Finetuning

All topics specific to modeling questions in Enterprise Dynamics
Post Reply
ditmar
Posts: 19
Joined: Friday 21 March, 2014 - 11:45

Finetuning

Post by ditmar »

aangepast impirical - 24 hrs frequentie.mod
(57.55 KiB) Downloaded 296 times
Hello,

Have some problems regarding my model which you find attached. So far tried everything to adjust and finetune it, but the cannot seem to get the right results.

Dataflow :
-107 average per day in my arrival list and 102 average per day in my departure list
- uniform distributed in both multiservice atoms.
- Both use the same server with cycletime 160 sec average (server has special input strategy: round robin adjustment; when one queue in empty it switches to the other).
- both queue capacities are 15 pieces.

The problem is when I analyse the results my histograms show some kind of neg.exponential line which means:
High percentage of time 0 in queue afterwhich it decends extremely to 14 and 15 gives another high percentage of total in queue. If I change the queue capacity to like 50 pieces it even gives a percentage of the total of 1%. Which seems odd when you look at the data.

The reasonable result should be when you look at the histogram any sort of normal distribution, which means low queues of zero and 15 and higher percentage of total in the middle which should be around 4-6 average in queue

How could I get this to work properly ? Tried everything already.

Best regards
Ditmar Rijk
ditmar
Posts: 19
Joined: Friday 21 March, 2014 - 11:45

Re: Finetuning

Post by ditmar »

Anyone ?

Regards

Ditmar
Nick
Posts: 48
Joined: Saturday 15 February, 2014 - 01:52

Re: Finetuning

Post by Nick »

Hi Ditmar,

The Histogram shows a high peak at 0 and 15 as the queue is empty or blocked most of the time.
I added some some graphs to illustrate this. 1000h run gives app. 27% empty, 28% full. The remaining 45% is 0>x>10.

102 + 107 trucks / 11 hours per day is on average 19 trucks per hour.
With 160 seconds trucks the server is busy for app. 50.6 minutes = 84% utilization.
If i run your model (1000h) then the utilisation rate of the weighing bridge is approximately 85%.
So, could you please explain me why you expect an average of 4-6?

"If I change the queue capacity to like 50 pieces it even gives a percentage of the total of 1%."
If you increase the size of your queue you also have to adjust the histogram segment sizes.

Regards, Nick
Attachments
aangepast impirical - 24 hrs frequentie with graphs.mod
(71.16 KiB) Downloaded 308 times
ditmar
Posts: 19
Joined: Friday 21 March, 2014 - 11:45

Re: Finetuning

Post by ditmar »

Hi nick

thank you very much for your reply, this makes some things a lot clearer for me.
I think I looked to much to the reality and made some statements where I focussed to much on.

However there is one thing, in the multiservice atom. The distribution of the 11 hrs takes place there via an uniform distribution. If you take an other distribution there what would be the effect? Or is it the best way to take uniform ? Or what should be the best distribution for this situation ?

Also if I want to change the priority rules of the weighingbridge. From the current situation to for example: 2 trucks in and 1 out. If you take pre defined priority rules it is hard to analyze. Does it have to be changed in the 4D script code ? or via the badge rule ?

Thanks in advance,

Regards
Ditmar
Nick
Posts: 48
Joined: Saturday 15 February, 2014 - 01:52

Re: Finetuning

Post by Nick »

Hi Ditmar,

Your distribution is uniform within the hour and emperical if you look at the whole day.
As a result trucks will always arrive/leave randomly but at the same time according to the daily pattern.
What kind of distribution you can or should use really depends on the data / situation.
I think your solution is feasible for this situation, based on the amount of data / the type of data / type of process.

In your model the trucks are distributed accros the day according to your data per hour.
As a result Trucks will never arrive the next day. However, the queue can be affected (blocked) by the previous day. On the long term this won't be an issue as the arrivals are planned per day (assuming sufficient capacity at the weighing bridge.)

2 in 1 out: this is an inputstrategy. If(And(content(in(1,c)) = 2, content(in(2,c) = 1), - code alternative strategy -, - code current input strategy - )

Regards, Nick
ditmar
Posts: 19
Joined: Friday 21 March, 2014 - 11:45

Re: Finetuning

Post by ditmar »

Hi Nick

thanks voor your reply. All well noted, however still trying to distribute the arrivals different in the multiservice atom. Because I think this will give me a better result. So autofitted some data and it came with: Lognormal incoming and neg.exponential outgoing. However when I simulate this my model takes an hour to simulate one year.

Little bit strange cause when distributed uniform it only takes like 3 minutes ?

Regards

Ditmar
Nick
Posts: 48
Joined: Saturday 15 February, 2014 - 01:52

Re: Finetuning

Post by Nick »

Hi Ditmar,

What is your cycle time formula after you changed the model from Uniform to Lognormal / NegExp? The Uniform distribution uses (min, max); The LogNormal(mean, standarddeviation) and the NegExp(mean).

Something like LogNormal(hr(Label([HourArrival], i)-1), hr(Label([HourArrival], i) ) ) slows your model down siginficantly.
For example, hour 3: Lognormal(2,3) has a mean of 2 hours and a standarddeviation of 3 hours.
This results in < 0 values.

If so, the following warning appears in the tracer:
"Warning: Atom ArrivalHour (ID:180) tries to create a negative event with EventCode '1'. It attempts to set time back by NAN seconds. Rounded to 0"

Regards, Nick
Post Reply