Energy community coalition formation: Why full-resolution data outweighs solver optimality under temporal and consumer compression

Depending on national regulation, energy community formation may require partitioning prosumers and consumers into size-constrained groups to maximise shared electricity. Existing studies mainly address capacity planning and operation, assuming predefined membership, leaving the large-scale formation problem unexplored. We propose a two-level framework (prosumer grouping, then consumer assignment) and a fast greedy algorithm that solves it on full-resolution annual data (35,040 quarter-hourly intervals) without commercial solvers. The approach is evaluated on a Czech community of 230 prosumers and 2470 consumers and compared against mixed-integer linear programming (MILP) on compressed surrogates of the year and the consumer set (representative-day clustering plus consumer-archetype reduction). All solutions are scored out-of-sample on the full uncompressed year. At a typical surrogate resolution the greedy algorithm on full-resolution data outperforms MILP by 0.8% in shared energy value at roughly 6× less compute time (solver plus surrogate construction). Across nine community configurations, greedy matches or exceeds MILP in five; in the remaining four greedy trails by at most 1.3 percentage points. Within the studied regime, full-resolution data matters more than optimal solving; the greedy pipeline requires no commercial solver and runs in minutes on commodity hardware, making it deployable as a default for early-phase Czech-style net-settlement communities.