A machine learning framework for construction planning and scheduling

Abstract

In building and infrastructure projects, construction planning and scheduling refer to a process of defining project policies and procedures and breaking them down into specific construction activities, which significantly affect various aspects including cost, time, safety, and quality. Construction planning and scheduling have been shifting from manual to automatic with the adoption of information and communication technologies, and numerous methods, such as optimization algorithms, have also been used in construction planning and scheduling. However, due to the multiplex, evolving, and unstructured nature of sites and tasks, construction planning and scheduling with previous technologies and methods do not work well for practical applications, especially during the execution phase of building and infrastructure projects. With the development of artificial intelligence in recent years, machine learning that is able to deal with complex, dynamic, and uncertain things shows the potential to assist with that problem. To structure and standardize construction planning and scheduling with the application of machine learning, this study proposes a framework with reinforcement learning, imitation learning, and transfer learning, and discusses their respective benefits and limitations. With the proposed framework, application effectiveness and efficiency could be enhanced and application clarity and repeatability cloud be promoted.