature), is proposed. To further reduce the complexity of tidal energy, the fluctuation and error of the mode component, combined with compressed sensing theory, an improved successive variational mode decomposition (SVMD) based on orthogonal matching pursuit (OMP) is proposed, and mantis search algorithm (MSA) is used to optimize it, named OMSVMD. To achieve more accurate forecasting, a Transformer structure with interactive fusion of attention mechanism, named AMI-Transformer (Attention mechanism interaction-Transformer), is proposed. To improve the reliability of forecasting and the level of risk control, the Bootstrap method is use to make interval forecasting and the interval forecasting results are obtained. In summary, an ocean tidal energy point-interval forecasting system based on enhanced auxiliary feature, mode decomposition combined with compressive sensing and attention interaction, named EAF-OMSVMD-AMI-Transformer-Bootstrap, is proposed. In the experimental study, taking two sets of real data from Port Angeles and Port Townsend as examples, and comparing the proposed system with 6 benchmark systems. The result shows that the proposed system has high accuracy and robustness. This study provides a new high-precision method and forecasting strategy for the future ocean tidal energy forecasting. • Propose an enhanced auxiliary forecasting method, named EAF. • Propose a mode decomposition method combing compressive sensing and parameter optimization, named OMSVD. • Propose a single encoder stack structure of Transformer, named AMI-Transformer. • Use Bootstrap for interval forecasting to get the interval forecasting results. • Propose a forecasting system named EAF-OMSVMD-AMI-Transformer-Bootstrap.