Evolution of Chemical Process by Using Neural Networks

本研究的目的是利用柔性神经网络和自学习机制来研究化工厂的进化。分布式和协调神经网络采用多个面向SISO过程的PENN控制器协同工作。所提出的神经网络控制器可以自动识别MIMO过程中控制变量之间的相互作用。此外，采用模型预测方法对长死区过程进行控制。通过这种方案，即使在复杂的化学过程控制中，自学习机制也变得非常有效。利用神经网络实现了聚苯乙烯本体聚合在不稳定区域内的状态预测。指示有关流程的一般信息的全局策略由几个不同的规则组成。此外，利用该过程的近似数学模型得到了详细的全局策略。该方案显著提高了建模能力。将所获得的过程模型作为正逆模型应用于控制过程，实现了良好的控制。支持分布式和协作系统的调度问题的解决方案。随着间歇式化工厂的广泛发展，组合问题经常出现。车间作业调度问题是典型的组合问题之一。迅速达成一项有效和实际的解决办法有望广泛提高生产率，并导致能源消耗的减少。提出了将遗传算法与机械搜索机制相结合的解决方案。与现有方法相比，该方法具有较强的寻优能力。

The aim of this research is the evolution of chemical plant by adopting flexible neural network and self-learning-mechanism as follows :1.Distributed and coorinated neural networkSeveral number of PENN controllers for SISO process are made to work together. The interaction among control variables of a MIMO process can be automatically recognized by the proposed NN controller. Furthermore, a process with long dead time is controlled by adopting model prediction method. By this scheme, The self-learning-mechanism becomes very effective even in the control of complex chemical prosesses.2.Progress on characteristics of neural networkThe state prediciton of bulk polymerization of polystirene within an unsatble region is achieved with PENN.The global policies that indicate the general information on the process consist of several distinct rules. Moreover, the approximated mathematical model of the process is utilized to get detailed global policies. In this scheme, the ability of modeling is significantly improved. The acquired process model is applied to control the process as forward and inverse model, and excellent contorl is achieved.3.A solution of scheduling problems supporting distributed and cooperative systemCombination problem is seen quite often as batch chemical plants grow extensively. The job-shop scheduling problem is one of the typical combination problems. The prompt achievement of an efficient and practical solution is expected to make extensive improvement of productivity and to lead to the decrease in energy consumption. The solution combining GA (Genetic Algorithm) with mechanical searching mechanism is proposed. This solution has high ability in searching preferable solutions than the existing method.