针对由被控对象非线性和优化目标函数非凸性带来的建模与实时优化问题求解的困难,提出一种基于相关向量机(RVM)的非线性多步模型预测控制算法.采用RVM建立非线性预测模型,并将差分进化算法引入非线性预测控制中发挥其全局最优、鲁棒、快速收敛等优点,在线求解多变量、多约束的非线性规划问题.利用实际生产数据进行聚丙烯牌号切换仿真,结果表明,该算法可大幅度减少切换时间,降低过渡料产量,提高经济效益. Aiming at the difficulty of solving the modeling and real-time optimization problems caused by the nonlinearity of the controlled object and the non-convexity of the optimization objective function, a nonlinear multi-step model predictive control algorithm based on the correlation vector machine (RVM) is proposed. Nonlinear predictive model, and introduces the differential evolution algorithm into the nonlinear predictive control to take advantage of its global optimal, robust and fast convergence, online solving multivariable and multi-constrained nonlinear programming problems.Producing polypropylene The simulation results show that this algorithm can significantly reduce the switching time, reduce the output of transition materials and increase the economic benefits.