結合の忘却によるニューラルネットワークDYC制御器からの制御則の抽出

概要

論文の詳細を見る
Recently, direct yaw moment control (DYC) using active brake has been applied on some vehicles. Its control law is determined from result of nonlinear simulation analysis. But, choosing significant inputs to DYC controller is difficult because of nonlinearity of tire-road force. The neural network (NN) can find control law of nonlinear system, but to understand the meaning is difficult because of its complex structure. In this paper, rule extraction from NN controller is presented. The NN controller is obtained by using nonlinear mathematical simulation of double lane change with genetic algorithm (GA). By addition forgetting genetic operator, the connection of NN are reduced and simplified. As a result, simple control law of DYC is achieved. The control law can be implemented by various means because of its simplicity. The DYC control law using side slip angle and time derivative of side slip angle is achieved. This control law has good performance and uses the same inputs as the DYC controller that was obtained by conventional analysis method. As a result, the effectiveness of rule extraction from NN is presented.
2001-09-25