Forecasting Road Traffic Using Kalman Filter Models

Providing users with accurate arrival times is key for improving the appeal of public transport. Research presented here incorporates social media data into a new model in order to improve accuracy of bus arrival time prediction. The model is intended to function at a pre-processing stage to handle real world input data in advance of further processing by a Kalman Filtering (KF) model. Arrival time is predicted using a KF model supplemented with information acquired from social networks. Social networks feed road traffic information into the model, based on information provided by people who have witnessed events and then updated their social media accordingly. Different KF models are compared and the best models identified using the road traffic simulator, Simulation in Urban Mobility (SUMO). SUMO simulates real world road traffic using digital maps and realistic traffic models. The combination of SUMO and social media information as inputs into KF models produces more accurate travel time predictions than is possible when using only one source of information. The combination of input data and modelling is done using MATLAB. The KF model predicts arrival time by filtering out disturbance during the journey. This research discusses modelling a road journey using KF and verifying results with a corresponding SUMO simulation. Integrating the SUMO measures with the KF model can be seen as an initial step to verifying our premise that real-time data from social networks can eventually be used to improve the accuracy of the KF prediction. In this research, X is used as an example social network technology. X offers an API to retrieve live real-time road traffic information and offers semantic analysis of X data. In order to acquire optimal estimation, verifying the trustworthiness of social network information is also crucial. Ideas on methods to establish a level of trust in social networks are discussed. This is important, as KF model prediction will suffer if incorrect information from social networks is used.