Urban Rail Transit Short-term Passenger Flow Prediction Usingdeep Learning Techniques

The development and advances in ICT(Information and Communication Technology)has revolutionized every sector and has given a completely different outlook;optimizing services based on in-depth understanding of both consumer and provider’s perspectives and bridging the gap between supply and demand.The transportation sector has also benefited significantly with its development as it has led to the emergence of ITS(Intelligent Transportation System).The enormous amount of data generated from the widely deployed sensors have given rise to the age of big data.Simultaneously,the computing advancements have enabled mining these ubiquitous data and carry out data-driven studies;enabling transportation planners,operators,and travelers to make informed and smart decisions through ITS applications like TDM(Transportation Demand Management),ATIS(Advanced Traffic Information System),APTS(Advanced Public Transportation System),etc.Urban rail transit has gained significant popularity among travelers in metropolitan cities due to numerous advantages like mass transit,fast,reliable,environmentfriendly operation.However,in recent times,the rapid increase in travel demands has led to over crowdedness and safety concerns both at platforms and trains.Thus,ITS has an even greater role to play in the coming future to improve and manage urban transit services.However,ITS requires accurate and robust forecasting to anticipate the dynamics of the transportation network prior and formulate strategies accordingly like feeder bus services during high demand,crowd controlling by facilitating orderly deploying of personnel and information,etc.This thesis studies the state of art-statistical and ML methods and their application in transportation domain.Four models namely;SARIMA,SVR,RF,LSTM is used in this study to predict urban rail transit passenger flow.With the aim to achieve a universal model;all these models are developed and tested on aggregated datasets i.e.complete dataset with both weekends and weekdays data.Further,different stations and scenarios(inbound/outbound)are considered to account for the heterogeneity in flow characteristics and other attributes like station located in urban/suburban areas,facilities available(regular or transfer station),etc.among different stations.Different combinations of features are used to prepare multiple datasets for each station,enabling the study to evaluate the models on different conditions while assessing the influence of the inclusion of different features on the performance of the model.Finally,the model performance is evaluated based on the predictions that are carried out for the three days,comprising a weekend and two weekdays.