一种基于RTRL的神经网络驾驶员巡航模型

针对大多数巡航模型未能充分考虑驾驶员的行为特性，文中设计了以实时递归学习算法的神经网络为核心的驾驶员巡航模型。该模型选取前车车速、本车车速、前车加速度和安全车间距共4个参数作为模型输入，以驾驶员控制自车所期望的加速度值为输出，通过真实环境下的巡航实验获取数据样本对RTRL的神经网络进行训练，并对该模型进行仿真验证。仿真实验结果表明，本车期望加速度的预测值与实际真实值基本一致，误差控制在005以内，说明该模型能较准确的模拟驾驶员的巡航行为。

A Driver Car Cruising Model Based on RTRL Neural Network

Most car cruising model could not give full consideration to the drivers behavior characteristic, this paper designed with real time recursion learning (RTRL) algorithm of neural network as the core of the driver model of the cruise. This model chooses four parameters as inputs, including the speed and acceleration of preceding vehicle, the following vehicle speed and the safety distance between two vehicles, meanwhile, the desired acceleration which driver controls cruise car is chosen as the output of model. Through cruise experiments under real environment, data samples used for training RTRL neural network are obtained, the model is verified by simulation. Simulation experimental results show that the predictive value of expected acceleration is consistent with the actual true value, error is controlled within 0.05 , it also shows that the model can accurately simulate the behavior of the driver cruising.