Recurrent neuro-fuzzy networks for the modelling and optimal control of batch processes

A recurrent neuro-fuzzy network based strategy for batch process modelling and optimal control is presented. The recurrent neuro-fuzzy network allows the construction of a "global" nonlinear long-range prediction model from the fuzzy conjunction of a number of "local" linear dynamic models. In this recurrent neuro-fuzzy network, the network output is fed back to the network input through one or more time delay units. This particular structure ensures that predictions from a recurrent neuro-fuzzy network are long-range or multi-step-ahead predictions. Process knowledge is used to initially partition the process nonlinear characteristics into several local operating regions and to aid in the initialisation of the corresponding network weights. Process input output data is then used to train the network. Membership functions of the local regimes are identified and local models are discovered through network training. In this paper, a recurrent neuro-fuzzy network is used to model a fed-batch reactor and to calculate the optimal feeding policy.