基于神经元网络模型的柴油调合非线性预测控制和在线优化

Nonlinear Predictive Control and Online Optimization Based on Neural Network Model in Diesel Oil Blending

针对炼油厂柴油调合生产这个强非线性、多输入、多输出、多扰动、大纯滞后对象 ,提出了以柴油调合倾点RBF神经网络模型为质量指标约束条件的柴油连续调合生产非线性在线最优化方法 ,提出了基于RBF神经网络预测模型、滚动优化目标函数中带有静态经济指标的柴油连续调合倾点非线性预测控制策略。仿真计算证明本文的非线性优化方法能计算柴油调合生产最优配方 ,在满足生产能力约束和产品倾点质量指标的前提下可实现组分油最优利用而获最大利润 ;预测控制策略在预测模型与实际对象模型失配时仍具有良好的跟踪性能和抗干扰性能 ,展示了其良好的工业应用前景。

The diesel oils blending possesses these characteristics:strong non linearity,multi perturbation,MIMO and big delay.Based on RBF model as the constraint of product quality and on its gradient,a nonlinear online optimization strategy applied continual blending in the production of diesel oil is developed.And based on the RBF predictive model and the objective function of roll optimization,in which a steady static optimization is included,a kind of nonlinear predictive strategy controlling the pour point of diesel oil product in the blending process is proposed too.Simulation results show that even if the RBF predictive model and the actual blending process don’t fit properly with their dynamic behaviors,the predictive control strategy still takes on the good performances of both setpoint tracking and anti perturbations.