Investigating the utility of artificial intelligence techniques for automatic generation of construction project plans.

摘要

Project planning involves the definition of those project activities and their sequential relationships that are required to complete a project. In this definition, a project plan includes all of the method-related, but not the resource-related, sequencing constraints. Network-based project scheduling techniques, invented during the late 1950's, have become indispensable as aids in scheduling construction projects, especially after being computerized. However, traditional network-based techniques can help only in analyzing a plan, not in generating it. The definition of activities and their predecessors must be provided to such systems by a human planner. Because of this, plan generation has been a challenging area of research for artificial intelligence (AI) researchers since the 1960's.;This thesis explores the utility of artificial intelligence techniques for automating the generation of construction project plans. First, we establish a set of criteria to be used in evaluating existing AI planners. Second, we describe the characteristics while meeting most of the evaluation criteria. We show that SIPE (System for Interactive Planning and Execution monitoring) is the most suitable of the existing AI planning systems for our purposes. Third, we present an experimental implementation of SIPE for planning construction projects with various degrees of activity repetition and across different abstraction levels. In this experimental implementation, we explore several planning issues relevant to the construction domain such as planning at multiple hierarchial abstraction levels, planning with uncertainty, planning repeated cycles of operation, and resolving interaction among activities in parallel branches of a plan. This experimental implementation aids in identifying both the power and some key limitations of SIPE in planning construction projects. Finally, we describe how SIPE has been extended to overcome these limitations, and demonstrate how the extended SIPE planning system can produce correct plans for a range of project planning applications.