Research Of Taxi Destination Prediction Method Based On Deep Learning

In recent years,with the extensive development of GPS positioning technology,location-based communication technology has become an important component of people’s daily life.Accurate destination prediction of taxi trajectory can benefit many location-based intelligent services,such as accurate advertising for passengers.The traditional methods used to predict the trajectory are mainly clustering algorithm,Markov and other destination prediction algorithms based on taxi historical trajectory.With the development of artificial intelligence,the prediction methods based on deep learning adopt Convolutional Neural Network(CNN)or Long Short Term Memory(LSTM)to process the trajectory,where CNN has better advantages in extracting the spatial features the trajectory and LSTM can well extract the temporal features of the trajectory.However,it is difficult to capture the complex spatio-temporal features of the trajectory by using CNN or LSTM alone.Firstly,by analyzing and studying the related technologies of deep neural network and its related work in taxi destination prediction,in this thesis,a general method of taxi destination prediction combining CNN and LSTM is proposed,where CNN and LSTM are used to capture the spatial and temporal characteristics of taxi trajectory,respectively.Besides,attention mechanism is introduced to better adapt to different application environments.It provides theoretical basis and feasible methods for the follow-up research of specific prediction models.On this basis,a new taxi destination prediction method based on Multi-stage Spatio-Temporal Network Model(MSTNM)is proposed.This method applies multiple independent CNNs and LSTMs to extract the spatial and temporal features of the trajectory respectively,and the number of CNNs and LSTMs can be flexibly adjusted to extract the trajectory features more accurately.In the process of mapping the trajectory into images,the historical traffic information of taxis is added to further consider the impact of traffic flow on the driving route of taxis.At the same time,the attention mechanism is applied to LSTM to enhance the influence of key trajectory segments on the final destination prediction,and improve the accuracy of the destination prediction by multiple methods.The experimental results show that the prediction error of MSTNM is significantly reduced compared with the existing algorithms,and then the effectiveness of MSTNM is validated.Finally,combining practical and theoretical algorithms,a prototype system of taxi destination prediction based on deep learning is designed and implemented,which adopts the proposed MSTNM and consists of four modules,including trajectory data management,model management,advertisment management and destination prediction.The system can realize the dynamic graphic display of taxi track and predicted destination.The system tests show that it has convenient operation and perfect functions.The research results of this paper provide a new idea for the research of taxi destination prediction based on deep neural network,which has high theoretical research value and wide application prospect.