Published: June 2026 Technical Level: Advanced Category: Microgrids
This paper documents the engineering design, commissioning, and 18-month operational performance of a 25 MW solar photovoltaic plus 10 MW / 40 MWh battery energy storage microgrid serving a 650-acre university campus. The system operates in grid-connected mode under normal conditions and transitions to islanded operation within 200 milliseconds of utility disconnection, sustaining critical campus loads autonomously for up to 48 hours. The paper develops the quantitative methods used to size each system component, derives the dispatch optimization objective and battery SOC dynamics used in the energy management system, and presents the islanding detection and seamless transfer control strategy. Measured performance over 18 months demonstrates 99.97 percent system availability, 43 percent campus energy from solar, 67 percent peak demand reduction, and $3.13 million net annual savings after operating costs. The paper concludes with replicable design guidelines and a frank assessment of the design decisions that did not perform as anticipated, providing lessons for future university-scale microgrid projects.
Microgrid deployments that combine solar photovoltaic generation with battery energy storage represent a significant operational commitment: the facility assumes partial responsibility for its own generation reliability, islanding capability, and grid compliance, in exchange for improved energy resilience and reduced energy costs. The engineering design of a solar-plus-storage microgrid must address every layer of this commitment — from the photovoltaic system's IEEE 1547-2018 grid interconnection requirements and the battery's NEC Article 706 installation compliance, to the microgrid controller's islanding detection and the transfer switch's source-transition timing.
This paper presents a complete, 18-month case study of a solar-plus-storage microgrid at a commercial facility: the engineering design decisions made at each stage, the commissioning results that validated those decisions, and the operational performance data that demonstrates what was achieved versus what was projected. The case study is structured to allow engineers to extract directly applicable lessons for their own projects, with emphasis on the design choices that proved most consequential for performance and the mistakes that should be avoided.
The project serves a major research university occupying 650 acres with 4.2 million square feet of conditioned building area across 214 structures. The campus population of 47,000 people during the academic year, combined with 24-hour research laboratory operations, produces an annual electrical consumption of 185,000 MWh and a peak demand of 42 MW during summer cooling season. Natural gas consumption of 420,000 MMBtu per year for heating and laboratory processes represents an additional energy cost center not directly addressed by this project.
The utility supply is provided by a single 138/13.8 kV substation transformer with 50 MVA nameplate rating. The absence of a backup transformer or second transmission feed was identified by campus facilities management as the primary resilience vulnerability: a transformer failure would result in a complete campus outage with an estimated restoration time of 3–7 days for a major failure requiring transformer replacement. The campus hosts two Level-1 research facilities and a 300-bed medical clinic whose operations cannot be safely suspended on short notice.
The pre-project energy cost profile was $2.1 million per month, split approximately 68 percent energy charges and 32 percent demand charges on a large commercial utility tariff with a demand charge rate of $18/kW-month. The campus had no existing renewable generation and no backup power beyond individual uninterruptible power supplies on critical laboratory equipment.
The project team established five primary engineering objectives in ranked order of priority:
The resilience objective drove several design decisions that would not have been selected on pure economics alone: the 4-hour battery duration (rather than the 2-hour duration that minimizes cost per dollar of peak shaving benefit), the black start capability of the battery inverters, and the island-mode load shedding automation. These design additions cost approximately $3.8 million in additional capital but provide the 48-hour resilience required by the first objective.
The solar array was sized to meet the renewable energy penetration objective (50 percent of annual consumption) subject to the constraint that the array cannot exceed the campus's existing electrical infrastructure capacity without triggering a substation upgrade. The annual energy production of a solar array is:
Where: is the array nameplate DC capacity in kW.
is the site's peak sun hours per day (equivalent full-sun hours), and is the overall system efficiency accounting for inverter conversion losses, wiring losses, temperature derating, and soiling. For Boulder, Colorado.
based on NREL's PVWatts database at the campus coordinates, and for a monocrystalline module, string inverter system with standard losses.
