A data-driven double-Gaussian wake model reflecting the wake evolution process

Accurate wake models are vital for optimizing wind farm efficiency, yet conventional analytical models struggle to adapt across operating conditions. This study proposes a data-driven double-Gaussian wake model (DDDGWM) that retains a double-Gaussian physical framework while replacing fixed empirical coefficients with a neural network trained on large volumes of high-resolution LiDAR field data. The network learns a nonlinear mapping from real-time conditions to the six core parameters of the double-Gaussian function. Systematic validation shows that DDDGWM delivers high accuracy and robustness across the full operating range, and markedly outperforms mainstream analytical models in reproducing the complex evolution from near-wake double-peak to far-wake single-peak under high-thrust conditions. Further analyses confirm physical consistency and interpretability: permutation feature importance highlights thrust coefficient, wind speed, and downstream distance as the most influential variables; derived wake-deficit and shape factors capture the mean statistical laws of wake recovery and morphological transition. DDDGWM therefore provides a precise, interpretable, and adaptive framework for refined wake modeling, with practical value for wind farm layout optimization and advanced control.