Identification of industrial air compression system using neural network

The existence of random variable in any industrial process is basically unavoidable. It occasionally creates nonlinearity behavior of a system and makes predictive control complicated. Such a random behavior must not be ignored as it may indicate any unknown event occurring during the process. System identification is an approach to construct the mathematical model of a dynamical system using the instrumentation signal of input and output of the system. This study performs system identification by using the NARX model as a base model with the nonlinear functions of a neural network for an industrial air compression system. The identification undergoes a series of analysis (number of neuron, delay and data division) to determine the most suitable NARX-NN model architecture configuration before coming up with a final model. Finally, the validation of model's predictive performance is carried out through several analyses, namely, mean square error and regression value. The predicted data are compared to the industrial data to verify its accuracy which shows that the final model had successfully ruled out the suspicious random event data.