基于长短期记忆网络的高速公路车辆变道轨迹预测模型

Lane Change Trajectory Prediction Model of Expressway Vehicles Based on LSTM Networks

高速公路车辆车速、车距、行驶方向等因素都是动态变化的,受外界环境干扰,采集到的目标车辆状态特征数据可能存在噪声,导致车辆变道轨迹预测存在误差,为此提出基于长短期记忆网络的高速公路车辆变道轨迹预测模型,有效预测高速公路车辆变道轨迹,改善车辆行驶条件,保障其安全运行;通过激光雷达、GPS等装置采集目标车辆交通数据,将其合理组合成目标车辆状态观测特征向量,并构建相应的特征向量矩阵,将所构建目标车辆状态观测特征向量矩阵作为1层卷积神经网络输入,提取目标车辆状态观测特征向量潜在特征后,以1层卷积神经网络输出结果为双向长短期记忆网络有效输入,经过模型迭代训练后,输出目标车辆变道轨迹预测结果;实验结果表明:该模型可有效预测高速公路车辆变道轨迹,预测出的轨迹横纵坐标误差极低,预测耗时较短;能够得到较为理想的高速公路车辆变道轨迹预测结果。

The factors of vehicle speed,distance,and direction of travel on highways are all dynamic changes and interfered with external environments.The collected data on the state characteristics of target vehicles perhaps contains noise,leading to errors in predicting lane-changing trajectories.To address this issue,a highway vehicle lane-changing trajectory prediction model based on long shortterm memory(LSTM)networks is proposed to effectively predict lane-changing trajectories on highways,and improve driving conditions to ensure safe operation.The target vehicle traffic data is collected by using devices such as laser radar and GPS.These da-ta are reasonably combined into the target vehicle state observation feature vectors and corresponding feature vector matrices.The constructed target vehicle state observation feature vector matrix is used as an input to a 1-layer convolutional neural network to ex-tract the latent features of the target vehicle state observation feature vectors.The output result of the 1-layer convolutional neural network is then used as the effective input to a bidirectional LSTM network.After the iterative training of the model,the predicted lane-changing trajectories of the target vehicle are obtained.The experimental results show that the model can effectively predict the lane-changing trajectories of highway vehicles with extremely low errors in both horizontal and vertical coordinates.Moreover,the prediction time is relatively short.It can achieve desirable prediction results for the lane-changing trajectories of highway vehicles.