SATURN OBA (MUC)
The Origin-Based Assignment (OBA) algorithm (Bar-Gera, 2002) was developed by Prof. Hillel Bar-Gera, while a PhD student at the University of Chicago in the late 1990’s and has been enhanced by Dr Yanling Xiang of Atkins to cater for multiple user classes, hence OBA (MUC).
The key difference of OBA from the link-based FW algorithm is that it stores the link flows as generated by each individual origin. In a certain sense, OBA is an intermediate between link-based and path-based algorithms but without requiring excessive RAM or CPU as typically characterized by the latter. The principle of OBA is that, by considering origin-based link flows from the a-cyclic sub network, it provides a computationally efficient route-building process as well as enabling the elimination of residual flows (i.e. small flows on sub-optimal routes) that have a detrimental impact on algorithm convergence.
OBA (Multiple User Class version) was released to all maintained users in the fourth quarter of 2009 with the intention of incorporating the algorithm as a standard part of SATURN from FY10/11 onwards.
Recent testing work on OBA (MUC) algorithm, as reported at the 2009 European Transport Conference, has demonstrated that, for the majority of networks, OBA (MUC) is able to achieve a higher level of convergence than the existing FW algorithm for the same CPU expenditure.
These higher levels of convergence now achievable in SATURN will provide practical benefits to other models that are sensitive to convergence including demand models (e.g. DIADEM) and cost-benefits models (e.g. TUBA) for example. In addition, as the algorithm stores the route proportions (in the .UFO file), any secondary analysis may be undertaken without needing to re-build the paths and thereby considerably reducing the CPU required.
The current OBA MUC development work has also reinforced the importance of good quality network coding (as enforced by WRIGHT=T). Convergence within SATURN is sensitive to the interactions between the assignment and simulation loops. The new OBA algorithm is able to achieve very high levels of convergence but only if the network coding enables robust and stable estimates of flow and travel costs to be calculated.
OBA MUC Presentation - 2009 SATURN UGM