Maximizing long-term gas industry profits in two minutes in Lotus using neural network methods

Generalized methods that are commonly used in neural-network research have made it possible for the US Energy Information Administration (EIA) to solve a gas-industry optimization problem on a personal computer that would previously have required a mainframe computer because of the run time required. The resulting model was used to produce EIA´s official energy forecasts published in 1988. It is shown how backpropagation can be used by modelers with no special training in neurocomputing. Earlier applications of backpropagation to modeling and to EIA problems are reviewed that antedate the practical applications to neural networks. Finally, the relations between backpropagation, the current EIA model, and economic issues related to modeling and the gas industry are discussed. Among these issues are optimization subject to constraints, and competition and efficiency in gas supply. It is also shown how more recent formulations of backpropagation are a special case of the proposed formulation