Origin Destination Flows
Understanding passenger origin-destination (OD) flows is crucial to improving the planning and operation of transit systems and enhancing passenger satisfaction. The Campus Transit Lab combines manually collected OD information and automatically collected APC (Automatic Passenger Count) data to conduct empirical evaluations of various methods that estimate passenger OD flows from boarding and alighting data. Methods have been developed to conduct these evaluations, and the scope of the data has allowed empirical comparisons on previously unobtainable scales. Theempirical investigations have also led to the development of a new estimation method that has outperformed traditional methods in all tests conducted so far. OD flow estimation from boarding and alighting data has been considered in the past, but the ability to collect massive amounts of these data with APC technologies has generated renewed interest in this topic. The benchmarking of the results produced by existing methods and the good performance of the newly developed method have attracted attention from transit agencies and metropolitan planning organizations, as well as from national and international researchers
- HEM new estimation techniques
- Field validate
- Homogeneous periods
- Stop Grouping
Transit lab has developed a new methodology, a Heuristic Expectation Maximization (HEM) algorithm, to estimate route-level probability OD flow matrices utilizing large volumes of APC (Automatic Passenger Counter) data. Unlike previous transit OD estimation approaches, this algorithm considers the distribution, rather than only the means, of the APC data. Numerical studies on several bus routes have demonstrated that the HEM algorithm produces more accurate estimates than the IPF (Iterative Proportion Fitting) and CM (Conditional Maximization) methods when there are a sufficiently large number of bus trips with APC data. It has also been shown to reduce biases in the estimation, such as the overestimation of short bus trips present in the IPF method.
Transit lab has collected over 20,000 passenger ODs across 450 trips on 5 different Campus Area Bus Service (CABS) routes since 2008. This extensive survey OD collection, along with a 95% sample rate of passengers on these trips, has given us a high degree of confidence in understanding spatial and temporal travel patterns around the campus. A survey OD collection of this magnitude is rarely undertaken on most bus routes, so we are given a unique ability to compare, evaluate, and validate results from APC-estimated ODs against the ground truth which we have collected in the field.
Transit lab has developed a method which utilizes Automatic Passenger Counter (APC) data to identify time-of-day periods of homogeneous bus route OD passenger matrices. This methodology has been implemented for both normalized and volume-based OD matrices. We aggregate these trip-level OD matrices into elemental matrices for a relatively short time period. These elemental matrices are the input for a traditional clustering procedure which indicates periods of homogenous OD flows continuous across the entire day. This method has been applied on bus routes (CABS, COTA, LA Metro) where temporal travel patterns are understood. The time periods identified in this study correspond well with our a priori knowledge of travel patterns on these routes.
This color plot presents the homogenous period zonal OD based on land use zone definitions for LA Metro Route 4 eastbound. It shows the Origin- Destination flow for passengers through 7 am- 7 pm.
- The darker the cell color is , the larger OD probability is;
- It shows how travel pattern changes through time of day.
Transit lab has developed a method which utilizes bus route OD passenger matrices determined from Automatic Passenger Counter (APC) data to group stops with similar OD passenger flows together. Long bus routes present difficulties in analyzing or visualizing OD flows. It was therefore beneficial to group stops which exhibit similar OD passenger flows to reduce the number of cells in an OD matrix, thus improving our understanding of passenger travel patterns. This method was applied across several CABS, COTA, and LA Metro bus routes and the stop grouping results were compared against grouping defined by land uses. Our stop grouping results generally corresponded with grouping defined by land uses with some interesting exceptions, such as single stop groupings at transfer points, which corresponded with our a priori knowledge of the route.