A Community Microgrid Architecture with an Internal Local Market

• DG means distributed generation, and DSO means distribution system operator.Source: Acronyms · p. 2

• The notation uses d for a device, t for a time period, and u for a community entity; T and U denote the time-period and entity sets.Source: Sets and indices · p. 2
• Device sets distinguish non-flexible loads, non-steerable generators, sheddable loads, steerable generators, and storage devices, with corresponding aggregate sets across entities.Source: Sets and indices · p. 2

• Demand and generation parameters include non-flexible and flexible consumption, non-steerable and steerable generation, and their associated shedding and generation costs.Source: Parameters · p. 2
• Grid and battery operating parameters specify each entity’s maximum import and export, each battery’s maximum charging and discharging power, and its minimum and maximum capacity.Source: Parameters · p. 2
• Battery parameters also include initial and final state of charge, unitary usage cost, and charging and discharging efficiencies.Source: Parameters · p. 2
• Economic parameters cover an entity’s optimal individual profit, the operator’s unitary fee, peak and reserve prices, and grid import and export prices.Source: Parameters · p. 2
• Load-shedding and steerable-generation costs are specified per unit and time period.Source: Parameters · p. 2

• The model includes a slack variable to be maximized and fractions for shedding flexible consumption, activating steerable generation, and charging or discharging batteries.Source: Variables · p. 2
• For each entity and period, variables represent energy exported to or imported from the community and the public grid.Source: Variables · p. 2
• Peak-related variables represent the microgrid’s peak power and the contribution assigned to each entity.Source: Variables · p. 2
• Reserve variables represent symmetric community reserve, entity contributions, and upward or downward reserve from entities and storage devices.Source: Variables · pp. 2–3
• Battery state of charge is tracked by device and time period, while entity profits are decomposed into community total, energy, peak-power, and reserve components.Source: Variables · p. 3
• The notation also includes the optimal profit of the community microgrid.Source: Variables · p. 3

Dual Variables

• The community energy-balance quantity is denoted by µt.Source: Dual Variables · p. 3
• The battery state-of-charge dynamics are associated with σd,t, and the final battery state of charge with ζd.Source: Dual Variables · p. 3
• The entries identify symmetric and downward reserve, storage downward-reserve state of charge, and storage downward-reserve power.Source: Dual Variables · p. 3

1. Introduction

• Community members pool generation, load, or storage resources and connect through a shared local bus for exchanges among themselves and with the public grid.Source: §1 · p. 3

1.1. Related work

• Prior work spans peer-to-peer and operator-managed markets, fairness allocation, and community systems involving storage, demand response, or mixed energy types –.Source: §1.1 · pp. 3–4
• The paper identifies market-oriented pricing for community trades as an open issue and also seeks to quantify the monetary value of community services.Source: §1.1 · p. 4

1.2. Paper contribution

• The proposed internal market determines executed quantities and prices through social-welfare maximization under marginal pricing.Source: §1.2 · p. 4
• The framework identifies three participation benefits: more favorable energy-trading prices, aggregate reserve provision, and lower peak-power costs through netting.Source: §1.2 · p. 4
• A Pareto superior-type criterion requires that community participation does not penalize entities relative to acting individually.Source: §1.2 · p. 4
• The benevolent planner shares community profits by non-discriminatory rules, while the framework also appraises services such as storage and the operator.Source: §1.2 · p. 4
• A real test case inspired by the MeryGrid project in Belgium reports significant yearly cost savings for community members.Source: §1.2 · p. 4

• The community acts as a virtual layer for energy exchanges among entities connected to the same local bus; those exchanges do not cross the bus boundary.Entities may have generation, load, and/or storage devices.Source: §2 · p. 4
• Internal community energy exchanges can be priced more advantageously than exchanges with the external grid.Source: §2 · p. 5
• Pooling entities can reduce the peak power penalty because it applies to the community peak rather than the sum of individual peaks.Reserve can also be pooled and exchanged with the grid at community level.Source: §2 · p. 5
• The operator optimizes day-ahead decisions using grid prices and entity forecasts, and manages energy flows and interactions with entities and the public grid.The operator must be contractually bound to the DSO and community entities, which pay it a fee.Source: §2 · p. 5

3.1. The bilevel model

• The bilevel model uses lower-level market clearing to determine quantities, reserve, and peak power for social welfare, while the upper level shares profits so no entity is worse off than acting individually.The lower level’s dual variables provide market prices; the upper level acts as a benevolent planner.Source: §3.1 · p. 6

3.1.1. Lower level problem

• The lower-level linear program maximizes community social welfare subject to energy, storage, reserve, and peak constraints, with dual variables yielding community market prices.At the optimum, internal monetary flows offset except for the community operator’s fee.Source: §3.1.1 · p. 7

3.1.2. Upper level problem

• The upper level shares the community’s optimal profit among entities and maximizes the minimum profit improvement over each entity’s individual benchmark.Its reserve-sharing policy can limit each entity’s credited contribution to its average reserve contribution.Source: §3.1.2 · p. 8

