GA学習形ニューラルネットワーク同定器を用いたトンネル機械の位置推定と制御

概要

論文の詳細を見る
A method has been developed for constructing a neural control system that can be used to control the machines used for tunneling underground conduits for telecommunication optical-fiber cables. The neural control system is based on a feedback controller whose gains are tuned adaptively using neural networks. The tunneling machine's displacement, which is used as a feedback signal, is estimated using a direct neural identifier. To avoid an undesirable local minimum problem within the learning process of the neural identifier, a learning algorithm is proposed that combines both genetic algorithm and back propagation. Simulation using a nonlinear system confirmed the algorithm's characteristics; identification results using the experimental data of a tunneling machine showed its usefulness for a practical system. Numerical simulation of the proposed neural control system confirmed its feasibility and showed that the direction of the tunneling machine can be controlled for various construction trajectories and soil types using the neural network's learning ability.
1996-12-25