The duck curve was a CAISO-specific concern in 2013. It's a multi-market problem in 2025. As solar penetration has crossed 15–20% of annual energy generation across CAISO, portions of PJM's eastern zones, and ERCOT's West Texas transmission area, the characteristic afternoon net load ramping pattern has emerged wherever solar capacity is high relative to grid flexibility resources. For commercial building operators, this is no longer an abstract grid planning problem — it's showing up in real-time price signals, demand response dispatch patterns, and the increasingly misaligned timing between when solar generation is cheapest and when building loads naturally peak.
Understanding the duck curve from a building load perspective — rather than a utility planning perspective — reveals a concrete set of operational strategies that convert solar intermittency from a problem into a managed resource. The buildings that navigate this well will have lower energy costs, better demand response performance, and more predictable operating expenses than those that continue running on pre-solar load profiles.
What the Duck Curve Actually Means for a Building
The classic duck curve shows California ISO net load (total grid demand minus solar generation) plotted against time of day. The "belly" of the duck occurs during midday hours when solar generation is highest and net load drops to a minimum. The "neck" rises steeply in the late afternoon as solar generation falls and evening load demand peaks simultaneously. The ramp rate from belly to peak — the duck's neck — represents the rate at which dispatchable generation (gas peakers, hydro, demand response) must activate to replace declining solar output.
For a commercial building, the duck curve's practical implications are:
- LMP volatility: Real-time locational marginal prices are increasingly negative or near-zero during midday solar-abundant hours in high-penetration markets, and spike sharply during the 4–8 PM ramping period. A building with flexible load that can shift consumption from the 4–8 PM window to the 10 AM–2 PM window captures real energy arbitrage value, even at small scales.
- Demand response dispatch timing: ISO emergency demand response events, which trigger demand charge deferrals and capacity payment opportunities, are increasingly concentrated in the late afternoon ramping window rather than the traditional hot-afternoon peak. The 5 PM demand event, not the 2 PM one, is becoming the critical performance window.
- On-site solar value dilution: Buildings with rooftop solar that export to the grid during midday hours receive NEM (net energy metering) credits at a time when grid-level LMPs are low. The financial value of solar generation has shifted toward self-consumption — using the solar output in the building rather than exporting it — because the export value is being compressed by the same overgeneration conditions that create the duck curve.
Cloud Passage Events: The Sub-Hour Intermittency Problem
The duck curve describes the daily shape of solar intermittency — the predictable rise and fall of solar generation over the day. Harder to manage is cloud passage intermittency: the rapid, unpredictable changes in solar output caused by partial cloud cover, which can cause a rooftop solar array's output to drop from full power to 20% of rated capacity in under a minute.
For a building with a 200 kW rooftop solar installation and a building peak demand near 600 kW, a cloud passage event that drops solar output by 160 kW in 45 seconds creates an immediate demand shock: the 160 kW that was being supplied by solar must be immediately picked up from the grid, and if this happens within a 15-minute demand measurement interval that was already running close to the building's monthly peak, it can set a new peak demand record for the month.
The problem is acute for buildings where the BMS treats solar output as a fixed offset to grid demand rather than a variable input. If the building's demand management strategy assumes 200 kW of solar output and pre-conditions HVAC setpoints accordingly, a cloud passage event that drops solar output to 40 kW creates a 160 kW demand spike that the building has no prepared response for.
Consider a mid-size mixed-use building in a PJM-East market — a 10-story, 220,000 sq ft building with 180 kW of rooftop solar, operating on a $17/kW demand charge rate. A March afternoon cloud bank passes over the building during the 3:45–4:00 PM demand interval. Solar drops from 165 kW to 22 kW in under 2 minutes. The grid demand spike registers as a 143 kW increase within the interval. If the building's demand management system was holding HVAC at elevated setpoints based on the expected solar offset, the combined effect — grid demand spike plus HVAC setpoint rebound — can push the interval average to a new monthly peak, adding $2,400 in demand charges for a 15-minute event that the facility manager won't even know happened until the invoice arrives.
Solar-Aware Load Shaping: The Operational Response
Solar-aware load shaping requires two things that are not standard in most commercial BMS configurations: real-time solar output telemetry integrated into the demand management loop, and a load response capability fast enough to compensate for sub-minute solar variability.
On the telemetry side, most commercial solar installations (20 kW and above) use inverters with Modbus TCP or SunSpec-compliant communication interfaces that report real-time output at 1-second resolution. This data is often available but not consumed by the BMS — it goes to the inverter manufacturer's monitoring portal and stops there. Pulling solar output telemetry into the demand management loop requires a simple integration: Modbus TCP read from the inverter, aggregation with the building's main meter data, and a demand forecast model that accounts for solar variability.
