Multi-objective optimization of turbine blade profiles based on multi-agent reinforcement learning

The aerodynamic design level of the blade profile directly affects the overall energy conversion efficiency of the turbine. However, the optimization process of the blade profile is a typical multi-objective and multi-constraint optimization problem. Traditional optimization algorithms tend to fall into local optima and have slow solving speeds when dealing with these types of problems. To address these issues, this study proposes a dynamic multi-objective optimization algorithm based on multi-agent reinforcement learning (DMORL). This algorithm describes the aerodynamic performance optimization process of the blade as a Markov decision process and employs a multi-agent collaborative optimization strategy to parallelize the solution for different optimization objectives. After the model training is completed, it can provide the Pareto front in real time under different geometric constraints and airflow incidence angles, accomplishing dynamic multi-objective optimization of the blade profile. Experimental results demonstrate that, compared to traditional multi-objective optimization algorithm (NSGA-II), DMORL can find a better Pareto front, with an average solving time of only 0.12 s per multi-objective optimization problem, improving optimization speed by 51 times.