Lift travel modelling using AI : assessing different lift modernisation solutions on energy consumption

Abstract

This thesis explores the use of artificial intelligence (AI) to advance lift traffic modelling, specifically to increase the accuracy of daily trip estimations as well as their impact on estimated annual energy consumption. Based on a dataset of more than 100,000 lifts supplied by one of the world's leading lift manufacturers, the study compares certain assumptions utilised by other articles based on the standard ISO 25745-2 to data-driven models. The paper applies machine learning methods such as XGBoost, LightGBM, and CatBoost to predict average daily trip numbers given building and lift characteristics. The findings reveal that in certain situations the ISO-based approach overestimates trip quantities, especially for tall buildings with greater trip numbers in a day, hence contributing to inaccuracy of energy consumption reporting. The machine learning models exhibit better predictive performance as well as a higher degree of conformity to real-world data fluctuations. This study also analyses the effect of improving trip estimation accuracy in five lift modernisation scenarios, indicating substantial saving by modernisation as well as the relative changes in running energy estimation with improved trip estimation. The implications of this study demonstrate the potential of AI-driven modelling in enabling more accurate environmental assessment as well as sustainable decision-making in the lift sector.