Modeling human operators using neural networks

Often in the process industries, human operators rather than mathematically based advanced algorithms are used to achieve product control. Typically, if there are three shifts of operators per day, one shift achieves superior control. If the expertise of the best operator can be captured easily and economically and made available to the other operators, significant economic benefits would accrue. This paper discusses a methodology that uses artificial neural networks for capturing the knowledge of process operators. For many operator tasks, only readily available information obtained from a process control computer is required. Once a converged network is available, a stripping technique can be employed to simplify the net and to gain knowledge about what a good operator is doing compared to a poor one. It is felt that the proposed approach is superior to traditional expert system techniques employing knowledge extraction approaches.