Adaptive forecasting of freeway traffic congestion.

Problems of forecasting freeway traffic variables a few minutes in advance, particularly lane occupancy and the difference between in flowing and out flowing traffic for a short section of freeway, are explored. Methods based on linear time series analysis were found to do reasonably well at forecasting mean values but not so well for those extremes corresponding to the onset of congestion. Techniques based on statistical pattern recognition principles were found to be promising. The most promising of the pattern recognition algorithms was put into use on a section of I-5 and is being field tested.