基于深度残差网络的智能换挡控制策略

Intelligent shift control strategy based on deep residual network

针对传统自动变速器挡位决策系统人机交互较差的问题,提出了一种基于深度残差网络的智能换挡控制策略。将节气门开度、车速及加速度输入多个自编码器对神经网络的权重进行无监督地预训练,重组自编码器的隐藏层并加入残差连接建立起深度残差网络。利用实车数据对深度残差网络进行微调,建立起基于深度残差网络的智能换挡控制策略。实验结果表明,相比基于传统人工神经网络的方法,该策略在实车工况下挡位识别率更高,为99.4%。

Aiming at the problem of poor human-computer interaction in the traditional automatic transmission gear decision system,an intelligent shift control strategy based on deep residual network is proposed.Firstly,the throttle percentage,vehicle speed and acceleration are input into several automatic encoders,and the weights of the neural network are pre-trained unsupervised.Finally,the hidden layers of several automatic encoders are reorganized and residual connections are added to construct a depth residual network.Using real vehicle data to fine-tune the depth residual network,an intelligent shift control strategy based on the depth residual network is established.Experimental results show that this strategy can achieve a higher gear recognition rate of 99.4%compared to the method based on traditional artificial neural network under actual vehicle conditions.