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 short-term memory network

摘要:

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层卷积神经网路输入,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,经过无数次模型训练后,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,能够得到较为理想的高速公路车辆变道轨迹预测结果。

Abstract:

The vehicle speed, distance, driving direction and other factors on the expressway are all dynamic changes, and the collected target vehicle status feature data may have noise due to the interference of the external environment, which leads to errors in the prediction of vehicle lane change trajectory. Therefore, a prediction model of expressway vehicle lane change trajectory based on long-term and short-term memory network is proposed to effectively predict the lane change trajectory of expressway vehicles, improve vehicle driving conditions, and ensure their safe operation. Collect target vehicle traffic data through laser radar, GPS and other devices, reasonably combine them into target vehicle state observation eigenvectors, and construct corresponding eigenvector matrix. The constructed target vehicle state observation eigenvector matrix is used as the input of 1-layer convolutional neural network. After extracting the potential characteristics of target vehicle state observation eigenvectors, the output results of 1-layer convolutional neural network are effective input of two-way short-term memory network, After countless model trainings, the predicted lane change trajectory of the target vehicle is output. The experimental results show that the model can effectively predict the lane changing trajectory of highway vehicles, and the predicted trajectory has extremely low horizontal and vertical coordinate errors, which can obtain ideal prediction results for lane changing trajectories of highway vehicles.