基于改进UKFNN和NSGA-Ⅱ的工业过程决策参数稳健优化

Generic hybrid dynamic modeling and robust optimizing of industrial processes using improved UKFNN and NSGA-Ⅱ for performance optimization

针对工业生产过程建模误差的不确定性和最优决策参数的执行误差的不确定性,提出采用改进无迹卡尔曼神经网络(unscented Kalman filter artificial neural network,UKFNN)动态建模保证建模精度;采用改进非支配排序遗传算法(non-dominated sorting genetic algorithm-Ⅱ,NSGA-Ⅱ)稳健优化设计最优决策参数保证执行效果,得到稳定最优输出。采用只需对输入/输出数据进行计算即可得到不可测的未知噪声统计信息的样本有效噪声估计(gamma test,GT)来计算观测噪声统计值,保证UKFNN的建模精度;再采用改进选择算子和交叉算子的NSGA-Ⅱ对工业过程进行稳健优化,得到能够保证系统稳健最优输出的决策参数。最后采用笔者的建模优化方案对氢氰酸生产过程进行实验研究,有效提高了氢氰酸转换率,为噪声不确定工业过程的建模优化提供了一条可行途径。

To get the precise model and the optimal decision parameters of the nonlinear dynamic industrial process with un- certain noise, this paper proved a generic hybrid strategy based on dynamic modeling and robust optimization using the im- proved （unscented kalman filter neural network, UKFNN） and （non-dominated sorting genetic algorithm-Ⅱ, NSGA-Ⅱ ）. It used the UKFNN to model for the industrial process, because it had abilities to adaptively approximate the nonlinear and dy- namic properties of the process. It used the （gamma test, GT） to calculate system measure noise covariance matrix, which only used the input-output parameters, then the precise model could be get by using the UKFNN. It designed the improved NSGA- Ⅱ with the new selection operator and new crossover operator to get the robust optimal decision parameters, which could ensure the optimum value was stable when the decision parameters deviated from optimum set point. The proposed UKFNN and NS- GA-Ⅱ was applied to improve the conversion of real nonlinear dynamic hydrocyanic acid （HCN） industrial process. Numeri- cal simulations show that the proposed approach is valid to obtain the high-precision dynamic modeling and the robust optimal parameters, improve the conversion of HCN. Therefore, provide a new solution to get the dynamic model and the optimal pa- rameters for the complex nonlinear dynamic system with uncertain noise.