I run the following test on the random rider insertion using the multiple config with the meta controller selected.
8floors, 2cars of 4capacity, 80people building, random seed 1 and I get an AVG wait time of 19.979.
If I reduce the number of people in the building to 40 and keeping all variables the same i receive a number 133.961 which is larger for the AVG wait time.
Now my question is how can a building with fewer people has to wait more for the elevator to arrive that if there are more people? Does this mean this specific building is better of with one Elevator?
Thanks
Last edit: KLEIDI 2014-12-13
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
The results are extremely dependent on the behaviour of the riders, which is pseudo-randomized. The potential range of results can be broad. Using your settings for 80 riders, but running a multiple simulation (instead of real-time), I used a range of 49000-49060, increment of 1 on the random seed. The results ranged from 595 to 304,383. This is substantial variation in the wait time with the only change being to the random seed that varies the behaviour of the riders.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I agree with you on the variable of seed which mimics the randomness factor of the riders and depending on this we have variable times. However, if we keep the randomness factor not random, thus keeping the seed 1, simply by changing the number of riders does this affect the randomness algorithm? I would suppose if seed stays the same, the randomness factor is the same. What I wanted to test was the avg wait time in two buildings with all parameters the same just the number of inhabitants in one was 80 and in the other 40 (assuming they use the elevator with the same rider pattern).
For your curiosity, I wanted to test if a building is better off with 1 Elevator, 2 Elevators that work even and odd floors or 2 elevators that work with just one controller and the closest to the floor responds. Hence the test to mimic the 2 elevators with even/odd was logically reducing the number of people to 1/2 the number of the building and reducing the elevator to 1 (thus serving only half the building).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I should add that we created to randomized rider option in order to test what we thought might be mid-day style behaviours where people move around to go to meetings or to perform tasks. Truthfully, it's probably not a good indicator of actual work day behaviour in a modern office building, although it allowed us to more thoroughly test the controller and elements of the simulator.
We added the morning and evening scenarios after this because these seemed like good peak-load tests. The behaviour there, although somewhat randomized, follows a Gaussian curve where there's a build up to a peak period and then a drop off. This model is a lot easier to justify, so I'd expect it to be a better indicator of efficiency.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Congratulations on your work! I read a bit the literature online about the elevator algorithms and it was extensive and with many variables, hard to understand if you simply want to test the effectiveness of just your own building elevator system. Not an easy task to put on a sim speaking from a programming point of view. Well done!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I run the following test on the random rider insertion using the multiple config with the meta controller selected.
8floors, 2cars of 4capacity, 80people building, random seed 1 and I get an AVG wait time of 19.979.
If I reduce the number of people in the building to 40 and keeping all variables the same i receive a number 133.961 which is larger for the AVG wait time.
Now my question is how can a building with fewer people has to wait more for the elevator to arrive that if there are more people? Does this mean this specific building is better of with one Elevator?
Thanks
Last edit: KLEIDI 2014-12-13
The results are extremely dependent on the behaviour of the riders, which is pseudo-randomized. The potential range of results can be broad. Using your settings for 80 riders, but running a multiple simulation (instead of real-time), I used a range of 49000-49060, increment of 1 on the random seed. The results ranged from 595 to 304,383. This is substantial variation in the wait time with the only change being to the random seed that varies the behaviour of the riders.
I agree with you on the variable of seed which mimics the randomness factor of the riders and depending on this we have variable times. However, if we keep the randomness factor not random, thus keeping the seed 1, simply by changing the number of riders does this affect the randomness algorithm? I would suppose if seed stays the same, the randomness factor is the same. What I wanted to test was the avg wait time in two buildings with all parameters the same just the number of inhabitants in one was 80 and in the other 40 (assuming they use the elevator with the same rider pattern).
For your curiosity, I wanted to test if a building is better off with 1 Elevator, 2 Elevators that work even and odd floors or 2 elevators that work with just one controller and the closest to the floor responds. Hence the test to mimic the 2 elevators with even/odd was logically reducing the number of people to 1/2 the number of the building and reducing the elevator to 1 (thus serving only half the building).
I should add that we created to randomized rider option in order to test what we thought might be mid-day style behaviours where people move around to go to meetings or to perform tasks. Truthfully, it's probably not a good indicator of actual work day behaviour in a modern office building, although it allowed us to more thoroughly test the controller and elements of the simulator.
We added the morning and evening scenarios after this because these seemed like good peak-load tests. The behaviour there, although somewhat randomized, follows a Gaussian curve where there's a build up to a peak period and then a drop off. This model is a lot easier to justify, so I'd expect it to be a better indicator of efficiency.
Congratulations on your work! I read a bit the literature online about the elevator algorithms and it was extensive and with many variables, hard to understand if you simply want to test the effectiveness of just your own building elevator system. Not an easy task to put on a sim speaking from a programming point of view. Well done!