After many months of development, the team is now in the process of calibrating the model. This involves comparing the speed and flow of vehicles in the model to the actual data collected in the corridor. For the I-210 Pilot, the model is being calibrated for multiple time periods including AM and PM peak periods, early morning, late morning, mid-day, late evening, and night, as well as weekday and weekend for each of these periods. This is necessary to reflect the various conditions and average amounts of congestion during those time periods.
The calibrated model is now being used to support the development of response plans and once the Pilot is live, the model will be used to evaluate and recommend response plans during incidents. If the model does not reflect how traffic moves through the corridor, the recommended response plans may not have the same impact as the model suggests. This could mean an effective response plan may not be recommended or an ineffective response plan is recommended by the model. Both of these situations degrade the success of the Pilot, which is why model calibration is so critical.
Model calibration is also one of the many reasons data quality and quantity are so important. The more data that the team has to compare the model to, the better and more accurate the model becomes. Stakeholders are also critical as they understand how the corridor flows and can make recommendations to improve the model. If the model shows a bottleneck, but the corridor data does not, stakeholders may be able to identify why there is a discrepancy.
Model calibration is an ongoing task, even once the Pilot is deployed. Changes in the corridor such as traffic signal changes, speed limit changes, or new lanes can change the speed and flow and must therefore be reflected in the model. Once response plans are being deployed, the team will have real data from those incidents that can then be used to improve the model for the next incident. Continuous improvement will be important at the beginning of the Pilot, with the long term goal being for the model to only need periodic calibration to maintain its accuracy.