Abstract: In view of large industrial processes, an alterable horizon predictive control algorithm based on reduced state-space decomposition is proposed. Operated in rolling-horizon optimization form, the size of optimization problem is decreased through the decomposition of interior and exterior states in each rolling horizon, on that basis, the coordinated optimization problem between multi-time-scale sub-systems is skillfully solved by using alterable horizon method. Simulation experiment is carried out using the actual production data from a certain iron works, and comparisons with three existing predictive control algorithms are made. Results show that this method deals extraordinarily well with performance index and real-time computation ability, and has the best overal performance.

Key words: multi-time-scale, decentralized predictive control, coordinated optimization

摘要： 针对大型工业生产过程，提出了一种基于约减状态空间分解的可变预测窗口长度预测控制算法。采用预测控制滚动优化的方法，在每个滚动窗口内，通过内外部状态分解减小优化问题求解规模，在此基础上，采用可变时间窗口长度的方式巧妙地解决了多时间尺度子系统之间的优化协调问题。以某炼铁厂的实际生产数据为基础进行了仿真实验，并与现有的三种预测控制算法进行了比较。结果表明，该方法很好地处理了计算实时性和优化性能之间的矛盾，综合性能最优。

关键词: 多时间尺度, 分散化预测控制, 协调优化