Origin Destination Matrix Estimation and Prediction in Vertical Transportation

The dispatching algorithm serving in a building with more than one lift can use passenger flow detailed information based on the requests made by the passengers to go up and down in order to improve the performance of the system. The requests can be analysed in order to extract detailed information about passenger arrival and destination floors and that information used to improve lift assignments. The control strategy can be optimised to look for ways to react to changes in traffic flow. In this paper the traffic flow is described as detailed origin-destination matrixes for 5-minute time intervals. The passenger flow analysis process is divided in two steps, first passenger detailed counting and then, a forecasting process applied to the data coming from the first step. For the first step, the passenger counting process, data obtained from a typical multi-storey office building is used and the results compared with actual manual counts from the same building. After this validation process, the next step uses the data as time series where prediction methods can be applied. Prediction methods can forecast the next time interval traffic flow. Neural networks are able to approximate different time series and are used in this paper with two different data resolution entries, 5-minute interval data and 2.5-minute interval data. The results of both forecasts are compared at a resolution of 5 minutes, and the results show that a methodology of working at a higher resolution to later aggregate the result at a lower resolution can be useful in this context.