Trajectory tracking with dynamic neural networks

Abstract

The application of artificial neural networks to dynarmcal systems has been constrained by the nondynamical nature popular network architectures. Many of the difficulties that ensue-large network sizes, long training times, the need to predetermine buffer lengthscan be overmmed with dynamic neural networks. The minimization of a quadratic performance index is considered for trajectory tracking or process simulation applications. Two approaches for gradient computation are discussed: forward and adjoint sensitivity analysis. The computational complexity of the latter is significantly less, but it requires a backward integration capability We also dlscuss two parameter updating methods: gradlent descent and a Levenberg-Marquardt approach.