AbstractsComputer Science

Elevators are used in a large amount of buildings all over the world for fast and comfortable transportation. Today it is becoming increasingly important for people and products to be time efficient, and with technological development new solutions are created to answer this rising demand. To do this in an elevator context, elevator control strategies are implementedas optimal as possible. Machine learning is a relatively new concept, but it is already used in attempts to improve the performance of elevator control strategies. In this report the impact of machine learning on elevator control strategies is investigated in terms of average squared waiting times for the users. Machine learning algorithms can learn from both the current and past environments. The impacts of these two environments are also investigated. Three static elevator strategies and two versions of a machine learning elevator control strategy are implemented and run through a simulator. The results of the investigation show that machine learning has a significant impact on elevator control strategies and is proven to increase time efficiency with at least about 12.5%. Another conclusion drawn is that the current environment is most valuable in the user travel pattern down-peak, while information about previous days especially can improve the performance in the user travel pattern up-peak.