Machine learning-assisted multiphysics coupling performance optimization in a photocatalytic hydrogen production system

Abstract The application of photocatalytic hydrogen evolution is plagued by low energy conversion efficiency. To achieve the maximum hydrogen production, a new optimization model with partial differential equations (PDE) as constrained conditions is constructed, and the Sobol’ method is employed to quantify the operation condition prioritization and determine the decision variables. To alleviate the computation efforts of daunting optimization problem, the Gaussian process regression (GPR) method is developed to approximate the original optimization problem. A three-dimension multiphysical coupling mathematical model for a photocatalytic reactor is built and validated by the experimental data, which is further employed to investigate the dynamic evolution behaviors of hydrogen yield under different working conditions and compute training samples. A new memetic algorithm integrating the performance advantages of whale optimization (WO) method and simulated annealing (SA) algorithm is developed to search for a high-quality solution. The encouraging results on typical reaction processes clearly imply that the proposed method achieves the coordination of reaction mechanism and operation parameters, reduces the computing complex and load, and successfully finds the optimal operation conditions that will maximize the conversion efficiency of solar energy to hydrogen energy. The research outcomes open new avenues to improve the hydrogen production and accelerate industrial applications of the technology.