A Reinforcement Learning Approach to 4D Crane Lifting Simulation for Tower Crane Cycle Time Estimation

A tower crane is critical equipment used to deliver resources from one place to another at height. Existing approaches to crane lift planning however are often limited as path planning and lift time prediction tend to be performed separately, which may result in infeasible lift time estimation in practice. This study thus proposes the integration of reinforcement learning-based simulation with lift time prediction, taking into account the spatio-temporal characteristics of crane movements. The resulting machine learning model using as an input the cycle time estimated from simulation achieves a correlation of 0.9548 with the actual time measured in a field. The proposed approach produces more feasible results, which may more likely occur in practice, compared to the traditional ones using only historical data. This study may contribute to the improvement of planning practice by providing insight into the use of a 3D model (e.g., BIM) for the automated generation of simulation and prediction of lift time at sites.