Dual-based feature selection for clustering and aggregation of power system components in market-clearing problems

Electricity market studies increasingly require system-scale simulations with large numbers of distributed and flexible assets, making dimensionality reduction via clustering and aggregation essential. The fidelity of these reductions critically depends on feature selection. This paper proposes a market-oriented, quantitative feature selection method that derives clustering features from dual information of a fast market-clearing linear program (LP) model. For each technology and bidding zone, an aggregated LP is solved to obtain parameter-level dual variables. These are combined with parameter variation ranges to compute unit-level effective elasticities, which quantify the relative influence of each parameter on total system cost. The resulting elasticities provide dimensionless weights that scale the candidate features, ensuring that the clustering reflects their true economic relevance. The methodology is compatible with standard LP-based market-clearing formulations without altering convexity or solver settings. Its objective is to retain operational behavior relevant to dispatch and prices while substantially reducing problem size. We validated the method on batteries, demand-side response, and thermal units of a Germany-focused case based on Network Development Plan 2037/2045 to substantiate scalability and benchmark against simpler aggregation baselines.