A surrogate-based algorithm with balanced infill sampling for ignition delay time optimization of a global RP-3 kerosene mechanism

To address the problem of excessive computation in fuel mechanism optimization, as well as to obtain the design results with the highest possible performance, this study developed a Surrogate-Based Algorithm with Balanced Infill Sampling (SBABIS). The computationally expensive combustion simulation is replaced with an inexpensive surrogate. The construction of the surrogate employs a proposed two-mode adaptive infill sampling technique, which allows the balance between local exploitation and global exploration, resulting in a significant increase in the efficiency of optimization. In order to compare the differences between the present algorithm and genetic algorithms, strengthen elitist genetic algorithm (SEGA) with multiple combinations of mutation and crossover probabilities, and an efficient global optimization (EGO) also based on a surrogate model (Kriging with Partial Least Squares, KPLS was employed), were used to optimize ignition delay times of a global RP-3 mechanism. Using the proposed algorithm, the accuracy of ignition delay time calculation of the RP-3 kerosene mechanism is significantly improved from the original error of 101.89%–19.76%, with a reduction of about 70% in the number of samples compared to that of SEGA, and only a difference of about 1.3% between the optimal solution found by the present method and the globally optimal solution found by SEGA. SBABIS also shows much stronger global optimization capacity than EGO.