It is common wisdom that while man should learn from history. At the same time, another prominent wisdom is that history never repeats itself. Both assertions are also true for public transport operations. When predicting future states of the system, it is advisable to account for historical records whilst bearing in mind that it will not evolve in the exact same fashion. Instead, the underlying dynamics of the current system as well as inherent stochastic will result with different performance. This makes the task of predicting vehicle trajectories and their arrival times at downstream stops a non-trivial task. Such predictions are useful both for operators for better managing the fleet in real-time as well as for passengers that wish to reduce their travel uncertainty and enable them to take more informed decisions.
Today and yesterday I presented on IEEE Intelligent Transport System Conference (ITSC) in Qingdao, China, two studies on prediction schemes for public transport and the generation of real-time travel information.One study together with Masoud Fadaei Oshyani, who is doing his PhD under my supervision, proposes a hybrid model to integrate several sources of information on downstream travel conditions for buses. The contribution of each information source (within-day instantaneous data, day-to-day historical data and static schedule data) and related parameters are estimated using a genetic algorithm optimization. Predictions improve considerably compared with the currently-used scheme and there are good indications on the transferability of the scheme. For the presentation slides click here: Real-time Bus Departure Time Predictions
The second study investigated real-time predictions for light rail trains. Light rail trains are often characterized by mixed-operations: running partially in mixed traffic (in some cases even in pedestrian street) and in segregated right-of-way, alternately as well as with different levels of priority in crossings. A prediction scheme that is based on constructing link-specific speed profile and then computing for a given current position and speed the remaining travel time was applied for a case study in Bergen, Norway (thanks to Edouard Naye). The scheme elevates the prediction accuracy to the same level that was reported by previous studies for metro systems. For the presentation slides click here: Real-time Predictions for Light Rail Train Systems