在线场景更新的多阶段非线性模型预测聚合反应控制

针对聚合过程中时不变不确定性参数不能直接估计的情况,导致的多阶段非线性模型预测控制中场景树生成的合理性问题,提出一种基于贝叶斯概率加权的在线场景更新算法.该方法利用前一时间步中每个场景的模型预测信息和过程状态测量信息计算对应场景的概率权重,然后通过合适的自适应步长在线更新场景树中不确定性的离散实现场景.所提方法在保证过程约束满足的同时,逐渐缩小不确定性集合逼近不确定性的真实值,从而降低保守性,提升控制器性能.通过多个批次的半间歇聚合反应过程实例仿真结果表明,所提出的方法可以有效降低批次反应时间,提高生产效率.

Multi-stage nonlinear model predictive polymerization reaction control with online scenario update

For the case where the time-invariant uncertainty parameter cannot be estimated directly in polymerization process, an online scenario update algorithm based on Bayesian probability weighting has been proposed to generate scenario trees in multi-stage nonlinear model predictive control (NMPC). The model prediction information and process state measurement information of each scenario in the previous time step are used to calculate the probability weight of the corresponding scenario. Then, using an appropriate adaptive step to update the uncertainty discrete scenario in the scenario tree. While ensuring that the process constraints are met, the uncertainty set is gradually reduced to approach the actual value of uncertainty, thereby reduce conservatism and improve controller performance. The simulation results of multiple batches of semi-batch polymerization reaction process examples show that the proposed method can effectively reduce the batch reaction time and improve the production efficiency.