Toward version 1.2: Improvements in SoundCast
This week we marked a milestone in the continuing development of our activity-based travel model, SoundCast. The latest official release, version 1.1, included big improvements to the truck model and the benefit-cost tool. Now, SoundCast “1.2 milestone 1” includes lots of new goodies.
Improvements in 1.2 milestone 1
Faster. Between code improvements and some new server hardware in the closet, we have significantly decreased runtime from 30 hours to 17 hours for a full run. This is mostly Moore’s Law in effect: incremental updates in computing power over the three years since our last hardware purchase are responsible for most of the improvement.
Bike trips are now assigned to the network. A completely new bike assignment algorithm is now implemented. The chosen bike routes are informed by auto volumes, amount of uphill climb, and presence of different bike facility types.
Source: PSRC SoundCast
We also have many more bike counts now, with which to validate the results. This model is just a first effort, and more tuning will be coming in the near future.
Simplified Toll/No-Toll representation. After extensive review and testing by both PSRC and our consultant team RSG, the toll/no-toll choice has been turned off in SoundCast. This results in:
- Fewer skim matrices
- Fewer modes
- Slight speed improvement
—without any significant loss in capabilities; as the choice of whether to choose a route containing a toll is still embedded in the EMME pathbuilder.
Big update to “external” trips. Trips that either originate or end outside of our four-county region are known as “external” trips, and those hadn’t been updated in a long, long time. This revision now bases the distribution of the external trips on actual observed behavior from LEHD flow data for work trips, which are about 70% of all external trips. We are still working on improvements to the other (nonwork) 30%.
Source: PSRC SoundCast
Model re-estimation based on 2014/2015 household survey. Many subcomponents of the travel model have be re-estimated using the latest survey data, including auto ownership, work tour and work-based subtour mode choice, school mode choice, other home-based, and more. Destination choice and day-pattern model re-estimations are still to come.
Can I use this?
This is just an internal milestone and as such, is not fully tested. You’re welcome to explore the code at GitHub or download tag v1.2m1 locally—everything SoundCast is fully open-source. The official release of SoundCast 1.2 will occur later this winter, after the model re-estimation, calibration, and validation is complete.