Predicting Dynamic Origin-Destination Matrix by Time Series Pattern Recognition

Dynamic Origin Destination (OD) matrix estimation is a classic problem that has long been a subject of scholarly investigation. OD estimation is an essential prerequisite for transportation planning and traffic management. Despite the plethora of research on this subject, most models available in the literature fail to present the elegant characteristics of OD time series data. The patterns in OD time series break down into regular and particular patterns. However, most studies in literature focused on the regular type.  Broadly the regular patterns are used to represent the general distribution patterns in time-dependent OD demands. Although, uncontrollable variables such as weather conditions, events, time, and crashes affect the OD patterns considerably. So, considering the impact of these uncontrolled variables, we developed a time series prediction algorithm model that can show both regular and particular patterns. The proposed model classifies historical data and estimates the class of coming demand. The clustering and association rule techniques are used in the proposed model to predict the coming OD. The bike riding data in Chicago was used to test the algorithm and the results suggest that the model can predict the class of OD with above 80% accuracy with a reasonable number of classes.