Inferential Development ofMLNG Depropanizer Bottom Product

This is an individual Final Year Project titled as 'Inferential Development for MLNG
Depropanizer Bottom Product' which carries four credits hours.
The main objective of this research project is to develop an appropriate inferential
model to predict the quality of a Depropanizer bottom product that consists ofbutane
and propane. In this research project, neural network technique was employed to predict
the property of the Depropanizer bottom product. There were twenty seven inputs and
one output used to develop the neural network model. This research project was carried
out in conjunction with MLNG whereby data were collected from the plant to construct
the network and training itto perform the property prediction. The software used for this
project is Matlab 6.1 especially neural network toolbox and Microsoft Excel.
The neural network used was of 'Feed Forward Backpropagation' type and suitable
configuration was tested and analyzed to achieve a minimum number of prediction
error. For this project, the error calculation used was Root Mean Square (RMS). The
network model were developed with the configuration of 3 layers which consist of 36
neurons in the first layer, 27 neurons inthe second layer and 1neuron inthe third layer.
The training function used for this network is 'Trainrp' and the adaptation learning
function is 'Learngdm'. This network was trained with 100 times iteration. The model
can be considered accurate to predict the concentration of the propane at the
Depropanizer bottom product with RMSE obtained at 5.36%.