Neural Network Application to Multi-Step Prediction for Generalized Heave Displacement of Shipborne Helicopter Platform

A back propagation algorithm is a supervised learning algorithm applied to train a neural network off line. This paper introduces a modified learning algorithm. The training algorithm, which is composed of a temporal difference (TD) method and a dynamic BP algorithm (DBP), can train an Elman network online. A gradient descent momentum and adaptive learning rate algorithm is applied to the TD-DBP algorithm. The modified TD-DBP training algorithm increases training speed and stability effectively. Using the collected real time data, the simulation suggests the modified TD-DBP learning algorithm is able to generate multi-step prediction for generalized heave displacements of a shipborne helicopter platform.