Calibration: stops models calibration (scheduled stops, vehicle choice, stop duration)

[ColSmith76]

Add table of specific calibration checks and status

[ColSmith76]

Stop duration model:

• choice of share by time (11 time ranges). Target data is the SEMCOG estimation dataset which does include breakdowns by activity (goods, service, non-scheduled stop for breaks etc) and vehicle type.
• to finalize calibration approach: add adjustments to vehicle type and activity specific parameters to match differences by vehicle and activity
• then add a final adjustment to match the observed overall duration distribution from the CMAP CPS data (check first how different it is from the SEMCOG one, if at all)

Iterative adjustments:

1. initially adjust constants to match the overall SEMCOG shares (iter 1-2?)
2. adjust the vehicle and activity parameters to match the vehicle and activity shares (iter 3-5)
3. final adjustments of the constants to match the CMAP GPS shares (iter 6-8)

Issue with stop duration adjustment: final application of model is in the intermediate choice models, calibrate for goods and service stops and then readjust for other stops only. Check if this distorts the overall distribution too much, and if so might need a final overall adjustment in the intermediate step, but application code is currently not set up to support this.

[ColSmith76]

Vehicle choice model:

• calibration function adjusts the asc's (light and medium vs heavy fixed to zero), the activity specific parameters (on light and medium seperate coeffs for service stops vs goods delivery) and by employment category (on light and medium seperately flags for emp category, with retail fixed to zero)
• targets are joint distribution of vehicle choice by activity and employment category from SEMCOG data
• function adjusts each marginal distribution (overall, by activity, by employment category) in turn through 10 iterations.
• Converges okay, with some conflicting targets meaning not all shares are hit (not enough degrees of freedom in the model to exactly hit all combinations)

[ColSmith76]

Vehicle choice model is not yet calibrated at this point using the stop distance distribution from the CMAP GPS data

[ColSmith76]

Left to do: Stops models: distance from business to stop distribution compared to CMAP GPS data Vehicle choice model: calibrate using the stop distance distribution from the CMAP GPS data Stop duration model: final adjustment to match the observed overall duration distribution from the CMAP CPS data (check first how different it is from the SEMCOG one, if at all)

Also remember to cross check the final distribution of stop durations after the addition of the intermediate stops with the CMAP GPS data

[ColSmith76]

Last couple of checks needed:

• add additional summaries to dash from GPS vs model
• some inconsistency in the vehicle choice by purpose/industry -- two or three combinations that are not close to the targets, look at error in terms of absolute difference in number of stops to see if magnitude is a problem

[ColSmith76]

After latest calibration run (number 8):

• Stops models, distance from the base to stop looks good overall, not capturing industry specific variation very well (compared to SEMCOG data). How to improve: adjust the industry specific distance impendance factors (d_bars)
• Vehicle choice: service/heavy a bit high, goods heavy a bit low -- can those be teased apart a bit? Add extra activity specific variable?
• stop durations look good in comp with activity and vehicle data, except need to add overall comparison with GPS (do on set of stops including the intermediate stops)

[ColSmith76]

With correction to stops targets, stop numbers look much better and resultant VMT better. Overall good, some minor remaining discrepancies between model and target data at segment level e.g. by industry or vehicle type. Might need to boost stop numbers and spread just a little more to hit the VMT targets.

[ColSmith76]

VMT looks okay relative to estimate. Close this issue and deal with any adjustments based on final review as a new issue