基于深度学习多网络融合的网联车辆驾驶行为预测

Prediction model of intelligent connected vehicles driving behavior based on deep learning multi network fusion

智能网联车在复杂道路场景中安全行驶需要对周围车辆驾驶意图和行驶轨迹的准确预测。将双向长短期记忆网络(Bi-directional Long Short-Term Memory)、卷积生成对抗网络(Deep Convolutional Generative Adversarial)和注意力机制(Attention Mechanism)相融合,提出一种基于BLSTM-DCG-ATT的网联车辆驾驶意图和行为预测模型。通过正反双向LSTM链路和注意力机制得到具有双重特征的数据,通过卷积生成对抗网络对特征数据进行卷积处理,迭代生成网联车辆及周围车辆的未来时刻变道意图和行驶信息。仿真结果表明:该模型能够在复杂路网、交通流密集的情况下,对网联车辆及其周围车辆的变道意图和行驶轨迹进行准确预测,预测精度达94%。

The accurate prediction of driving intention and driving tracks of surrounding vehicles is the basis to ensure the safe driving of Intelligent Connected Vehicles in complex road scenes.This paper proposes a driving intention and behavior prediction model of intelligent connected vehicles based on BLSTM-DCG-ATT by combining Bi-directional Long Short-Term Memory,Deep Convolutional Generic Adversary and Attention Mechanism.The data with dual characteristics are obtained through the forward and reverse Bi-directional LSTM link and attention mechanism,and then the characteristic data are convolved through the Deep Convolutional Generic Adversary to iteratively generate the lane change intention and driving data of the intelligent connected vehicle and surrounding vehicles in the future.The simulation results show that the model can accurately predict the lane change intention and driving track of the intelligent connected vehicle and its surrounding vehicles under the condition of complex road network and dense traffic flow,and the prediction accuracy reaches 94%.