复合粒子群优化算法在模型参数估计中的应用

化工非线性模型的参数估计是较为困难的寻优问题，经典方法常会陷入局部极值。粒子群算法操作简便、容易实现且全局搜索功能较强，适用于非线性参数估计。但其参数值的确定与问题相关，若设定不当，会严重影响全局搜索的性能。今提出引入遗传算法，在粒子群算法的搜索过程中，逐代优选参数，包括惯性权值，加速常数，以此构建为复合粒子群优化算法。分析与测试表明，其全局搜索性能有显著改善。进一步的工作又将两种粒子群算法成功地应用于重油热解模型的参数估计。采用复合粒子群优化算法估计参数构建的重油热解模型，其预报相对误差比常规粒子群优化算法降低了8.97％，比简单遗传算法降低了23.21％，效果明显。

Estimation of Model Parameters Using Composite Particle Swarm Optimization

Estimation of nonlinear model parameters in chemical engineering is a tough searching problem. Unfortunately, the traditional approaches easily get stuck in a local minimum. Considering that the particle swarm optimization (PSO) algorithm is quite simple and easy to implement, it was used to estimate the nonlinear model parameters in this paper. However, PSO needs several particular control parameters, such as inertia weight and acceleration constants, which are usually problem dependent and affect the PSO performance significantly. In order to overcome these troubles, a composite particle swarm optimization (CPSO) using simple genetic algorithm (SGA) to optimize the control parameters was proposed. Two benchmark functions illustrate that the performance of both CPSO and PSO are better than SGA, in particular the CPSO is extremely effective. Finally, the two types of PSO were applied successfully to the nonlinear parameter estimation of heavy oil thermal cracking model. The model used CPSO algorithm reduces the relative prediction error about 8.97% than PSO algorithm, and 23.21% than SGA.