Neural Network Modeling of Heat Exchanger

The controlling of heat exchanger using conventional PID controller always face the problem of having limiting performance due to unpredictable unsteady state thermal behavior of heat exchangers. In recent decades, the applications of neural network for thermal analysis of heat exchangers have been extensively studies by great amount of researchers and institutions. The non-linear characteristic of neural network is deemed to possess great potential to accurately predict heat exchangers performance. A multilayer feedforward network has been chosen as the base network architecture to construct a neural network model for heat exchanger. The network has been trained by using backpropagation algorithm to obtain simulated result. The result was compared with the raw data. The network architecture without partitioning of data was fully developed using MATLAB simulation which number of neurons and transfer functions at each layer were determined before a neural network model can be constructed in Simulink Model. The neural network developed in Simulink will be integrated with PID controller based on Internal Model Control model-based design strategy. The tuning relations used for the PID controller settings are now set to be Cohen Coon, ITAE and IAE tunings. The result of this research will help to improve the prediction of heat exchanger performance.