Minimising waste generation using neural networks based dynamic optimisation

Abstract

In this paper, a novel optimisation method that combines iterative dynamic programming, IDP, (Luus,1989) and neural network (NN) models is analysed in the framework of the minimisation of waste generation in a discontinuous reactor. Although IDP has shown to be effective in solving difficult batch reactor optimisation problems, it requires precise knowledge of the kinetic mathematical model. In cases where the systems are highly nonlinear, neural networks have the ability to self adjust their parameters to match input -output mapping rules of unmodeled systems. So, combining dynamic optimisation and neural network models allows to optimise dynamically those complex systems whose kinetic models are difficult to represent mathematically, as it happens in many industrial cases. In such cases, neural networks offer the possibility of using data obtained from the original process to model it.