Hybridization of reinforcement learning and agent-based modeling to optimize construction planning and scheduling
Introduction
Construction planning and scheduling is the process of determining what activities are performed and establishing how and when these activities are conducted within the limits of the available time, budget, and resources [1]. According to the Project Management Institute (PMI), planning activities consists of transforming the scope of work to establish a hierarchy of manageable work packages, also called a work breakdown structure (WBS) [2,3], and then determining the sequence of activities' execution according to project constraints including work environment layout, available resources, and scope. In the same manner, construction planning enables a project to accomplish a set of required objectives that can be considered as a two-part problem. First, the solution needs to capture the dynamic construction environment with activities representing project scopes that can be defined as a hierarchy of executable work packages. Second, the solution is a result of estimating duration requirements for activities and optimizing activity sequencing based on multiple and pre-determined constraints that also incorporate decision makers' knowledge and experience. Construction planning includes scheduling and other forms of planning, such as material handling, site layout planning, equipment path planning, and site logistics planning [4]. Scheduling problems are an important part of construction planning activities in terms of planning physical construction project components that have a specified set of start and finish timelines and an estimated duration.
Researchers have proposed multiple decision-aid methods, such as simulation, optimization, multi-criteria decision-making, and automation, to tackle activity sequencing and WBS formations in construction scheduling problems [4]. Some methods include linear programming, heuristic or meta-heuristic approaches, and hybrid simulation approaches such as discrete event simulation-genetic algorithm (DES-GA). These methods have proposed solutions by solving mathematical objective functions that optimize a given metric, such as time, cost, resource, or quality. These approaches have some shortcomings in capturing uncertainty in the construction environment, raising computational burdens, and not being easily generalizable to multiple construction projects. In a scheduling problem, the optimization process needs to consider multiple constraints tied to each activity, such as time, budget, and resources. These constraints can include 1) precedence relationships, 2) project manager preferences, such as activity associated with a rented crane may need to take precedence to minimize equipment rental costs, and 3) interruptions, such as equipment breakdowns. To tackle these constraints, methods are needed that can capitalize on the simulated environment to understand complex behaviors and derive more sufficient decisions.
Reinforcement learning (RL) is very effective for decision-making processes in construction problems. RL algorithms are able to solve optimization problems with higher constraints [5] and perform efficiently with increasing complexity and number of activities [6]. The RL agent learns to implement better actions, including optimal sequencing of activities, through training achieved from exploiting local rewards and exploring random actions despite lower rewards. Hence, RL can help fill the aforementioned shortcomings of current decision-aid methods in construction planning by developing a local decision-making policy for each agent, based on communication channels, and by breaking down the problem into sub-problems, all of which contributes to computational efficiency. Using RL assists construction practitioners in facilitating generalizations through the learning process, because different problems can be broken down into similar sub-problems. Moreover, RL facilitates agent communications and enables agents to arrive at a set of decisions involving a set of joint actions. This results in a faster convergence to the optimum global policy. However, an RL process does not capture the dynamic nature of modeling in the construction environment, because of the complexity caused by various interactions between system components [7]. In a construction setting, however, having a model of the construction environment is crucial.
Simulation techniques have been used to capture the dynamic nature of the construction environment as well as uncertainties in the modeling process [8]. Compared to other simulation techniques, such as DES and system dynamics (SD), agent-based modeling (ABM) is able to handle these complexities and capture emerging behaviors. ABM is capable of handling very complex real-world systems often containing large amounts of autonomous, goal-driven, and adapting agents [9]. ABM uses a bottom-up approach where the system is described as interacting objects with their behaviors, which allow complex emergent behaviors to be captured. ABM enables tracking of agent interactions in their artificial environments to understand overall processes that lead to global patterns [10]. By incorporating ABM in an RL process, necessary features that support environment modeling, such as system parameters, system behaviors, and rules, are provided in order to enable an efficient representation of the dynamic construction environment and provide the RL platform with the necessary features to support environment modeling.
The objective of this paper is to propose an RL-ABM method with graph networks that can be used to support decision-making in construction planning by providing optimum work package sequencing to schedule activities based on project constraints. The application of the proposed model can be extended to establishing a WBS for a construction project. Three case studies were used to demonstrate the proposed model and discuss the applicability of RL-ABM to addressing similar problems related to activity sequencing. The developed RL-ABM method enables construction decision-makers to evaluate project objectives, facilitates the optimization of multiple types of resources during planning through the RL agent's learning ability, is able to incorporate resource planning during schedule development, and can be generalized to other construction planning problems. Moreover, the applications of the method can be extended to scope definition (WBS formulation) at the project level in future work that will extend this study.
The rest of this paper is structured as follows. First, as background, a literature review section is presented, which discusses decision-making in construction planning and shortcomings of current decision-aid approaches to scheduling problems, followed by an introduction of simulation approaches and RL to address the gap in the literature. Next, the theoretical development of RL-ABM is presented as part of the proposed methodology, which also includes the steps of problem definition, ABM simulation, and development of the RL model. Three case studies are then presented to demonstrate application of the proposed RL-ABM method. Finally, conclusions are presented and recommendations for future work are discussed.
Section snippets
Background
This section provides an overview of decision making in construction planning. Simulation approaches and RL are then discussed along with the knowledge gap existing in the construction planning literature.
Methodology
The research methodology of this study consists of four steps: 1) development of the RL model, 2) problem definition, 3) ABM simulation process, and 4) development of the RL process for construction planning.
