Neural network based approach for optimization of industrial chemical processes

Process optimization involves the minimization (or maximization) of an obje ctive function, that can be established from a technical and/or economic vi ewpoint. In general, the decision variables are subject to constraints such as Valid ranges (max and min limits) as well as constraints related to saf ety considerations and those that arise from the process model equations. U sually in chemical engineering problems, both the objective function and th e constraints are non-linear. Computational methods of non-linear programmi ng with constraints usually have to cope with problems such as numerical ev aluation of derivatives (Jacobian, Hessian) and feasibility issues. The bas ic idea of the optimization method using neural network (NN) is to replace the model equations or plant data by an equivalent NN, and use this NN to c arry on a grid search on the region of interest. As an additional benefit, the full mapping of the objective function allows one to identify multiple optima easily, an important feature not presented by conventional optimizat ion methods. Moreover, the constraints are easily treated afterwards since the points with violated constraints can be recognized and classified (acco rding to weak or hard constraints). This approach was applied in some indus trial chemical process: the process of nylon-6,6 polymerization in a twin-s crew extruder reactor and an acetic anhydride plant. (C) 2000 Elsevier Scie nce Ltd. All rights reserved.