Research On Environmental Vehicle Trajectory Prediction Based On Physical Model And Data Driven

More reasonable decision-making and planning of intelligent vehicles depend on accurate prediction of future behavior and trajectory of peripheral dynamic obstacles.Reasonable and accurate trajectory prediction of intelligent vehicles is a problem that has to be solved when automatic driving vehicles develop to a higher level especially for moving environment vehicles.The paper will explore the trajectory prediction of vehicles in driving environment Considering that driving by people will still be the main method for a long time in the future.Because of the individualization of driving and the complexity of road conditions,the potential operation of environmental vehicles presents short-term variability and long-term intention.It is key to establish a prediction algorithm of long-term and short-term integration.The paper mainly researches on four levels: short-term trajectory prediction based on physical models such as vehicle kinematics,long-term trajectory prediction based on vehicle driving intention recognition,medium-term and long-term trajectory prediction based on improved long-term and short-term memory network and output fusion.In the field of physical model-based environmental vehicle trajectory prediction,Considering environmental noise,different vehicle kinematics prediction model is established,Kalman filter,extended Kalman filter,unscented Kalman filter and particle filter algorithm are introduced for effective short-term trajectory prediction.The output performance of different algorithms are compared.The simulation results show that the CTRA-PF physical model-based environment vehicle trajectory prediction algorithm can predict the target vehicle trajectory in the environment with high accuracy in the range of 0.5 to 1 second.In terms of long-term trajectory prediction based on environmental vehicle driving intention recognition,the intention recognition module is constructed by Support Vector Machine,Hidden Markov Model and Graph Neural Network.the driving intention of the target environmental vehicle in the future is judged.Combining the quintic polynomial with the constraints,the optimal long-term trajectory is output to predict the long-term trajectory of vehicles in target environment for 4 seconds or more.in the area of medium and long-term trajectory forecasting based on improved long-short term memory network,standard network structure is established,Dropout layer is embedded to enhance network generalization,attention mechanism is introduced to increase the weight of time series data which has a greater impact on the prediction effect,thus improving the reliability of the forecasting result.The improved LSTM model is combined with GRU model to build LSTM-GRU environment vehicle trajectory prediction model to further improve the prediction accuracy.This outputs the vehicle forecast trajectory for the target environment over a medium to long time horizon for 1 to 3 seconds.on the basis of trajectory forecasting based on physical model and data driven,On different time domains,the predicted trajectory of vehicles in target environment is fused by weight distribution principle based on the above-mentioned algorithm models of different principles,and the prediction trajectory of vehicles in the next 8seconds is output.Finally,relevant simulation tests are carried out for different operating conditions and the output results are analyzed qualitatively and quantitatively.The results show that the proposed fusion model has better prediction performance and can predict the vehicle trajectory in target environment under different conditions with higher accuracy and stronger robustness.