The objectives of this study were to develop a predictive algorithm for freeway congestion and investigate and evaluate the current TSMC definition of freeway congestion or "bottleneck" conditions. Data were collected along a section of the I-5 mainline northbound beginning at Downtown Station 108 and ending at Montlake Terrace Station 193 using two approaches: (1) time series modeling, and (2) pattern recognition. A pattern recognition approach was used to identify the best criteria for "bottleneck" definition and also to identify the best criteria for predicting "bottleneck" conditions. The time period for collection was 2:30 to 6:30 p.m. with a data time interval of 20 seconds.
The study concludes that: (1) The current definition of "bottleneck" conditions misses true forced-flow conditions approximately half of the time. A new definition is proposed. (2) A simple method for predicting congestion that can be easily incorporated into the TSMC computer system is proposed. (3) An alternative method of selecting the appropriate metering rate is proposed and further investigation of this criterion is suggested. (4) An improved method of identifying "chattering" errors in loop detectors was discovered as a by-product of the current study. It is recommended that the new criterion be incorporated in the TSMC error analysis routine.