On the response side, the loads that can compensate for rapid solar variability are those with fast actuation and high enough capacity to cover the solar swing range. In a typical commercial building, the candidates are:
- EV charging stations: Modern commercial EV chargers (Level 2 and DC fast chargers with OCPP 2.0 or direct demand management integration) can reduce charging power from full rate to near-zero in under 10 seconds. A 50 kW DC fast charger that backs off to 5 kW during a solar cloud event provides 45 kW of near-instantaneous demand compensation. For buildings with 100 kW or more of installed EV charging capacity, this alone covers a substantial fraction of typical cloud event magnitude.
- Battery storage dispatch: A BESS in solar smoothing mode discharges proportionally to solar output drop, maintaining a constant net power draw from the grid regardless of solar variability. This is computationally straightforward — the BESS inverter tracks solar output and modulates discharge to maintain a target net metering point power level. The challenge is that a BESS used for solar smoothing may not be fully charged during peak demand shaving windows.
- HVAC thermal mass: Pre-conditioning building temperature upward (warmer) in the hours before a predicted cloud passage event provides thermal buffer — the building can reduce HVAC output during the solar shortfall period because it has stored cooling capacity in the building structure. This approach works on timescales of 15–30 minutes, not seconds, which limits its utility for sudden cloud passage events but makes it effective for predictable daily solar generation curves (morning and evening shoulder periods).
The Duck Curve Arbitrage Window
Markets with significant solar penetration increasingly exhibit midday periods where wholesale energy prices are very low or negative, followed by steep ramping periods where prices spike. CAISO's real-time market has seen midday prices drop below $0/MWh on multiple days per spring season, and late-afternoon prices climb above $100/MWh during the same days. This price spread represents an arbitrage opportunity for buildings with flexible load or storage: consume more (or charge storage) during the cheap midday window, consume less (or discharge storage) during the expensive ramping window.
In PJM markets, this pattern is less extreme but increasingly visible in high-solar zones (PJM-East, particularly Maryland and New Jersey pricing nodes during summer). Day-ahead LMP spreads between midday and evening peak hours have widened by roughly 30–50% in these zones over the 2021–2024 period as solar capacity additions have accumulated.
For a building enrolled in a real-time energy market DR program, this spread directly translates to revenue. Shifting 100 kW of flexible load — pre-cooling, EV charging, water heating — from a 6 PM window (LMP $80/MWh) to a 12 PM window (LMP $15/MWh) yields $6.50/hour in energy cost savings, or roughly $1,600 per month in a building that can execute this shift consistently across 250 operating hours. Not transformative on its own, but meaningful in the context of a portfolio where this optimization runs automatically across 30 buildings.
We're not saying that every commercial building should be actively arbitraging wholesale energy markets — the administrative complexity of direct market participation has a minimum scale threshold, and most buildings access these benefits through DR aggregators rather than direct ISO market participation. The point is that load shaping aligned with solar generation patterns captures economic value that static, schedule-based building operations leave entirely on the table.
Forecasting Solar Output for Predictive Load Management
The distinction between reactive and predictive solar load management is significant. A reactive system responds to solar output changes after they occur — it reduces EV charging power when solar output drops, then restores it when solar recovers. A predictive system forecasts solar output 15–45 minutes ahead and adjusts building loads in advance to prevent demand spikes from occurring.
Predictive solar forecasting at the building level uses a combination of: local irradiance sensor data (pyranometer or global horizontal irradiance from a weather station), sky imaging or satellite-derived cloud motion vectors, and numerical weather prediction (NWP) model output for the site's location. The forecast accuracy achievable at 15–30 minute horizons using sky imaging or cloud motion vector approaches is typically ±10–15% of installed capacity — sufficient to take pre-emptive action on EV charging deferral and BESS pre-charging decisions, even if not precise enough to eliminate all solar smoothing requirements.
The operational benefit of a 30-minute solar forecast over a purely reactive system is the avoidance of demand spikes caused by the transition from high-solar to low-solar conditions. A predictive system that knows a cloud bank will arrive in 25 minutes can pre-charge the BESS from the grid (at low-cost midday rates), defer EV charging starts, and adjust HVAC setpoints before the solar output drops — smoothing the demand profile across the transition rather than scrambling to respond after the fact.
This is a solvable engineering problem with existing technology. The forecasting models exist; the inverter and BMS integrations exist; the gap for most commercial buildings is the software layer that connects solar forecast data to building load control decisions in real time. That gap is closing as grid-edge intelligence platforms move from proof-of-concept to production deployment in commercial building operations.