Research And Implementation Of Real Traffic Flow Forecast Model Improved Based On LSTM Model

In order to reduce the impact of traffic congestion and other issues on urban construction and development,Intelligent Transportation System(ITS)has been widely used in the process of urban traffic management,traffic flow prediction technology is used as the core component of the urban intelligent transportation system.In part,it is of great significance to urban planning and road traffic management.Long short-term memory network(LSTM)is an important prediction model in deep learning.It has structural advantages in time series prediction problems.Therefore,this model has been widely used in the field of traffic flow prediction.Traditional LSTM-based traffic flow prediction models usually use global positioning system(GPS)data as the data source,only use traffic flow data as the model input during the prediction process,ignoring the impact of travel mode and time information on the traffic flow.In addition,GPS data coverage and collection methods are limited,the acquisition cost is high and it is difficult to adapt to the needs of traffic flow prediction on large-scale road networks.For the above-mentioned problems,this article mainly focuses on the following aspects:Firstly,this thesis proposes a method for identifying user travel mode based on signaling data.Compared with GPS data,the positioning accuracy and frequency of signaling data is lower and it is challenging to accurately extract the user’s travel mode from the signaling data.This thesis preprocesses the signaling data through spatio-temporal continuity and uses the preprocessed continuous signaling trajectory to calculate and extract the user’s travel characteristics.Furthermore,the multi-dimensional spatio-temporal trajectory characteristics of the signaling user are constructed and input these characteristics into the XGBoost model to realize the identification of the user’s travel mode.The experimental results show that the method used in this thesis can achieve more than 90% recognition accuracy on both the public data set and the mixed data set(public data set +signaling data set).Secondly,this thesis proposes a traffic flow forecasting model based on vehicle detectors.The model obtains time features by converting the time information in the vehicle inspector data set.At the same time,in order to better obtain long-term traffic flow information,the model adds an attention mechanism layer before the input layer of the LSTM model to obtain the predictive value.Influencing traffic flow information,while using traffic flow data,influential traffic flow information and time characteristics to predict traffic flow.The experimental results show that,compared with the gated recurrent unit(GRU)model with the best prediction performance,the root mean square error(RMSE)and mean absolute percentage error(MAPE)of this model are increased by 4.69 and 0.25% respectively.In addition,the model can be used to deal with missing data.Finally,because traditional prediction models usually use GPS data as the data source and do not consider the impact of travel mode and time information on traffic flow.This thesis uses signaling data and LSTM model to construct a traffic flow prediction model,which combines travel mode characteristics,time characteristics and attention mechanism to predict regional traffic flow and conducts experiments with real signaling data.The experimental results show that compared with the traditional traffic flow prediction model,the model in this thesis has the best prediction performance.Compared with the LSTM model,the average absolute error(MAE),RMSE and MAPE of this model are increased by 1.1,3.7 and 0.6% respectively.