Accurate solar power forecasting transforms a variable renewable asset into a manageable, predictable power source. For grid operators, solar forecasting is essential for dispatch scheduling and balancing. For battery storage operators, generation forecasts directly drive BESS charge/discharge schedules and determine how much arbitrage or ancillary service revenue can be captured. As solar penetration increases globally, forecasting capability is becoming a core competitive differentiator.
Why solar forecasting matters for BESS operations
A BESS co-located with solar can only be optimally operated if the operator knows, with reasonable confidence, how much solar generation to expect in the next 2–48 hours. Without a forecast:
- The BESS might discharge during the day (expecting solar will handle load) when cloud cover reduces solar output — resulting in simultaneous grid import and BESS discharge
- The BESS might hold charge (waiting for a peak demand event that doesn't materialise), missing TOU arbitrage opportunities
- Wholesale market bidding requires forecasting 24–48 hours ahead; poor forecasts lead to under/over-production imbalance penalties in real-time markets
Forecast horizons and applications
| Horizon | Name | Application | Best method |
|---|---|---|---|
| 0–2 hours | Nowcasting / intra-day | Real-time BESS dispatch, ancillary services | Sky cameras, satellite imagery |
| 2–6 hours | Short-term | Intra-day market bidding, BESS scheduling | Numerical weather prediction (NWP) + ML |
| 6–48 hours | Day-ahead | Day-ahead market bidding, maintenance scheduling | NWP ensembles, statistical post-processing |
| 7–30 days | Medium-term | Resource planning, O&M scheduling | Seasonal NWP models, climatological data |
| 1–30 years | Long-term (resource assessment) | Project financing, energy yield analysis | Satellite TMY datasets (Solargis, Meteonorm) |
Irradiance forecasting methods
1. Numerical Weather Prediction (NWP)
NWP models (GFS, ECMWF, NAM) simulate the atmosphere from first principles, producing gridded forecasts of global horizontal irradiance (GHI), temperature, wind and cloud cover at 1–25 km spatial resolution. ECMWF is generally the most accurate NWP model globally; GFS is freely available. Limitations: NWP models have low spatial resolution and systematic biases in complex terrain and coastal zones. Horizon: 6 hours to 10+ days.
2. Satellite-based nowcasting
Geostationary satellites (GOES-16/17 in Americas, Meteosat in Europe, Himawari in Asia) provide cloud motion vector (CMV) forecasts — tracking cloud movement to predict irradiance 0–6 hours ahead. Satellite-based methods outperform NWP at short horizons (0–4 hours) because they see current cloud state at high spatial resolution (1–4 km).
3. Machine learning / statistical post-processing
Modern forecasting systems combine NWP and satellite data as inputs to ML models (gradient boosting, LSTM neural networks) that correct for local biases — site-specific terrain effects, aerosol patterns, shading from nearby hills. ML post-processing typically improves RMSE by 10–25% compared to raw NWP. Requires 1–2 years of on-site irradiance data for training.
Forecast accuracy metrics
- RMSE (Root Mean Squared Error): the primary forecast accuracy metric; expressed as % of installed capacity or kWh. Typical day-ahead RMSE for a 10 MW solar plant: 8–15% of capacity for clear-sky sites; 15–25% in cloudy/variable climates
- nRMSE (normalised RMSE): RMSE expressed as % of mean production, allowing comparison across sites with different capacities
- MAE (Mean Absolute Error): less sensitive to outliers than RMSE; useful for contract penalty calculations where errors above a threshold incur costs
- Skill score: improvement relative to a persistence forecast (naive assumption that current generation equals next-hour generation); a skill score >30% indicates meaningful forecasting value
Integrating forecasts into BESS dispatch
A model predictive control (MPC) BESS dispatch system uses a rolling 24–48 hour solar forecast to optimise charge/discharge scheduling:
- Receive updated NWP-based solar forecast every 6 hours (or every 15 minutes for satellite-based)
- Run optimisation over the forecast horizon, given BESS state of charge, TOU tariff schedule, demand forecast and solar forecast
- Execute optimal schedule; update every 15–30 minutes as forecasts are refreshed
- Track actual vs. forecast generation; feed actuals into ML model for ongoing bias correction
Sungrow's iSolarCloud platform integrates power forecasting with BESS EMS, allowing BESS operators to implement forecast-driven dispatch without third-party systems. For projects participating in wholesale markets, a certified forecast service provider (e.g., Solargis Energy, Solcast, DNV GL Energy Transition) provides bankable forecasts for market bidding.