Selection of Neural Model Order and Time-Delay for MIMO Non-Linear Systems and a Case Study on a CSTR Process

Abstract A model order and time-delay selection method for multi-input multi-output (MIMO) non-linear systems is developed in this paper based on the selection for the SISO case previously proposed by the authors. The MIMO form of the NARX model is considered and the model order and time-delay for each input are determined by identifying linearised models of the system. The method can be applied to many approximation of a MIMO non-linear system, such as neural network models, fuzzy logic models, etc. A case study on a continuous stirred tank reactor (CSTR) process using this method was investigated to demonstrate the selection procedure. A neural model is developed for the process based on the order and time-delay identified using the proposed method and is compared to other neural models with different structures to demonstrate the effectiveness of the method.