Research On Automatic Driving Lane Prediction And Decision Method Based On LSTM Algorithm

In recent years,with the rapid development of China’s urbanization process,traffic safety and road congestion have become increasingly serious.Autonomous vehicle has become an important research direction in the field of traffic intelligence because of its advantages of effectively reducing traffic accidents caused by human errors and reducing energy consumption caused by road congestion.In the actual traffic scene application of autonomous vehicles,they will face a highly uncertain and dynamic interactive complex driving environment.This requires autonomous vehicles to detect and collect the movement information of surrounding vehicles to ensure the safe driving of vehicles in real time,and to predict the driving trajectory of surrounding vehicles in real time.However,it is difficult for a vehicle to obtain all the information about its surrounding vehicles,and the surrounding vehicles do not have a specific motion law.Therefore,it is necessary to predict the vehicle trajectory and make safe and effective lane changing decisions,which has important research significance.In view of the problems existing in the current research on trajectory prediction and lane changing decision,this thesis selects long short term memory neural network(LSTM)and fuzzy logic as the main application algorithms,and constructs an LSTM trajectory prediction model based on convolution neural network(CNN)and a lane changing decision model based on fuzzy logic algorithm.The research contents are as follows:(1)Track data preprocessing.This thesis starts from the trajectory data and studies the characteristics of the data.Considering the problem that the training time is too long,in order to effectively improve the speed,efficiency and accuracy of model training,this paper will use clustering algorithm to fully mine the relevant features of the trajectory data,and input the extracted features and trajectory into the model for training.The supervised learning method is used to construct the data,the sliding window method is used to construct the supervised learning data of time series,and the symmetric exponential moving average filtering algorithm is used to reduce the noise of the data.The results of data preprocessing show that the quality of vehicle trajectory data is further improved while maintaining the original undulation characteristics of the data.(2)The C-LSTM trajectory prediction model is established.In this paper,a CNN-LSTM(C-LSTM)vehicle trajectory prediction model is constructed.The model is based on CNN and LSTM,and includes four layers: input layer,convolution layer,LSTM layer and full connection layer.Different layers have different functions.The convolution layer is mainly to accurately extract the characteristics of trajectory data and improve the efficiency of model training.The LSTM network layer can effectively complete the extraction target of timeseries characteristics of motion trajectory,The establishment of the full connection layer is to predict the motion trajectory of the vehicle.In this thesis,NGSIM Expressway data set is selected as the verification data set.The results show that the average mean square error of the C-LSTM model constructed in this paper is lower than that of the single-layer LSTM model and the traditional CTRA model,so the application effect is better.(3)A lane changing decision model based on fuzzy logic is established.In this thesis,the complex nonlinear relationship between the vehicle running environment and lane changing decision is studied through fuzzy logic algorithm,and different lane changing scenarios and processes are analyzed.At the same time,the motion state of all vehicles is explored,and the lane changing safety distance model is constructed.After a comprehensive analysis of various factors affecting free lane changing,the corresponding speed difference coefficient and vehicle distance expectation coefficient are input into the lane changing decision-making model,and the free lane changing fuzzy rules are established with the intention of lane changing as the output.The conclusion is obtained by reasoning according to the fuzzy rules.The results show that the proposed lane change decision-making model has higher accuracy than the naive Bayesian network lane change model,can effectively reduce the potential safety risks of planning results,and is effective and reliable.Finally,according to the trajectory prediction and lane change decision-making model constructed in this thesis,simulation verification is carried out on MATLAB and Simulink platforms.MSE,RMSE,MAPE and loss function are used to evaluate the prediction model,and different models are compared and verified in terms of accuracy and time.The simulation results show that the method described in this paper can meet the requirements of real-time prediction and accuracy,realize dynamic and safe lane changing of vehicles,meet the safe and reasonable interaction between autonomous vehicle and other traffic vehicles,and promote the improvement of the intelligence of autonomous vehicle.