Use this URL to cite or link to this record in EThOS:
Title: Modelling and forecasting UK organic coated steel demand using neural networks
Author: Rees, R. H.
Awarding Body: University of Wales Swansea
Current Institution: Swansea University
Date of Award: 2000
Availability of Full Text:
Access from EThOS:
The aim of this research was to develop a practical model capable of medium to long-term forecasting of organic coated steel (OCS) demand in the United Kingdom. Previous methodologies in steel forecasting were reviewed as well as artificial neural network forecasting techniques. The theory of neural networks and backpropagation of error is described in detail along with genetic algorithm optimisation techniques. Two neutral software packages, Neuralyst and Trajan, which incorporated genetic algorithms were reviewed and Trajan selected to build a forecasting model of OCS demand. Market analysis was carried out to find the key determinants of OCS demand. This identified several candidate variables that could be incorporated into a neural network model of OCS demand. A genetic algorithm technique was adopted to select the most appropriate variables as inputs to the neural network models. Three neural network models were constructed using 12, 8, and 3 explanatory variables as network inputs in a bid to find the simplest network without compromising forecast accuracy. Two univariate neural network models were also constructed for comparison purposes. The networks were trained using the conjugate gradient descent algorithm and incorporated a verification set to avoid overfitting. A further analysis of network weight distributions was carried out to weed out overtrained models. The study concludes that neural networks are robust and are capable of forecasting OCS demand to within a mean absolute percentage error of 7% over an eight quarter period whilst also being able to predict longer periods for scenario forecasting. The model is also easy to apply requiring little statistical or econometric model building knowledge and is therefore easy to implement and update as new data becomes available.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available