TL;DR
This paper introduces a scalable mathematical framework for designing "collective self-consumption" loops—communities where neighbors share locally produced solar energy. By combining Mixed-Integer Linear Programming (MILP) with Benders and Dantzig-Wolfe decomposition techniques, the authors demonstrate how to optimally select participants and energy flows across large territories and long time horizons that would otherwise crash standard optimization solvers.
What Problem Does This Paper Address?
As we shift toward renewable energy, "collective self-consumption" (CSC) is becoming a cornerstone of the smart grid. In a CSC setup, a group of "prosumers" (producers + consumers) share electricity locally to reduce transmission losses and grid strain.
However, most existing research assumes the community members are already chosen and focuses only on managing the day-to-day energy flow. This paper tackles the much harder design problem: Out of hundreds of potential actors in a semi-rural area, who should be grouped together into which loops?
The challenge is two-fold:
- Legal & Physical Constraints: In countries like France, CSC loops are restricted by geographical distance (e.g., members must be within a certain radius) and total installed power limits.
- Computational Complexity: Designing these loops over a six-month horizon with 100+ actors creates a "combinatorial explosion." A standard "compact" model becomes too massive for even the best solvers to handle in a reasonable timeframe.
Methods
The authors formulate the problem as a Mixed-Integer Linear Programming (MILP) model. To make it scale, they employ two heavy-hitting decomposition strategies:
- Benders Decomposition (Temporal Scaling): This splits the problem into a "Master" problem (deciding loop membership) and a "Sub-problem" (optimizing energy flows for each hour). This allows the model to handle long time horizons by solving the flow optimization independently for different time steps.
- Dantzig-Wolfe & Clique Generation (Spatial Scaling): Since legal rules limit how far apart members can be, the authors represent the territory as a "neighborhood graph." They use a clique generation algorithm to pre-identify feasible groups of actors. This effectively "prunes" the search space, focusing the solver only on groups that are legally and physically compatible.
They compare five variations, ranging from a basic "Single-Loop Compact" (SLCpct) model to a "Benders Extended Multi-Loop" (MLColExt) model that combines both decomposition types.
Datasets
The authors generated realistic test instances to validate their models:
- Consumption Data: Based on reference profiles from Enedis (the French national grid operator), categorized into "Household" and "Professional" profiles.
- Production Data: Solar irradiance data was pulled from the PVGIS database, considering different panel tilts and orientations (Best vs. Worst case scenarios).
- Synthetic Scenarios: They created 100 replicates for various configurations, varying actor density and the legal power limits (from 1 MWc to 5 MWc). These instances are publicly available as a benchmark for the research community.
Key Insights and Ideas
- The "Sparsity" Advantage: Counter-intuitively, the legal distance constraints make the problem easier to solve if you use the right math. By treating the territory as a sparse graph, the authors can use clique generation to eliminate millions of impossible combinations before the solver even starts.
- The Design-Operation Link: The model doesn't just pick members; it ensures they are compatible. A good loop needs a mix of "complementary" profiles—for example, pairing a school (which uses power during the day) with residential homes (which use more in the evening).
- Self-Consumption vs. Self-Production: The paper highlights a crucial distinction. High self-consumption (using all the solar you produce) doesn't always mean high self-production (covering all your needs with solar). The optimal design depends heavily on whether the goal is to maximize profit or maximize energy independence.
Key Results
- Performance Breakthrough: For small problems (10 actors), the standard compact models are fine. However, once the group grows to 30+ actors, the compact models fail to find an optimal solution within an hour. The decomposed models (MLCol), meanwhile, solve these larger problems in seconds or minutes.
- Scalability: The Benders decomposition allowed the models to scale from a 1-day horizon to a 360-day horizon with a nearly linear increase in solve time, whereas the compact model's complexity grew exponentially.
- Sensitivity to Policy: The results show that increasing the "legal power limit" allows for larger, more profitable loops, but it doesn't necessarily increase the number of actors per loop. Instead, the model tends to swap out many small producers for one or two large professional ones to maximize efficiency.
Limitations and Future Work
- Storage: The current models do not include batteries. Adding storage would introduce "inter-temporal" constraints (what you do at 10 AM affects what you can do at 6 PM), which complicates the Benders decomposition.
- Fairness: The objective function currently maximizes global profit. The authors acknowledge that future work needs to address "equity"—ensuring that the financial benefits of the loop are distributed fairly among all participants so they actually want to join.
- Stochasticity: While the model can handle scenarios, more work is needed to make the designs robust against highly unpredictable weather patterns over many years.
Summary updated