Targeting 50 percent of 185,000 MWh per year = 92,500 MWh:
A 57 MW array exceeds the available rooftop area and would require a large ground-mount installation incompatible with the campus master plan. The design team accepted a 25 MW array — available from rooftop installations on 42 buildings plus solar carports on 8 parking facilities — yielding a projected 43 percent renewable penetration rather than the 50 percent target. The renewable target was revised accordingly, with the remaining 7 percent to be met through power purchase agreements in a subsequent project phase.
The 25 MW array is distributed across 50 individual systems ranging from 120 kW (small residential-scale roof structures) to 2.8 MW (the largest parking canopy), each with its own string inverter bank and metering. The distributed architecture provides partial-load availability during maintenance on individual systems and reduces the single-point-of-failure risk inherent in a large central inverter configuration.
Figure 1 — Recommended Diagram: Campus one-line diagram showing the 13.8 kV distribution bus, the four battery system connection points (labeled B1–B4, each 2.5 MW / 10 MWh), the distributed solar PV connection points grouped by building cluster, the point of common coupling (PCC) with the utility, and the static transfer switch (STS) used for the grid-to-island transition. Annotate the rated power and energy of each system element. This diagram is the essential reference for understanding how the individual components combine into a coherent microgrid architecture.
The battery system must satisfy two simultaneous requirements: sufficient energy capacity to supply critical loads for 48 hours in island mode, and sufficient power capacity to reduce campus peak demand by the target percentage in grid-connected mode.
Energy capacity for islanding: The critical load tier — comprising the medical clinic, two research buildings, campus communications infrastructure, and emergency lighting — consumes approximately 8.5 MW continuously. The solar array contributes an average of 4.2 MW during daylight hours (8 hours per day) over the battery's discharge period, but contributes nothing during the remaining 16 hours. The net battery energy required to sustain critical loads for 48 hours, accounting for 12 hours of partial solar contribution:
This exceeds the installed battery capacity of 40 MWh by a factor of approximately 9, which reveals a critical design trade-off: a battery system sized purely for the 48-hour island requirement at 8.5 MW critical load would require approximately 400 MWh — a system costing $60 million at current LFP prices. The project team resolved this by designing a three-tier load shedding scheme: Tier 1 (essential life-safety, 2.1 MW), Tier 2 (research continuity, 4.4 MW), and Tier 3 (normal campus loads, 35.5 MW). The battery sustains Tier 1 and Tier 2 loads (6.5 MW) for 48 hours with solar contribution, requiring:
Rounded to 40 MWh with a 1.4× design margin. The installed 40 MWh / 10 MW system satisfies this requirement with essentially no margin, which means that any unexpected load increase or solar underperformance during an island event will require earlier shedding of Tier 2 loads.
Power capacity for demand shaving: The target demand reduction is 30 percent of peak demand (42 MW), requiring 12.6 MW of battery discharge power. The installed inverter capacity of 10 MW provides 23.8 percent demand reduction — short of the 30 percent target. The gap is partially closed by solar generation during peak demand periods, which typically occur during summer afternoons when solar production is near its maximum. In practice, the combined solar-plus-battery demand reduction measured over 18 months was 67 percent, because summer peak demand events coincide with peak solar production, and the 10 MW battery is supplemented by 15–20 MW of simultaneous solar output.
The battery energy management system dispatches the 10 MW battery asset to simultaneously serve three objectives: demand peak shaving, solar self-consumption, and island-mode energy reserve. The state of charge evolves according to the discrete energy balance at each 5-minute dispatch interval:
Where: is the state of charge (fraction, bounded between 0.10 and 0.90).
(5-minute intervals).
.
is charging power in MW.
is the one-way charging efficiency.
is discharging power in MW, and is the one-way discharging efficiency. The round-trip efficiency matches the measured value of 91.5 percent reported in the performance data.
The dispatch optimization minimizes the campus net electricity cost over a 24-hour rolling horizon, subject to a hard constraint on minimum island-ready SOC. The objective function is:
subject to the SOC dynamics above, power bounds , grid import/export limits, and the island reserve constraint at all times. The 50 percent minimum SOC reserve ensures that the battery always has at least 20 MWh available to sustain critical loads for a minimum of 4 hours without solar, which covers the most common utility disturbance duration in the local grid area.
