Abstract:
Traditional optimization methods, such as simplex method, steepest descent method, and conjugate gradient method, have the advantages of high computational efficiency, strong reliability, and mature theory. However, it is difficult for them to solve the optimization problem without mathematical model. Moreover, for many high-dimensional optimization problems, they easily fall into the local minimum. Due to the characteristics of strong nonlinearity, high dimension, long computation time, and high precision, the parameters identification problem of the first-order reaction equation in Kumar model cannot be solved effectively by means of traditional optimization algorithm. In recent years, various stochastic optimization algorithms such as genetic algorithm and differential evolution algorithm have rapidly developed. Among them, differential evolution algorithm(DE) shows the strong ability to nonlinear optimization problem, due to its advantages of fast convergence, few control parameters, and simple parameter setting principle. Nevertheless, the standard DE algorithm and its improved algorithm have some shortcomings, e.g., slower convergence speed, easy falling into the local minimum, etc. Aiming at the above problems, this paper proposes a new classification variation parameter adaptive differential evolution algorithm (CVPADE). By combining the classification variation strategy with the control parameter adaptive evolution strategy, CVPADE algorithm can not only increase the convergence speed, but also avoid the shortcoming that the population diversity is not rich enough due to the small population size. Besides, the adaptive evolution strategy of parameters is controlled to make the algorithm more stable and avoid falling into local minimization. It can effectively solve the selectivity coefficient identification problem of Kumar model primary equation. Finally, the comparison with the standard differential evolution algorithm and the control parameter adaptive differential evolution algorithm (jDE) are made, showing that the time consumption of the proposed algorithm in this paper is 40% and 35% lower than that of DE and jDE, respectively. Hence, the proposed algorithm has a significant improvement effect on the parameter identification of the first-order reaction equation in Kumar model.