基于神经网络算法的驾驶员制动需求预测研究

Research on Driver Braking Demand Prediction Based on Neural Network Algorithm

由于线控制动系统在结构上的解耦关系,驾驶员的制动需求识别成为线控制动系统研究中的焦点。本文建立了一种基于动态时间规整DTW (Dynamic Time Warping)算法和长短时记忆模型LSTM (Long-Short-Term Memory)融合的驾驶员制动需求识别模型。该模型主要包括数据收集、数据处理、分类匹配、需求预测四个模块。在搭建的线控底盘实验台上进行了实验,采集了大量的驾驶员制动数据,数据经过处理后首先利用动态时间规整算法进行驾驶员制动习惯分类匹配,然后将分类后的数据分别用长短时记忆模型进行训练,在完成训练后对模型性能进行了测试。同时我们还将本文所建立的模型与其它方法进行了对比实验,结果表明,本文所提出的模型能够准确地对不同驾驶习惯的驾驶员实现高准确度的制动需求预测。

Due to the structural decoupling relationship of the brake-by-wire system, the identification of the driver’s braking demand has become the focus of research on the brake-by-wire system. This paper establishes a driver’s braking request recognition model based on the fusion of dynamic time warping algorithm and long-short-term memory model. The model mainly includes four modules: data collection, data processing, classification matching, and request predication. The experiment was carried out on the test bench of chassis controlled-by-wire, and the driver braking data was collected. Using the conditioned sample data, dynamic time-warping algorithm was used to classify and match the driver’s braking habits, and then the classified data is trained with the long- short-term memory model. The model is validated after the training is completed. The proposed prediction model performance is compared with other approaches and the effectiveness is verified with the expected driver brake habit as per driving situations. The results show that the model proposed in this paper can accurately predict the braking request of drivers with different driving habits.