Everyone has shared the vexation of waiting for an elevator that ne'er seems to arrive. "I pressed the button, why isn't it coming?" you ask. For thing arsenic commonplace arsenic elevators, they are acold much analyzable than meets the eye.
Over the people of this article, we'll unravel the mysteries of elevators. The measurement you push their buttons, and really they push yours.
One Car
The simplest elevator algorithm is called SCAN and was patented successful 1961. The elevator starts astatine the lobby and goes each the measurement to the apical level earlier reversing and coming backmost down. It picks up and drops disconnected anybody connected the way.
Most of the clip you don't really request to spell to the TOP floor. If the elevator goes only arsenic precocious arsenic requested earlier reversing, the algorithm is called the LOOK algorithm. This is the algorithm astir group cognize and expect.
Multiple Cars
Here's wherever the enigma begins. If location are aggregate elevators, really do the cars coordinate who picks up who?
In the astir basal system, there's a cardinal scheduler that tells each elevator which floors to extremity on. When a caller petition comes in, it's assigned to the closest elevator. As we'll soon spot however, we tin do better.
Long Waits
How do you really measurement really bully an elevator algorithm is? The evident metric is really agelong you hold for the elevator to arrive.
A very elemental measurement is "how often does the elevator get wrong 30 seconds?" Or "how often does the elevator get wrong 90 seconds?"
-
wait < 30s
-
wait < 90s
flow14/min
Applied Stats
More rigorously, we want to look astatine the DISTRIBUTION of hold times. If we crippled the hold clip crossed thousands of rides, we get the histogram below.
010s
p50—p90—
flow14/min
A p90 of 2m intends 90% of the time, riders hold 2m aliases little for the elevator. A p50 of 1m intends half the clip the elevator arrives wrong 1m.
People don't usually retrieve the mean magnitude of clip they wait. They fixate connected those times erstwhile the elevator took FOREVER, the p90 case.
Morning Rush
Not each rider postulation is created equal. Imagine a ample firm agency building. In the mornings, astir each postulation is dominated by trips from the lobby to the precocious levels.
In the evening this flips arsenic everyone leaves the building. The luncheon unreserved is simply a spot of both, and the remaining postulation is often from level to floor.
-
wait < 30s
-
wait < 90s
flow14/min
The distribution of hold times varies drastically depending connected the clip of time and the postulation patterns the elevators are facing. Morning unreserved notoriously has the worst hold statistics.
Smarter Elevators
When analyzing the LOOK elevator algorithm, we LOOKED (ha ha) astatine really riders are assigned to cars. We naively assigned each petition to the nearest car but said we could do better.
What if the nearest car is full? We tin get smarter pinch Otis' RSR (Relative System Response) algorithm. RSR scores each car for really good suited it is to prime up a passenger. Lower scores being better.
RSR pickup score
Score=ETA to pickup+onboard load penalty+same-direction anti-bunching penalty-direction-match bonus-idle-nearby bonus-low-load bonus
Anti-BunchingPenalize a car if different car is already headed to the aforesaid level successful the aforesaid direction.
Idle NearbyReward idle cars wrong 2 floors of the caller.
RSR besides re-optimizes each 5 seconds. A rider that's going to beryllium picked up by elevator A tin beryllium re-routed to elevator B if elevator A encounters delays. This re-optimization turns retired to beryllium cardinal for streamlining postulation flow.
In the schematic below, each elevator lights up erstwhile it's the champion prime to work a telephone from level 3 if the fastener happened to beryllium pressed astatine that nonstop moment. This perpetually changes arsenic the elevators move, showing the optimizer successful motion.
LOOK vs RSR
Armed pinch our elevator study toolkit, we tin benchmark the capacity of LOOK vs RSR to spot really overmuch a smarter elevator algorithm really improves hold time.
LOOK
-wait < 30s
-wait < 90s
RSR
-wait < 30s
-wait < 90s
flow8/min
Interestingly arsenic the travel complaint gets higher, LOOK really starts to outperform RSR. When the elevators are ever afloat and stopping connected each floor, the other rules don't matter arsenic much.
LOOK besides tends to outperform RSR successful mini buildings pinch less elevators per bank. Sometimes it's amended to conscionable support things simple.
Another metric you tin way is travel time, really agelong you're really waiting successful the elevator earlier getting to your floor. RSR and LOOK person different characteristics present arsenic good but that's beyond the scope of this article.
Destination Dispatch
Not each elevators person buttons successful them. Some of the fancy caller elevators person a stalls connected each level that allows you to specify what level you're heading to earlier the elevator moreover arrives. The stalls past points you to which elevator you should hold for.
This is called Destination Dispatch. At first glance, it seems great. The elevator optimizer now has afloat knowledge of who is going where, surely we tin usage this to trim hold times right?
RSR
-wait < 30s
-wait < 90s
Destination Dispatch
-wait < 30s
-wait < 90s
flow8/min
It turns retired these fancy kiosks are successful wide worse for hold times compared to the accepted bully ol' up and down buttons. There are surely separator cases erstwhile the kiosks tin triumph retired (extremely gangly buildings pinch 8+ cars per elevator bank) but for the mostly of cases, elemental up down buttons reign supreme.
This counterintuitive consequence is each acknowledgment to the rebalancing measurement wherever each 5 seconds, the strategy re-optimizes each elevator's path. The stalls enforces rigidity, you must get successful the assigned elevator.
The authorities of the world 30sec aft you called your elevator mightiness beryllium very different but the strategy is incapable to adapt. Turns retired the nonaccomplishment successful elasticity is not worthy the other accusation for the optimizer.
Full Sim
Here's a simulation pinch each the buttons and knobs to play with. Go crazy!
-
wait < 30s
-
wait < 90s
floors8cars4flow18/min
Conclusion
This article conscionable scratches the aboveground of elevator algorithms. Next clip you're stuck waiting for an elevator, effort not to return it personally. The elevator did perceive you, it conscionable has a batch to deliberation about.
English (US) ·
Indonesian (ID) ·