考虑前车信息的CNN-BiLSTM的短时车速预测

Research on short-time speed prediction based on WSO-optimized CNN-BiLSTM

提出一种考虑跟车信息的基于卷积神经网络(CNN)和双向长短时记忆神经网络(BiLSTM)车速预测模型,引入白鲨优化算法(WSO)对模型的超参数进行优化。综合考虑跟车时的前车信息和其他影响车速的因素,通过驾驶人在环平台采集相关数据,确定了加速踏板开度、制动踏板开度、自车车速、相对车距、相对车速、自车加速度6种变量作为WSO-CNN-BiLSTM模型的输入。通过数据的样本熵值确定变分模态分解的模态个数对数据进行降噪处理。仿真结果显示,考虑前车信息的多输入预测模型相比单一输入预测精度有所提高,且所建立的模型与SVR(support vector regression)、LSTM、CNN和TCN(temporal convolutional network)相比,RMSE值分别降低了63.39%、11.45%、58.45%、42.58%,MAE值分别降低了59.09%、8.09%、57.29%、38.99%,提高了车速预测精度。

Accurate prediction of vehicle speed is of vital importance for vehicle safety and control.In this paper,a Convolutional Neural Networks(CNN)and Bidirectional Long Short-Term Memory(BiLSTM)based vehicle speed prediction model considering the following vehicle information is proposed.And the White Shark Optimisation(WSO)algorithm is introduced to optimize the hyperparameters of the model.With thorough consideration of the information of the front vehicle and other factors affecting the driving speed when following a vehicle,the relevant data are collected through the driver-in-the-loop platform,and six variables(accelerator pedal opening,brake pedal opening,self-vehicle speed,relative vehicle distance,relative vehicle speed,and self-vehicle acceleration)are determined as inputs to the WSO-CNN-BiLSTM model.The number of modes for the variational modal decomposition is determined by the sample entropy value of the data for noise reduction of the data.Our simulation results indicate the multi-input prediction model considering the information of the front vehicle improves the prediction accuracy compared to the single-input prediction.Compared to SVR(Support Vector Regression),LSTM,CNN,and TCN(Temporal Convolutional Network),it reduces the RMSE values by 63.39%,11.45%,58.45%and 42.58%and cuts the MAE values by 59.09%,8.09%,57.29%,and 38.99%respectively,markedly improving the accuracy of vehicle speed prediction.