City-scale Taxi Demand Analysis And Prediction Using Multisource Data

Traffic prediction is the key technology in intelligent transportation and the basis of intelligent traffic management and control.Taxis are an important part of urban public transportation.The development of online taxis has gradually increased the number of taxis but the imbalance between supply and demand still exists.Taxi demand forecasting is helpful for reasonable vehicle pre-allocation,assisting traffic dispersion,reducing the time waste of passengers and drivers and fuel waste caused by empty vehicles,which accelerates the process of ecological transportation.However,the significant differences in demand patterns in different regions,people's travel habits and weather,the temporal and spatial dependency of taxi demand,all leads to that cityscale taxi demand prediction is still a challenge.Urban areas have accumulated a large amount of heterogeneous data from multiple sources,which contains region characteristics,crowd travel and other information.The information is related to taxi demand.Thus,this study uses urban multisource data to analyze the spatiotemporal characteristics and correlation of taxi travel,builds a region-adaptive spatiotemporal network model to predict city-scale taxi demand.The main tasks are as follows:1)Urban multisource data processing and analysis.This research uses taxi historical order data,urban spatial data(POI data,road network data,demographic data),weather data,and time attribute data.Different methods are designed to process each type of data,which lays the foundation for data fusion and obtains region features.Using processed data,this study analyzes the correlation between taxi travel,urban spatial data and weather,explores the temporal and spatial features of taxi travel.Data correlation and spatiotemporal characteristics will be used as the basis for the prediction model and experiment design.2)Construction and verification of region-adaptive spatiotemporal network prediction model.This paper proposes a region-adaptive spatiotemporal network prediction model based on multisource data to predict the taxi demand in various regions of the city.The model learns spatiotemporal features through convolutional LSTM;Model branches are based on the pixel-adaptive convolution and calculate the region-adaptive kernel according to the POI,road network and population characteristics of regions.Spatially varying parameters and region adaption and be obtained in branches;The branches are fused based on the weight matrix method.The weather and time attribute information are further integrated.Comparison experiments between models verify the superiority of the proposed model and the effectiveness of multisource data and region-adaptive operation.3)Result and analysis of city-scale taxi demand prediction.Using real data from Chengdu,this paper comprehensively discusses the prediction results of the model.First,prediction results and prediction errors in different regions and times are displayed and analyzed.The prediction results are close to the true values,and the prediction errors are low and stable,indicating that the model stability is high.Then we change the spatial resolution and time series input to analyze the influence of different experimental settings on the prediction results.Finally,the role of the region-adaptive guidance template and the adaptive kernel is analyzed.The region-adaptive model can ensure regional heterogeneity and improve the prediction accuracy.The focus of this paper is analysis and prediction of taxi demand for the entire city using multisource data.Correlation analysis and spatiotemporal feature mining of taxi demand are implemented.This paper applies pixel-adaptive convolution technology and proposes a region-adaptive spatiotemporal network to achieve accurate and stable city-scale taxi demand prediction.