Deep multi-objective reinforcement learning for utility-based infrastructural maintenance optimization

• DOI: 10.1007/s00521-024-10954-0
• Publication Date: 2025-01-10T17:38:00Z
• Authors: Jesse van Remmerden; Maurice Kenter; Diederik M. Roijers; Charalampos Andri...; Yingqian Zhang; Zaharah Bukhsh
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Machine Learning (cs.LG)

Abstract In this paper, we introduce multi-objective deep centralized multi-agent actor-critic (MO-DCMAC), a reinforcement learning method for infrastructural maintenance optimization, an area traditionally dominated by single-objective (RL) approaches. Previous RL methods combine multiple objectives, such as probability of collapse and cost, into singular reward signal through reward-shaping. contrast, MO-DCMAC can optimize policy objectives directly, even when the utility function is nonlinear. We evaluated using two functions, which use cost input. The first threshold utility, in should minimize so that never above threshold. second based on failure mode, effects, criticality analysis methodology used asset managers to assess plans. MO-DCMAC, with both environments, including ones case study historical quay walls Amsterdam. performance was compared against rule-based policies heuristics currently constructing Our results demonstrate outperforms traditional across various environments functions.