Elevators

Elevators, despite their ubiquity, are far more intricate than they initially appear. This article aims to demystify the inner workings of these devices. The basic algorithm used by elevators is called SCAN, which was patented in 1961. The elevator follows a straightforward route, starting at the lobby and heading to the top floor before reversing its direction and descending.

It picks up and drops off passengers as it travels. However, most people don't need to travel to the top floor. If the elevator stops only at the required floor before reversing, it employs the LOOK algorithm, which is the standard algorithm most people are familiar with.

When multiple elevators are present, coordinating their movements becomes a challenge. Initially, a central scheduler assigns each elevator to specific floors. When a new request arises, it is allocated to the closest available elevator. However, more efficient algorithms exist. The RSR (Relative System Response) algorithm, developed by Otis, assigns a score to each elevator based on its suitability for a particular passenger request.

The algorithm then re-optimizes its route every 5 seconds, allowing it to adapt to changing circumstances.

The performance of elevator algorithms can be measured by the wait time for an elevator to arrive. The most common metric is the percentage of times an elevator arrives within 30 or 90 seconds. A more rigorous approach involves analyzing the distribution of wait times. By plotting the wait time across thousands of rides, one can create a histogram to visualize the data.

For instance, a p90 of 2 minutes indicates that 90% of the time, riders wait 2 minutes or less for the elevator, while a p50 of 1 minute means half the time, the elevator arrives within 1 minute.

The performance of elevator algorithms can vary depending on the time of day and traffic patterns. In large corporate office buildings, morning rush hours typically see the highest traffic, with riders primarily traveling from the lobby to higher floors. In the evenings, the flow reverses as people leave the building. The distribution of wait times changes significantly based on the time of day and the elevators' workload.

The LOOK algorithm tends to perform poorly during morning rush hours due to its rigid nature, as it assigns each request to the nearest car, regardless of its load.

In contrast, the RSR algorithm demonstrates superior performance during peak times by dynamically adjusting to the changing demands. However, as the flow rate increases and elevators become fuller, the advantage of RSR diminishes. In smaller buildings with fewer elevators per bank, the LOOK algorithm often outperforms RSR. Additionally, journey time, or the duration spent inside the elevator before reaching the desired floor, differs between the two algorithms.

The LOOK algorithm tends to have shorter journey times, while the RSR algorithm may have longer wait times but faster overall travel speeds.

A newer technology, Destination Dispatch, involves kiosks on each floor that allow users to specify their destination before the elevator arrives. These kiosks then direct passengers to the appropriate elevator. However, this advanced system often results in worse wait times compared to the traditional up and down buttons. While the kiosks provide the elevator optimizer with comprehensive information about passenger destinations, they lack flexibility.

The re-balancing step, which occurs every 5 seconds, prevents the system from adapting to changing circumstances. Consequently, the rigid nature of Destination Dispatch negates the potential benefits of the optimizer.