HCA Framework Improves Explainability in Nonlinear Model Predictive Control
A new framework called Hierarchical Causal Abduction (HCA) aims to make nonlinear Model Predictive Control (MPC) more interpretable for human operators. MPC is widely used in safety-critical infrastructure but its opacity undermines trust. HCA combines physics-informed reasoning via domain knowledge graphs, optimization evidence from Karush–Kuhn–Tucker multipliers, and temporal causal discovery via the PCMCI algorithm. Tested on greenhouse climate, building HVAC, and chemical process engineering, HCA achieved 53% higher explanation accuracy than LIME (0.478 vs. 0.311) using a single cross-domain parameter set.
Key facts
- HCA combines domain knowledge graphs, KKT multipliers, and PCMCI algorithm.
- Tested on greenhouse climate, building HVAC, and chemical process engineering.
- HCA improves explanation accuracy by 53% over LIME (0.478 vs. 0.311).
- Uses a single set of cross-domain parameters.
- Addresses opacity in nonlinear MPC for safety-critical infrastructure.
- Published on arXiv as 2605.10624v1.
- Involves expert validation.
- Aims to generate faithful, human-interpretable explanations.
Entities
Institutions
- arXiv