Deconstructing High-Density Wave Farm Hydrodynamics Using Wave-Aligned Explainable AI

Optimizing the spatial layouts of high-density wave energy converter (WEC) arrays is critical for unlocking marine renewable energy. However, high-fidelity numerical solvers suffer from severe computational bottlenecks when scaling to dense layouts (N 100 devices). Conversely, existing machine learning (ML) surrogates are undermined by optimization-track data leakage due to randomized row splits, yielding artificially inflated accuracies that fail on unseen arrangements. Furthermore, standard ML frameworks lack coordinate invariance, rendering them incapable of extracting physical fluid dynamics. This study proposes a wave-aligned explainable AI (XAI) framework that is least affected by translation and rotation to address these limitations. The pipeline introduces a strict block-shuffle stratified split to establish a leakage-free validation baseline while tested on the UCI wave energy farm dataset. The model maps continuous trigonometric components to eliminate boundary discontinuities by projecting global coordinates onto a wave-aligned polar frame relative to the oncoming wave vector (β_site). Additionally, a non-local macro-feature, the upstream wedge sector density, is engineered to capture forward-facing fluid starvation. The resulting LightGBM architecture achieves a robust predictive accuracy (R² = 0.897) for multi-directional settings. Game-theoretic feature attribution via SHAP deconstructs exact physical interference boundaries, revealing a steep local starvation penalty when neighbors align inline with the wave vector, contrasted against a constructive scattering bonus when staggered. This computationally instant layout blueprint provides a potential deployment pathway for dense, high-yield nearshore arrays across southeast Asia and India's vulnerable coastline. This advances blue-economy initiatives by lowering capital barriers, decentralizing clean electricity for remote coastal communities, and accelerating climate resilience in typhoon-prone, energy-stressed maritime sectors.