A hybrid expert system-neural networks methodology for anticipatory control in a process environment

A methodology is presented that couples expert systems to neural networks for the purpose of monitoring and control in a process environment. This is achieved within the framework of the anticipatory paradigm. The basic assumption of the anticipatory paradigm is that a complex system can modify its behavior on the basis of present as well as anticipated future states. The complex systems contemplated in this research use measures of performance to represent current as well as anticipated states in such a manner that decisions about change of state are related to a search for maximizing the performance associated with a state variable. Current performance is computed on the basis of a comparison between measured(sensor) and calculated(model) data. Anticipated performance is computed on the basis of predictions made by pre-trained neural networks. A fuzzified Bayes formula is used as the computational link between present and future states. The implementation of such a system calls for a coupling between expert systems(ESs) and artificial neural networks(ANNs). This coupling allows a knowledge-based system to make control decisions through an assessment of the current performance of the process and an estimate of its anticipated performance a &Dgr;t time latter; pre-trained neural networks provide fast estimates of future states.