The application of hybrid neural network architectures can have advantages over stochastic time series analysis techniques and other regression techniques in forecasting. These nonlinear systems can exhibit the ability to represent nonlinear mapping networks, to manage multivariate data, to manage temporal relationships, and to function without the prior selection of a basis function. This Phase I program will identify and characterize potential hybrid neural network architectures which provide significant performance advantages over presently used techniques in forecasting large scale nonlinear systems. In addition, this program will compare the performance of currently used time series techniques with the performance of specialized neural network architectures. This project has strong commercial applications. Nonlinear forecasting systems are particularly well suited for electric power forecasting, econometric modelling, and transportation forecasting. With the addition of a powerful user interface to the optimized nonlinear systems of Phase II, the final system is expected to be very useful to the above mentioned industries. Therefore, follow-on funding is fully expected.