A critical review of the most widely accepted demand prediction algorithms is presented. Based on data collected at four intersections, sensitivity analysis of the best existing algorithms indicates that very little improvement in their performance could be achieved. A new, simpler algorithm, which requires considerably less information and fewer computations, is subsequently proposed and compared with the best of the existing algorithms. The results suggest that for 5-min prediction the second-generation Urban Traffic Control System predictor (UTCS-2) is usually better. However, in cycle-by-cycle prediction the proposed algorithm is considerably (as much as 41 percent) better than the best of the existing algorithms. (Authors).
Abstract