Case studies
To demonstrate the proposed RL-ABM methodology, this study utilized construction planning case studies elaborated from three scheduling problems. The first two are described in Lu and Li [77]. Case study 1 illustrates how to utilize the proposed RL-ABM method to address a simple scheduling problem. Case study 2 demonstrates the applicability of the proposed model in construction planning to address a more complicated scheduling problem from a bridge construction project. Case study 3 is a more
Conclusions and future work
In construction planning, the optimal solution for sequencing activities is often selected from a set of finite solutions. However, the optimization problem is everchanging, because the environment, which includes the number of activities, type, and number of allocated resources, changes during execution of the project. Agents in RL algorithms learn better solutions even as the environment changes. A review of the literature emphasizes the need for an effective decision-making tool that can be
Data availability statement
All data, models, and code generated or used during the study appear in the published article.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
This research is funded by the Natural Sciences and Engineering Research Council of Canada Industrial Research Chair in Strategic Construction Modeling and Delivery (NSERC IRCPJ 428226–15), which is held by Dr. Aminah Robinson Fayek.
References (78)
- et al.
Automating look-ahead schedule generation for construction using linked-data based constraint checking and reinforcement learning
Autom. Constr.
(2022) - et al.
Barriers to adopting simulation modelling in construction industry
Autom. Constr.
(2020) - et al.
A review of application of multi-criteria decision making methods in construction
Autom. Constr.
(2014) - et al.
On the effect of subjective, objective and combinative weighting in multiple criteria decision making: a case study on impact optimization of composites
Expert Syst. Appl.
(2016) - et al.
Construction site layout planning using multi-objective artificial bee colony algorithm with Levy flights
Autom. Constr.
(2014) - et al.
A new heuristic algorithm for the operating room scheduling problem
Comput. Ind. Eng.
(2011) - et al.
Hybrid of genetic algorithm and simulated annealing for multiple project scheduling with multiple resource constraints
Autom. Constr.
(2009) - et al.
Knowledge-based schedule generation and evaluation
Adv. Eng. Inform.
(2010) - et al.
Integrating discrete event simulation and genetic algorithm optimization for bridge maintenance planning
Autom. Constr.
(2021) - et al.
Analysis of earth-moving systems using discrete-event simulation
Alexandria Eng. J.
(2015)
Developing an evacuation evaluation model for offshore oil and gas platforms using BIM and agent-based model
Autom. Constr.
Agent-based modeling and simulation of earthmoving operations
Autom. Constr.
An energy-aware, agent-based maintenance-scheduling framework to improve occupant satisfaction
Autom. Constr.
Modeling technological innovation risks of an entrepreneurial team using system dynamics: an agent-based perspective
Technol. Forecast. Soc. Chang.
Asynchronous n-step Q-learning adaptive traffic signal control
J. Intell. Transp. Syst.
Reinforcement learning based process optimization and strategy development in conventional tunneling
Autom. Constr.
Reinforcement learning for building controls: the opportunities and challenges
Appl. Energy
Whole building energy model for HVAC optimal control: a practical framework based on deep reinforcement learning
Energy Build.
Artificial intelligence based smart energy community management: a reinforcement learning approach
CSEE J. Power Energy Syst.
A distributed approach solving partially flexible job-shop scheduling problem with a Q-learning effect
IFAC-PapersOnLine.
Deep reinforcement learning-based controller for path following of an unmanned surface vehicle
Ocean Eng.
Particle swarm optimization for resource-constrained project scheduling
Int. J. Proj. Manag.
Factors affecting construction-planning outcomes
J. Constr. Eng. Manag.
A guide to the project management body of knowledge (PMBOK® guide)-fifth edition
Proj. Manag. J.
Knowledge-based project planning
Automated methods and systems for construction planning and scheduling: critical review of three decades of research
J. Constr. Eng. Manag.
Experimental evaluation of agent-based approaches to solving multi-mode resource-constrained project scheduling problem
Cybern. Syst.
Fuzzy agent-based modeling of construction crew motivation and performance
J. Comput. Civ. Eng.
Agent-based simulation tutorial - simulation of emergent behavior and differences between agent-based simulation and discrete-event simulation
Using agent-based modeling to study construction labor productivity as an emergent property of individual and crew interactions
J. Constr. Eng. Manag.
Fuzzy agent-based multicriteria decision-making model for analyzing construction crew performance
J. Manag. Eng.
Synthesis of decision-making research in construction
J. Constr. Eng. Manag.
A review of methods and algorithms for optimizing construction scheduling
J. Oper. Res. Soc.
Construction progress control
Resource supply-demand matching scheduling approach for construction workface planning
J. Constr. Eng. Manag.
Optimal path planning for UAVs using genetic algorithm
Resource optimization using combined simulation and genetic algorithms
J. Constr. Eng. Manag.
Earthwork optimization system for sustainable highway construction
The logical precedence network planning of projects, considering the finish-to-start (FS) relations, using neural networks
Int. J. Adv. Manuf. Technol.
Cited by (8)
Multi-agent deep reinforcement learning based decision support model for resilient community post-hazard recovery
2024, Reliability Engineering and System SafetyExtensions of the resource-constrained project scheduling problem
2023, Automation in ConstructionApplying Machine Learning and Particle Swarm Optimization for predictive modeling and cost optimization in construction project management
2024, Asian Journal of Civil EngineeringAgent-Based Simulation of Multi-Crew Allocation to Scattered Repetitive Projects
2024, Construction Research Congress 2024, CRC 2024