A Deep Multi-Objective Reinforcement Learning Approach for Infrastructural Maintenance Planning with Non-Linear Utility Functions
Published in Multi-Objective Decision Making Workshop (MODeM) at ECAI 2023, Kraków, 2023
Authors
Jesse van Remmerden, Maurice Kenter, Diederik M. Roijers, Yingqian Zhang, Charalampos Andriotis
Summary
This workshop paper set out the case for treating infrastructural maintenance planning as a genuinely multi-objective problem rather than a single-objective one with a shaped reward, and introduced the deep multi-objective actor–critic approach that was subsequently developed into MO-DCMAC (Neural Computing and Applications, 2025).
The central argument is that the utility functions asset managers actually use are non-linear — a maintenance plan that keeps the probability of collapse below a threshold at moderate cost is not interchangeable with one that achieves a lower expected cost while occasionally exceeding it. Methods that scalarise objectives linearly cannot represent such preferences, and so cannot recover the policies practitioners want.
Links
- Paper: https://modem2023.vub.ac.be/papers/MODeM2023_paper_15.pdf
- Presented at: MODeM 2023, ECAI 2023, Kraków
- Journal version: MO-DCMAC, Neural Computing and Applications (2025)
Recommended citation: van Remmerden, J., Kenter, M., Roijers, D. M., Zhang, Y., & Andriotis, C. (2023). A Deep Multi-Objective Reinforcement Learning Approach for Infrastructural Maintenance Planning with Non-Linear Utility Functions. Multi-Objective Decision Making Workshop (MODeM) at ECAI 2023, Kraków, Poland.
Download Paper