A method for adaptive spectral clustering in flexibility markets

Market-based frameworks require aggregating flexible, granular resources into stable trading products to achieve efficient grid congestion management. The heterogeneous nature of demand-side flexibility, which entangles topological constraints with dynamic user preferences, renders static clustering ineffective. Frequent reclustering could capture these dynamics, but it introduces operational volatility and remains mathematically intractable for near-real-time operations. To resolve this stability-efficiency trade-off, this paper proposes a grid-aware State-Adaptive Hybrid Clustering (SAHC) method. Unlike rigid topological partitioning, SAHC employs a spectral approach to capture non-linear interactions between grid physics and economic behavior. The framework integrates a day-ahead initialization—Affinity Aggregation with Spectral Clustering (AASC)—with a dynamic intra-day engine that autonomously switches between a lightweight stateful update (Power Iteration Clustering) and a stateless re-estimation (Nyström approximation) based on spectral structural integrity. Simulations on the IEEE 33-bus system under an abrupt behavioral shift demonstrate that SAHC achieves lower computational latency with minor effects on economic efficiency. The approach attains flexibility costs comparable to high-fidelity benchmarks while reducing computation time by more than one order of magnitude relative to full spectral decomposition, enabling operation under strict real-time constraints. These results support SAHC as a scalable solution that ensures market continuity without the limitations of static approaches.