3.2. Solution strategies

• With a unique lower-level solution, the bilevel model can be solved as two sequential linear programs; non-uniqueness permits a single nonlinear reformulation using lower-level optimality and strong duality.Source: §3.2 · p. 8
• The lower-level problem is feasible when each entity’s individual benchmark problem is feasible, since setting community exchanges to zero gives a decoupled feasible solution.Source: §3.2 · p. 8

4.1. Illustrative examples

• The illustrative cases demonstrate the proposed approach and the soundness of its solutions.The paper then turns to a real-size test case to assess yearly cost savings for community members.Source: §4 · p. 8

4.1.1. Excess of local generation, one time period

• The grid sets the local price for the exporting generator, while the buyer saves by purchasing below the grid price and avoiding the community peak penalty.The generator’s 2 kWh excess is exported to the grid; the community supplies the load’s 3 kWh demand.Source: §4.1.1 · p. 9, Figure 3 · p. 9

4.1.2. Shortage of local generation, one time period

• The load buys 5 kWh from the community and 3 kWh from the grid; the upper-level optimizer assigns the community’s entire peak cost to the non-importing generator.The grid is the marginal producer and sets the local price for the load.Source: §4.1.2 · pp. 9–10, Figure 4 · p. 9

4.1.3. Storage, two time periods, no shared peak

• Storage charges from local generation in period 1 and supplies the load in period 2, avoiding its peak penalty even though its community energy price exceeds the grid price.Reducing storage capacity narrows the feasible set and reduces overall community cost savings; the storage still gains 0.159 e by participating.Source: §4.1.3 · p. 10

4.1.4. Storage, two time periods, shared peak

• In the shared-peak storage case, all three entities benefit from the community market solution.The optimization solution uses equal grid imports across the two periods to reduce peak costs, storing period-1 imports for period 2.Source: §4.1.4 · pp. 10–11, Figure 6 · p. 11

4.1.5. Flexible demand and generation, one time period

• With flexible demand and generation, the load continues to shed all demand, while the other load is fully supplied by the steerable generator through the community, avoiding peak costs.The community’s entire gain corresponds to the gain of the supplied load.Source: §4.1.5 · p. 11, Figure 7 · p. 11

4.1.6. Steerable generation, one time period, reserve

• With reserve provision, marginal prices account for generators’ opportunity cost of reserve, making community energy sales preferable; all three entities benefit from the market solution.The community also allocates some reserve revenue to an entity that produces at maximum capacity and does not contribute to symmetric reserve.Source: §4.1.6 · p. 12, Figure 8 · p. 12

4.2. The MeryGrid project test case

• The MeryGrid test uses 2017 demand and generation profiles at 15-minute intervals, solving one 96-period optimization instance per day.Each instance has around 5,500 variables and 5,900 constraints, takes around 6.36 seconds on average, and excludes reserve provision because the storage tends to provide reserve instead of participating in the community as reserve prices rise.Source: §4.2 · pp. 12–13
• Monthly storage-use costs and operator fees are proportional to the monthly energy exchanged with storage and among community entities, respectively.Monthly gains are cumulative daily gains, and the stacked bars show the community’s total monthly gain.Source: §4.2 · p. 13
• On a yearly scale, entities 1 and 2 achieve cost savings of 28.36% and 37.98%, entity 3 has a revenue increase of 73.47%, and entity 4 gains 180 e.Entity 4 typically sells energy at a price that only balances its energy-purchase costs, including efficiency losses, and storage usage.Source: §4.2 · p. 13
• Median percent daily gains range from 19% to 62% across the reported cases and vary significantly day to day; entity 4’s percent gains cannot be computed because its individual daily profit is always zero.Source: §4.2 · p. 13

5. Conclusions

• Bilevel programming structures the internal market, using marginal pricing to maximize social welfare and a Pareto-superior condition to protect each entity against its individual outcome.Source: §5 · p. 13
• Members gained 28%–74% on a yearly scale compared with acting individually in a real Belgian test case.The framework imposes a Pareto-superior condition so no entity is penalized by participating in the community.Source: §5 · p. 13
• Extensions to curtailment, community reserve and peak sharing, and advanced demand-side models are possible while the lower-level problem remains a linear program.Source: §5 · p. 13

Appendix B. Strong duality

• The strong-duality condition equates the primal and dual lower-level objective values at the optimum.Source: Appendix B · p. 14

Appendix C.1. Sum of bilinear terms involving community prices and community import/export

• Complementary-slackness conditions and equation (12) are used to establish identity (25) at the lower-level optimum.Source: Appendix C.1 · p. 14

Appendix C.2. No simultaneous import/export

• An entity cannot simultaneously import from and export to the grid when the stated grid-price condition holds; nonzero community fees also rule out simultaneous community import and export.Source: Appendix C.2 · p. 15