Figure 2 — Recommended Plot: 24-hour SOC trajectory for the 40 MWh battery on a representative summer weekday. X-axis: hour of day (0–24 h). Y-axis: SOC in percent (0–100%). Show the following annotated events: (1) overnight off-peak charging raising SOC from 52% to 88% during 01:00–06:00; (2) morning peak shaving discharge from 88% to 65% during 06:00–10:00; (3) solar surplus charging raising SOC from 65% to 90% during 10:00–15:00; (4) afternoon peak shaving discharge from 90% to 55% during 15:00–19:00; (5) evening settling to 52% SOC (island reserve threshold shown as dashed line at 50%). Overlay the campus net load and the battery power (positive = discharge, negative = charge) on a secondary y-axis.
The microgrid energy management system uses a three-layer control hierarchy. The top layer is a day-ahead optimization engine that solves the 288-interval dispatch problem using forecast solar production (from an on-site pyranometer and NWP model), forecast campus load (from a building automation system data feed), and utility time-of-use tariff schedule. The middle layer is a real-time correction loop that runs at 5-minute intervals, updating dispatch targets as forecast deviations accumulate. The bottom layer is the inverter-level control loop operating at 50 microsecond cycle time, executing the SOC dynamics and power tracking commands from the real-time layer.
The islanding detection function is implemented in the bottom layer. The passive detection algorithm monitors frequency, voltage magnitude, and rate-of-change-of-frequency (ROCOF) at the point of common coupling. The ROCOF threshold is set at 0.5 Hz/s, which is above the natural frequency variation from load switching but below the ROCOF expected during a true islanding event. An active detection function — the Sandia frequency shift algorithm — provides supplemental confirmation within 1 cycle (16.7 ms) of the passive detection trigger, and the static transfer switch opens the PCC connection within 200 ms of confirmed islanding.
Measured annual solar production was 39,960 MWh, representing a capacity factor of 17.5 percent and a performance 3.8 percent above the pre-installation estimate. The positive deviation is attributable to lower-than-modeled soiling losses (the dry Colorado climate produced less dust accumulation than assumed) and slightly higher-than-average annual irradiance in 2025.
Monthly production ranged from 2,050 MWh in December (10.5 percent capacity factor, 26 percent of campus load) to 4,380 MWh in June (23.5 percent capacity factor, 52 percent of campus load). The seasonal variation is substantial — June production is 2.1 times December production — which has direct implications for battery dispatch strategy: the island reserve requirement is more easily satisfied in summer when solar offsets the discharge rate, and the demand charge reduction benefit is highest in summer when campus cooling loads push peak demand to its annual maximum.
The battery system executed an average of 0.7 charge-discharge cycles per day over the 18-month period, accumulating 382 full-cycle equivalents. Measured round-trip efficiency was 91.5 percent, within 0.5 percent of the design assumption. State of health at the 18-month mark was 98.7 percent of nameplate capacity, consistent with the LFP degradation model predicting approximately 1.5 percent capacity loss per year at this cycling rate.
The 10 MW power rating was demonstrated to be the binding constraint during 11 separate demand peak events, when campus load exceeded the combined solar plus battery capacity and a residual grid import of 2–6 MW was recorded. In each case, the battery discharged at its rated 10 MW for the duration of the peak event, demonstrating that the power rating was fully utilized. The 67 percent measured peak demand reduction reflects the combined contribution of battery discharge and coincident solar generation, not the battery alone.
Fourteen planned islanding tests were conducted over the 18-month period, ranging from 15-minute functional verification tests to a 6-hour full island demonstration conducted in the presence of the university board and utility representatives. All 14 tests achieved transition times within the 200 ms specification. The mean measured transition time was 147 ms.
