faq 303693825

initial agent's plans

by Rizki Nagari on 2018-07-19 02:27:24

---

Dear, experts.

Let's say i have information at sub-area level on at what time individual leaving home, the duration of individual travel from home to the activity location, and at what time individual arriving back at home. also the number of individual change mode and main mode.

my questions are :

• how to assign agent's activity location based on that travel time duration and home location information ?

• and how to make initial route plan ? Is it that important ? because in my simulation i have not set the initial route and it worked too.

---

Comments: 2

---

Re: initial agent's plans

by Thibaut Dubernet on 2018-07-19 07:19:27

Dear Rizki,

how to assign agent's activity location based on that travel time duration and home location information ?

If the only information you have is where agents live and how long they travel, I am afraid this will be difficult. A first approach would be to search for candidate facilities (to which travel time would be approximately the one you have) and sample one at random. What you really want though would be additional data to validate/constrain this process (for instance in the form of OD matrices and/or traffic counts).

In any case, with such data, one will often have to rely on heuristics to disaggregate the demand into realistic plans.

and how to make initial route plan ? Is it that important ? because in my simulation i have not set the initial route and it worked too.

Initial route does not matter if you use "Reroute" at replanning (and if you only have one innovative replanning strategy, it should be this one).

---

Re: initial agent's plans

by Kai Nagel on 2018-07-19 07:25:36

(1) You don't need to compute initial routes; MATSim does that.

(2) I would just select random points within the home zone and work zone. Cf.{{RunDemandGenerationFromShapefileExample}} (but it reads like you already figured this out).

(3) There is in principle a method to come up with better location choices if your zones are rather big and you have marginal distributions (e.g. of travel times). We are still working on this. There is a paper by Ziemke, Bhat, Nagel explaining the methodology.

(4) In your case, it sounds like you have individual travel times, which is again a different situation. You might be able to program it yourself: Give each agent multiple plans, with different home and/or work locations within the given zone, then run iterations with those plans and in the long run only keep those where the simulated travel times are closest to the one from your data. This is similar to (3), but in fact a bit simpler since you already have the individual objective values.