PV Yield Forecasting for Aggregation: GHI, Cloud Cover, and Soiling Models

🎯 Learning Objectives

✓ Calculate clear-sky GHI using the REST2 model and adjust for cloud cover using satellite-derived cloud optical depth
✓ Apply empirical soiling rate models to quantify daily energy loss due to particulate accumulation on PV modules
✓ Analyze forecast error propagation across an aggregated fleet using ensemble-weighted RMSE metrics
✓ Design a multi-input forecasting pipeline that fuses ground-based pyranometer data, NWP outputs, and real-time soiling sensor feedback
✓ Explain the impact of spatial correlation decay on aggregation-level forecast uncertainty at 15-minute vs. hourly intervals

📖 Why This Matters

In modern grid operations, aggregators manage thousands of rooftop and commercial PV systems as a single controllable resource—but inaccurate yield forecasts cause costly imbalance penalties, inefficient dispatch, and missed arbitrage opportunities. For mining sites with off-grid solar-diesel hybrids or remote mineral processing plants relying on solar microgrids, forecasting errors directly impact fuel logistics, battery cycling, and production uptime. This lesson bridges meteorology, PV physics, and aggregation architecture to build robust, auditable forecasts.

📘 Core Principles

Forecasting begins with Global Horizontal Irradiance (GHI) modeling: clear-sky GHI is derived from solar geometry and atmospheric transmittance (e.g., REST2 or Ineichen-Perez), then attenuated by cloud cover via cloud optical depth (COD) or cloud fraction scaling. Soiling introduces time-varying, location-specific losses modeled as exponential decay (e.g., η_soil(t) = exp(−k·t)) or stepwise loss per rainfall event. Aggregation requires spatial downscaling of NWP data, accounting for terrain shading, panel orientation diversity, and statistical correlation between sites—critical because uncorrelated errors partially cancel, improving aggregate accuracy over individual site forecasts.

📐 Cloud-Attenuated GHI with Soiling Correction

This formula computes aggregated hourly PV yield per kWp, combining irradiance physics and degradation dynamics. It’s used in day-ahead bidding and real-time dispatch loops within DERMS platforms.

💡 Worked Example

Problem: Given: 500 kWp fleet across 3 mining sites (A: 200 kWp, B: 150 kWp, C: 150 kWp); clear-sky GHI = 850 W/m²; cloud optical depth = 4.2 (moderate overcast); soiling rates: A=0.15%/day, B=0.22%/day, C=0.18%/day; days since last rain = 12; module efficiency = 20%; system PR = 0.82.

1. Step 1: Compute cloud-transmitted GHI using REST2-derived transmittance: τ_cloud = exp(−0.27 × COD) = exp(−0.27 × 4.2) ≈ 0.32 → GHI_cloud = 850 × 0.32 = 272 W/m²

2. Step 2: Calculate soiling factor per site: η_soil,A = exp(−0.0015 × 12) = 0.982; η_soil,B = exp(−0.0022 × 12) = 0.974; η_soil,C = exp(−0.0018 × 12) = 0.979

3. Step 3: Compute weighted yield: Y_agg = Σ [Pp,kWp × GHI_cloud × η_module × PR × η_soil,k] / Σ Pp,kWp = (200×272×0.20×0.82×0.982 + 150×272×0.20×0.82×0.974 + 150×272×0.20×0.82×0.979) / 500

4. Step 4: Numerator = (200×44.6) + (150×44.2) + (150×44.4) = 8920 + 6630 + 6660 = 22,210 → Y_agg = 22,210 / 500 = 44.42 W/kWp

5. Step 5: Convert to kWh/kWp: 44.42 W/kWp × 1 h = 0.0444 kWh/kWp → Total forecast = 500 × 0.0444 = 22.2 kWh

Answer: The aggregated forecast yield is 22.2 kWh for the hour, which falls within the typical operational range of 18–28 kWh/kWp under moderate cloud cover and <15-day dry spell conditions.

🏗️ Real-World Application

At the Tasiast Gold Mine (Mauritania), Kinross deployed a 34 MWp solar farm integrated with a 20 MW diesel plant. Their DER aggregator uses a hybrid forecast: ECMWF NWP feeds a REST2-GHI engine, corrected hourly by Meteosat Second Generation cloud motion vectors and on-site soiling sensors (Kipp & Zonen SMP11). During the Saharan dust season (Mar–May), soiling rates peak at 0.45%/day; the system dynamically adjusts cleaning schedules and bids conservatively into the West African Power Pool (WAPP) market—reducing imbalance penalties by 63% year-on-year (IEA PVPS Task 15 Report, 2023).