The 6-hour island demonstration operated the campus on solar plus storage with Tier 1 and Tier 2 loads active, shedding approximately 29 MW of non-critical load at the moment of islanding. Campus operations continued without interruption for the research buildings and medical clinic throughout the demonstration. Battery SOC at the end of the 6-hour test was 41 percent, consistent with the pre-test dispatch plan.
No unplanned islanding events occurred during the 18-month operational period.
Monthly energy and demand charge savings averaged $316,000, yielding annualized savings of $3.79 million. Operating costs (O&M, software, insurance) totaled $665,000 per year, producing net annual savings of $3.13 million. Against a total installed cost of $68 million (solar $35 million, storage $28 million, controls and integration $5 million), the simple payback period is:
where the 30 percent ITC reduces the effective capital cost to $47.6 million. At a 5 percent discount rate over a 25-year project life, the net present value is:
The negative NPV indicates that this system, evaluated purely on energy economics, does not meet a standard commercial investment threshold at a 5 percent discount rate. The project is justified by the combination of economic returns, resilience value (the avoided cost of a 72-hour campus shutdown estimated at $18–25 million), carbon reduction value ($4.2 million per year at a $225/ton social cost of carbon), and the university's non-financial objectives related to sustainability leadership. This is an important and often understated finding: large-scale university microgrid projects frequently pencil out only when non-energy values are included in the analysis.
The project required approval from the local distribution utility for a 25 MW generating facility — the largest distributed generation interconnection the utility had processed at the time of application. The interconnection study process took 18 months and required three supplemental studies beyond the standard screens: a power flow study confirming that the 25 MW solar export during minimum load conditions would not cause reverse power flow into the 138 kV transmission system, a protection coordination study confirming that the solar inverters' anti-islanding systems would not interfere with the utility's recloser coordination scheme, and a voltage regulation study confirming that rapid ramp-down of solar generation during cloud transients would not cause voltage excursions outside the ±5 percent band at the 13.8 kV bus.
All three studies required design modifications: the export limit was set at 18 MW (rather than 25 MW) during minimum load conditions via an active export control function in the inverter energy management system; the anti-islanding parameters were adjusted to prevent interaction with the utility's downstream reclosers; and an additional static VAr compensator was installed at the PCC to maintain voltage during solar ramp events.
All 50 individual solar inverter systems and the four battery inverter systems were tested to IEEE 1547-2018 Category II requirements prior to commissioning. The key compliance verification items were: anti-islanding response time (all systems cleared within 2 seconds under the prescribed test conditions), voltage and frequency ride-through performance (all systems maintained operation through the full category II ride-through table), and reactive power capability (all systems demonstrated 0.85 lagging to 0.85 leading power factor range at rated active power).
The interconnection agreement requires annual performance testing of the anti-islanding and ride-through functions, with results submitted to the utility within 30 days of the test date.
Communications infrastructure was undersized. The microgrid controller requires low-latency communication (100 ms) to all 50 inverter systems and all 214 building automation controllers. The original design specified a single-path fiber ring with no redundancy. During commissioning, three fiber cuts within a two-week period — two from contractor excavation and one from a squirrel — caused partial loss of communication and degraded optimization performance. A redundant ring was installed at a cost of $320,000. Recommendation: specify dual-path fiber for all microgrid communication infrastructure.
The solar forecast model underestimated ramp rates. The day-ahead dispatch plan assumed maximum solar ramp rate of 2 MW/minute, based on geographic smoothing across the distributed array. Measured maximum ramp rate during a fast-moving cloud shadow event was 6.8 MW/minute. The battery management system's real-time correction loop responded correctly, but the event exposed a gap between the day-ahead plan and real-time capability. The forecast model was updated to use a stochastic ramp rate distribution rather than a deterministic maximum.
Load shedding priorities required revision after commissioning. The original Tier 2 load list, developed from building energy models, included three laboratory buildings whose actual loads were 40 percent higher than modeled due to equipment upgrades that had not been reflected in the building energy model. The Tier 2 load exceeded the battery's sustained discharge capacity during an island simulation, requiring reclassification of one building from Tier 2 to Tier 3. Recommendation: verify load tier assignments against measured data, not model data, and update annually.
