Multi-Agent Reinforcement Learning for Dynamic Resource Scheduling in Engineering Project Networks

Abstract

This study examines the problem of ineffective dynamic resource scheduling in engineering project networks, where limited labor, machinery, materials, technical expertise, and budget must be continuously allocated across interdependent project activities under uncertainty. The purpose of the study was to determine how Multi-Agent Reinforcement Learning (MARL) capabilities influence dynamic resource scheduling performance by improving decision-making, coordination, adaptability, resource optimization, and intelligent scheduling support. A quantitative, cross-sectional, case-based research design was used, focusing on enterprise engineering project cases involving project managers, project engineers, resource and scheduling planners, operations managers, and digital or AI project officers. From 220 distributed questionnaires, 190 were returned, and 180 valid responses were used for final analysis, producing an 81.8% usable response rate. The key independent variables were MARL-based decision-making, agent coordination, dynamic scheduling adaptability, MARL-enabled resource optimization, and intelligent scheduling system capability, while the dependent variable was dynamic resource scheduling performance. Data were collected through a structured five-point Likert-scale questionnaire and analyzed using descriptive statistics, Cronbach’s Alpha reliability testing, Pearson correlation analysis, and multiple regression modeling. The findings showed strong positive perceptions of all constructs, with mean scores of 4.18 for MARL-based decision-making, 4.11 for agent coordination, 4.23 for dynamic scheduling adaptability, 4.16 for resource optimization, 4.09 for intelligent scheduling capability, and 4.21 for scheduling performance. Reliability was strong, with Cronbach’s Alpha values ranging from 0.82 to 0.89. Correlation results showed significant positive relationships with scheduling performance, including MARL-based decision-making (r = 0.68), agent coordination (r = 0.64), dynamic scheduling adaptability (r = 0.72), resource optimization (r = 0.69), and intelligent scheduling capability (r = 0.61), all at p < 0.001. Regression analysis revealed that the model explained 66.6% of the variance in scheduling performance, R² = 0.666, F (5,174) = 69.42, p < 0.001. Dynamic scheduling adaptability was the strongest predictor (β = 0.29), followed by resource optimization (β = 0.24), MARL-based decision-making (β = 0.21), agent coordination (β = 0.18), and intelligent scheduling capability (β = 0.14). The study implies that engineering organizations should adopt adaptive MARL-enabled scheduling systems to reduce conflicts, improve resource utilization, strengthen coordination, and enhance project cost-time performance.