Driving Behavior Prediction Based on Combined Neural Network Model
Runmei Li, Xiaoting Shu, Chen Li
Abstract
Accurate behavior prediction of surrounding vehicles can greatly improve the operating safety of autonomous vehicles. However, in real traffic scence, the complexity and uncertainties of traffic flow bring great challenges to driving behavior prediction. This article proposes a driving behavior prediction model using a wide-deep framework that combines gradient boosting decision tree (GBDT), convolutional neural network (CNN), and long short-term memory network (LSTM) algorithm to fully mine driving behavior characteristics while improve interpretability of the CNN-LSTM model. The GBDT algorithm can quantitatively describe the interaction between the autonomous vehicle and its surrounding vehicles during the driving process, obtaining a series of driving behavior rules, and integrating the driving behavior rule features into the CNN-LSTM neural network. The CNN-LSTM neural network model is constructed to find the spatial features in driving trajectory by CNNs and the temporal features by LSTM networks. The accuracy of the driving behavior prediction model is further improved. Simulation experiments show the rationality and validity of themodel and algorithm.