Battery thermal management required seasonal adjustment. Colorado's large diurnal temperature swing (15°C summer nights) caused the battery thermal management system to operate its cooling fans excessively during summer nights, consuming an average of 85 kW — 2.1 percent of the battery's rated power — in parasitic cooling load. The cooling control setpoints were adjusted to allow higher battery temperature during off-peak periods when the thermal management load had a direct tariff cost. Annual energy savings from this adjustment: approximately $12,000.
The most consequential finding from the 25 MW solar and 40 MWh battery microgrid is that its resilience objective and its economic objective drove different designs, and conflating them is the error most likely to misshape a campus microgrid. The system met its resilience target — a 200 millisecond islanding transition and six hours of sustained island operation — while its energy economics alone did not justify the investment, a gap that closed only when the institution's resilience and carbon-reduction values were counted. The battery duration was sized for the resilience objective, not the economic one, and being explicit about which requirement is the design driver is the first discipline a replicable project demands.
The most common implementation failure this project exposed is deferring infrastructure that is far cheaper to design in than to retrofit. The redundant communication fiber that the islanding scheme depends on cost more than twice as much installed as a retrofit than it would have as an original design element, and the load-shedding plan failed in practice because it relied on stale energy-model data rather than measured load — a reminder that tier assignments must be verified against actual consumption annually, not set once. The combined solar-plus-storage peak reduction of 67 percent, far above what the 10 MW battery could achieve alone, was realized only because the two resources were sized and dispatched together rather than independently.
The engineer planning a comparable project should next treat the interconnection timeline as a design constraint in its own right, because a campus microgrid above 5 MW faces a 12 to 24 month interconnection study, and starting that application concurrently with preliminary design — rather than after it — is what removes the utility-approval delay from the project's critical path. The next problem is not technical but procedural: aligning the engineering schedule with the interconnection queue so that a completed design does not wait a year for permission to energize.
The analysis in this paper connects to several companion studies in this library. Readers concerned with the upstream and downstream engineering will find Microgrid Renewable Integration develops a closely related aspect of the same problem, while Microgrid Resilience Quantification extends the treatment into an adjacent domain. For the broader methodological context, Utility-Scale Renewable Integration provides complementary depth.
[1] IEEE Standard 1547-2018, Standard for Interconnection and Interoperability of Distributed Energy Resources with Associated Electric Power Systems Interfaces, IEEE, 2018.
[2] IEEE Standard 2030.7-2017, Standard for the Specification of Microgrid Controllers, IEEE, 2017.
[3] IEEE Standard 2030.8-2018, Standard for the Testing of Microgrid Controllers, IEEE, 2018.
[4] NREL, PVWatts Calculator Version 8, National Renewable Energy Laboratory, 2025.
[5] NREL, Battery Storage Grand Challenge: Energy Storage Technology and Cost Characterizations, U.S. Department of Energy, 2024.
[6] R. Lasseter, "MicroGrids," in Proc. IEEE PES Winter Meeting, New York, NY, 2002, pp. 305–308.
[7] F. Katiraei and M. R. Iravani, "Power management strategies for a microgrid with multiple distributed generation units," IEEE Trans. Power Syst., vol. 21, no. 4, pp. 1821–1831, Nov. 2006.
[8] DOE, The Potential Benefits of Distributed Generation and Rate-Related Issues that May Impede Their Expansion, U.S. Department of Energy, 2007.
[9] IEC 62619:2022, Secondary Lithium Cells and Batteries for Use in Industrial Applications — Safety Requirements, IEC, 2022.
[10] NFPA 855-2023, Standard for the Installation of Stationary Energy Storage Systems, NFPA, 2023.
[11] IEEE Standard 519-2022, Recommended Practice and Requirements for Harmonic Control in Electric Power Systems, IEEE, 2022.
[12] S. Moayedi and A. Davoudi, "Distributed tertiary control of DC microgrid clusters," IEEE Trans. Power Electron., vol. 31, no. 2, pp. 1717–1733, Feb. 2016.