基于神经网络的多变量协调预估控制器

A COORDINATED PREDICTIVE CONTROLLER BASED ON NEURAL NETWORK

针对操作变量多于被控变量的“胖”系统，提出了一种多变量协调控制系统．本控制系统在保证对关键变量实施定点控制的同时，还尽可能将操作变量协调至理想操作区内,以达到优化操作变量的目的。文中将生产过程中复杂的操作要求和优化规则归纳整理为一系列多变量优化协调逻辑,并进而用人工神经元网络予以实现，用预估控制技术实现控制．这种协调控制器结构充分发挥了神经元网络和预估控制算法两方面的优势．在石油分馏塔上实现的多变量协调预估控制及仿真结果说明了本方案是可行的,易于在生产过程中应用.

A Multivariable coordinated predictive control system is presented for process systems with more inputs than outputs. This control system can not only maintain key PVs (Process Variables) at their set points but also coordinate some MVs (Manipulate Variables) to their ideal resting regions so as to optimize industrial process. In this control system a neural network is used to generate comprehensive optimal coordinated logic for industrial processes, and the controls are implemented by a predictive control algorithm. Based oil tills control strategy a multivariable coordinated control system is developed for the top temperature control of a petroleum distillation column. The feasibility and applicability of this control systems is shown by the simulation results of a case study.