Design of Multivariable Neural Controllers Using a Classical Approach

Abstract

In the present study, the neural network (NN) based multivariable controllers were designed as a series of single input-single output (SISO) controllers or multi variable SISO (MVSISO) controllers utilizing the classical decoupled process models. Multilayer feed forward networks (FFNN) were used as direct inverse neural network (DINN) controllers, which used the inverse dynamics of the decoupled process. To address the disturbance rejection problems, the IMC based neural control architecture was proposed with suitable choice of filter and disturbance transfer function. Multi input - multi output (MIMO) non-linear processes like interacting tank systems, temperature and level control of a mixing tank with hot and cold input streams & a (2×2) distillation process were considered as case studies for that purpose. Simplified as well as ideally decoupled process as well as disturbance transfer functions was used for neural controller design. DINN/IMC based NN controllers performed effectively well in comparison to conventional P/ PI/IMC based PI controllers for set-point tracking & regulator problems.