Improving elevator call time responsiveness via an artificial neural network control mechanism

Elevator traffic comprises the movement of individuals from the floor from which they called the elevator to their destination floor. This project seeks to improve elevator call time responsiveness by utilizing the concept that traffic flows generally form definable patterns that can be used to predict future traffic flow behaviors. A feed-forward neural network-based control algorithm has been developed that can approximate elevator call patterns by learning to associate time of day with specific call locations. This algorithm was tested against fuzzy patterns of elevator calls in which the randomly generated calls were biased towards certain floors at certain times of day. When the average neural network controlled call times of 10 such fuzzy sets were compared to the typical scenario of the elevator returning to the first floor after each call, a 42% improvement in elevator call time responsiveness was observed. It is thereby suggested that a machine learning enabled-elevator control system could result in increased user satisfaction by reducing wait times by helping to ensure that the elevator is at the most likely place the elevator will be called from prior to an individual even pushing the call button. The utility of such an algorithm is likely further enhanced, however, by the fact that having the elevator in the most likely call location can also lead to significant energy savings in that the elevator will need to travel less to pick up prospective passengers.