Urban Taxi Demand Prediction Based On LSTM And Optimized Dispatching Model

Taxi is an important means of transportation in urban public transport,which provides a strong support for alleviating urban traffic pressure.In order to solve the high empty-loading rate problem and difficulty in finding passengers in urban taxi operation at present,demand-based taxi dispatching systems are proposed in many works.However,the existing dispatching algorithms only include the number of realtime demand and the number of empty taxis.There are some dispatching delays in those dispatching algorithms,and the passenger loaded taxis are not taken into account.In-depth research on this problem is proposed in this thesis from two aspects: regional taxi demand prediction and destination prediction.A lot of research has been carried out on the trajectory data of about 15,000 taxis in Shenzhen.The data set spans 37 days from October 20,2014 to November 25,2014.The average number of trajectories per day is about 400,000 in the data set.The main contents of this thesis are as follows:1)Aiming at the influence of urban traffic weekday periodic pattern on the change of regional taxi demand,an urban regional demand prediction algorithm based on time periodic pattern and spatial correlation feature is proposed.Based on the taxi data of Shenzhen,the experimental sample data of demand prediction model is constructed.Two-dimensional Convolutional Neural Network(2D-CNN)is used to mine the spatial characteristics of demand distribution in the region and its surrounding regions.Long Short-Term Memory(LSTM)network is used to extract the characteristics of time-series dependence of samples,and a demand prediction model is proposed.In this thesis,the time periodicity characteristics of regional demand data are analyzed,and the time information is added into the prediction model as an auxiliary information in the form of embedded vectors.The experimental results show that the root mean square error(RMSE)and the mean absolute percentage error(MAPE)of the regional demand prediction model proposed in this thesis are about 14.37 and 0.1825 respectively.The experimental comparison shows that the urban traffic weekday periodic pattern can improve the prediction effect of taxi demand.2)Aiming at the lack of spatial feature extraction of trajectory in LSTMbased destination prediction algorithm,one-dimensional CNN is added to extract local features and semantic information of trajectory,so as to enhance the expressive ability of destination prediction model to spatial features of trajectory.By clustering destination based on DBSCAN(Density-Based Spatial Clustering of Applications with Noise),the taxi trajectory is labelled to obtain the experimental sample data.The taxi trajectory is mapped to the grid area.Point of interest(POI)information is mined in the grid area,and POI semantic information is added to the taxi trajectory information.One-dimensional CNN is used to extract spatial local features from taxi trajectory information and semantic information,and LSTM-based prediction model is proposed.In addition,the time and period information of the trajectory are embedded in prediction model to enhance the expressive ability of the prediction model.The experimental results show that the prediction error of destination is about 3.8km.Compared with the LSTM model,the prediction accuracy is improved.3)Aiming at the problem of ignoring the arrival of passenger loaded taxi in the existing dispatching algorithms,combining the regional taxi demand quantity prediction and destination distribution prediction,the dispatching demand/supply model is improved,and NSGA-II is used to solve the multiobjective optimization problem.In this thesis,two objectives of maximizing regional demand satisfaction and minimizing driving distance of dispatched vehicles are optimized.The time delay of taxi dispatching is reduced based on the prediction of urban regional demand.Taxi destination prediction can improve the calculation of regional demand,thereby reducing unnecessary taxi dispatch and reducing the distance of dispatched vehicles.Using NSGA-II algorithm,multi-objective optimization of regional demand satisfaction and dispatching distance in the dispatching model is solved.The experimental results show that the combination of urban regional demand prediction and taxi destination prediction can schedule taxis ahead of time,maximize regional taxi demand satisfaction and minimize dispatching distance.In this thesis,regional demand prediction algorithm and destination prediction algorithm based on LSTM are proposed to improve the accuracy of the two prediction models.Two prediction models are used to improve the problems of dispatching delay and neglect of passenger loaded taxi in taxi dispatching system.A multi-objective optimization algorithm is used to construct the taxi dispatching model.The experimental results show that the proposed demand prediction model and destination prediction model are improved compared with the existing methods.The dispatching optimization model has a great improvement in the experimental results of multiobjective tasks with demand satisfaction and dispatching distance.