THE USE OF CONNECTIONIST SYSTEMS TO RECONCILE INCONSISTENT PROCESS DATA

Since measurements of variables in chemical and metallurgical plants generally violate the conservation and other constraints of these systems owing to random measurement errors, these data have to be reconciled with the constraints prior to further use. In multicomponent systems the reconciliation of process data normally results in a non-linear constrained optimization problem, which can constitute a formidable computational burden when large systems have to be solved by conventional techniques. Connectionist systems, such as artificial neural networks, can be implemented to considerable advantage for the solution of optimization problems such as these and in this paper their use is explored. Three variants of crossbar feedback connectionist systems were investigated, two are based on gradient descent techniques and one on a direct search method. The results of simulations, as well as a comparison with traditional computational procedures, indicate that systems such as these based on gradient descent techniques can be used to solve large systems efficiently.