The preparation and analysis of input and model data was carried out. The
importance of the correct technique of data filtering was highlighted with
particular focus being emphasised on the removal of outliers in raw data.
An important process in the use of Artificial Neural Network (ANN) models
was identified as being the selection of the right input variables.A comparison
between using factor analysis and statistical analysis in the selection of
inputs and it was observed that the former gave significantly better results.
The training and testing phase of Artificial Neural Network (ANN) model
development was shown to be an important step in Artificial Neural Network
(ANN) model development. If this phase was wrongly done then the ANN
model would not be accurate in its predictions.
Optimisation of the ANN model architecture was carried out with the amount
of hidden layers, amount of neurons in the hidden layers, the transfer function
used and the learning rate identified as key elements in obtaining an
Artificial Neural Network (ANN) architecture that gave fast and accurate
Fresh water addition and demulsifier addition were identified as key parameters
in the economic performance of the desalting process.
Due to a scarcity of water and the high cost of the demulsifier chemical it
was important to try and optimise these two input variables thus reducing
the cost of operations.
A Doctoral Thesis. Submitted in partial fulfilment of the requirements for the award of Doctor of Philosophy of Loughborough University.