The current project addressed two major weak points of the existing WSDOT ramp control system. One weak point in the system is the fact that it reacts to the problem (congestion), rather than preventing the problem. The other weak point in the system is its reliance on detector data that may be in error. Both of these problems can be minimized by developing methods to accurately predict short-term traffic data. By predicting the onset of congestion early enough, the ramp metering system can act to prevent or delay occurrence of the problem. Also, if a detector has failed or is malfunctioning, the data from the detector can be estimated from short-term predictions based on neighboring detectors.
At the beginning of the current project, the researchers had hoped that the same model would provide a basis for both forecasting congestion (for predictive ramp control) and replacing erroneous data (predicting actual values). However, the best method for filling in missing detector data turned out to be multivariate time series analysis.
Several pattern recognition and time series models were tested for further development. In both cases, the simpler models turned out to be the best choices, and in both cases, further model testing and development were recommended.
The research on both model types continues in follow-up studies that are expected to lead to incorporation of these models in the new TSMC computer system.