自适应混合粒子群约束优化算法及其在软测量模型参数估计中的应用

摘要

Tranditionally the problems of parameter estimation of soft sensing model are solved by using the traditional gradient methods to optimize the nonlinear least squares models. However, there exist the disadvantages of the dependence on the initial solution, the requirement of additional trend analysis to verify model correctness and the insufficiency in solving complex problems. Therefore, a new idea of transforming the original parameter estimation problem into a constrained optimization problem and using a hybrid optimization algorithm to solve it is developed. An adaptive hybrid particle swarm optimization algorithm for constrained optimization problems (AHPSO-C) is proposed. In AHPSO-C, an update strategy of maximum function evaluations for the interior-point method (IPM) is imported to balance the global search (chaotic particle swarm optimization) and local search (IPM). The simulation results for the 12 classic benchmark functions indicate that AHPSO-C is an effective algorithm for solving constrained optimization problems. The feasibility and superiority of the method is illustrated with the challenging parameter estimation of soft sensing model for melt index in a cascade slurry HDPE reaction process.%对于软测量模型参数估计问题，针对传统梯度法求解非线性最小二乘模型时依赖初值、需要追加趋势分析进行验证和无法直接求解复杂问题的缺陷，提出将参数估计化为约束优化问题，使用混合优化算法求解的新思路。为此提出一种自适应混合粒子群约束优化算法（AHPSO-C）。在 AHPSO-C 算法中，为平衡全局搜索（混沌粒子群）和局部搜索（内点法），引入自适应内点法最大函数评价次数更新策略。对12个经典测试函数的仿真结果表明，AHPSO-C是求解约束优化问题的一种有效算法。将算法用于淤浆法高密度聚乙烯（HDPE）串级反应过程中熔融指数软测量模型参数估计，验证了方法的可行性与优越性。