HAD-BO: A history-aware dynamic Bayesian optimization strategy and its applications in laser-driven plasma high-harmonic generation

An enhanced Bayesian optimization method, named History-Aware Dynamic Bayesian Optimization (HAD-BO), is proposed and applied to optimize the ellipticity in laser-driven plasma surface high-harmonic generation (SHHG). While Bayesian optimization is a cornerstone of black-box function optimization, established methods like the Gaussian Process Hedge (GP-Hedge) algorithm often fail to effectively leverage historical performance when selecting acquisition functions. This results in suboptimal strategies and unstable convergence. To address this, HAD-BO introduces a global, time-series-based dynamic adjustment mechanism that integrates historical gain updates with a weight decay factor, significantly improving the flexibility and adaptability of acquisition selection. We first validate HAD-BO’s effectiveness through numerical experiments on classical benchmark functions and then demonstrate its performance in a three-dimensional parameter optimization of SHHG ellipticity. The results indicate that HAD-BO not only accelerates convergence but also achieves superior stability in complex and multi-dimensional systems. This work provides a novel and efficient optimization tool for complex physical systems.