Estimating intersection turning movement flows with a NETFLO algorithm: weight constraint calibration

Traffic management strategies grounded on reliable traffic demand data can reduce the costs associated with traffic congestions. While traffic video imaging can deliver real-time turning movement (TM) information, their cost can be prohibitive, especially to smaller agencies. The Network Flow (NETFLO) model uses the Minimum Cost Flow optimization by generating network flows along weighted arcs from a sparse set of detected link flows. The network is constructed so that weighted arcs represent both turning movements and links. When constrained, the NETFLO algorithm can infer reliable TM estimates from link flows detected in quasi-real-time. The research presented here establishes a consistently robust constraint regime that enables reliable estimation of urban intersection TMs. Turning movements are identified as through, left, and right. The optimal NETFLO constraint regime assigns a weight of 35 for lefts, 2 for throughs, 28 for right intersection turns. A weight of one applies to all other arcs. The optimal weighting scheme consistently outperforms a non-weighted regime, when tested estimating 5-minute TM flows on a real network in Orem, Utah, U.S.A. The optimal weights yield a mean coefficient of determination (R2) of 93% between observed and modeled TM flows, while the non-weighted regime produces an R2 of 78%. With the optimal weight set, almost all through, 50% of right and 40% of left movements at intersections are reliably predicted. The lower reliability of left and right estimates can be attributed to their low observed flows, when compared to the dominant through flows. This paper shows the details of how NETFLO is constrained to give consistent urban intersection TM estimations.