全息搜索策略的改进及在参数优化中的应用

Estimating reaction kinetic model parameters based on holographic research strategy

反应动力学模型往往是非线性多极值函数,对其参数优化,传统的确定性优化算法易陷入局部最优。全息搜索策略(HRS)是一种寻优效率较高的确定性优化算法,但只能用于离散系统的优化。通过对连续变量进行离散化处理,并运用迭代计算逐步缩小离散系统与原连续系统的偏差,将复杂的多维连续变量优化问题转化为多个串联的较为简单的离散变量组合优化问题,使HRS只需在有限个离散解中寻优,实现HRS的连续变量优化,并在HRS确定性寻优过程中引入随机性交叉操作,以进一步提高算法的全局搜优效率,由此建立了一种改进的HRS。八维Alpine函数测试表明,改进的HRS的全局优化性能明显优于常规遗传算法(SGA),将其应用于2,3-二氟-6-硝基苯酚的合成反应动力学参数优化,得到的动力学方程与原方程相比,拟合偏差平方和减小了17.2%,验证的平均相对误差由5.37%降至3.44%,参数优化结果较为满意。

Reaction kinetic models usually are nonlinear and multi-extremal function, the traditional deterministic optimization methods easy to fall into local optimum for its parameters estimation. Holographic research strategy （HRS） is a determinate optimization method proposed newly. The principle of HRS is based on a special, two-dimensional presentation of a multidimensional space. This presentation was termed two-dimensional hologram. HRS translated the optimization operation in multidimensional space into finding better points in the neighborhood around the current best data points. In this way, HRS can find the global optimal parameters in all probability. However, HRS can＇t be applied to optimize continuous variables, it was only used in optimizing discrete systems. HRS was improved in this paper. The improved HRS changed continuous variables into discrete variables in the searching region firstly, and then found the optimum in the discrete system. In order to reduce the deviation between the continuous system and the discrete system, it adopted iterative algorithm to shrink the searching region gradually according to the location of the current optimal value. Furthermore, in order to improve the efficiency of HRS in searching for the global optimum, random crossover operator was added to the optimizing process. Eight-dimensional Alpine function was applied to testing the improved HRS. The testing results demonstrated that its global optimization performance is superior to simple genetic algorithm （SGA）. Further, the improved HRS was applied to estimate the kinetic model parameters of 2,3-difluoro-6-nitrophenol. Compared with the original equation, the fitting deviation square sum of the kinetic equation obtained by the improved HRS reduced by 17.2%, and the verifying average relative error reduced from 5.37% to 3.44%, parameter optimization with satisfactory results.