Translate numerical weather prediction fields into site-level wind speed and solar irradiance forecasts.
Oil, Gas & Energy
Short-Term Wind and Solar Generation Forecasting with Machine Learning
Build short-term wind and solar generation forecasts with machine learning, from weather inputs and power curves through to probabilistic forecasts used in market bidding and balancing.
🗓️Duration5 training days
📚Content4 modules · 8 sessions
🎓On completionAccredited attendance certificate
About the programme
Course Overview
Wind and solar output can swing sharply within a single hour, and a trading or balancing decision made on a poor forecast turns directly into imbalance costs or unnecessary curtailment. This course examines how short-term generation forecasts are actually built: translating numerical weather prediction fields into site-level wind speed and irradiance, applying turbine power curves and photovoltaic derating factors, and training machine learning models, from gradient boosting to recurrent neural networks, that correct systematic weather-model error using recent site history. Participants work through nowcasting horizons of a few hours, day-ahead horizons used for market bidding, and the probabilistic forecasts, expressed as quantiles or ensembles, that balancing and risk teams need rather than a single number. Sessions cover forecast combination across multiple weather providers, satellite-derived irradiance nowcasting, and the verification metrics used to judge whether a new model actually improves on the current one. Teaching uses historical wind and solar datasets, a forecasting model-build exercise, and backtesting against realised generation. Participants leave able to specify a forecasting pipeline, choose a model appropriate to the horizon, and communicate forecast uncertainty to trading and grid operations teams.
Expected Learning Outcomes
Apply turbine power curves and photovoltaic derating factors to convert weather forecasts into generation forecasts.
Train machine learning models, including gradient boosting and recurrent neural networks, to correct systematic forecast error.
Produce probabilistic forecasts expressed as quantiles or ensembles rather than a single deterministic value.
Combine forecasts from multiple weather providers into a single blended forecast with lower average error.
Verify forecast performance using error metrics such as mean absolute error, RMSE and skill score against a reference.
Communicate forecast uncertainty to trading, balancing and grid operations teams in terms they can act on.
Who Should Attend
Renewable energy traders and market analysts who bid wind and solar output into day-ahead and intraday markets.
Forecasting analysts and meteorologists supporting wind and solar generation portfolios.
Grid balancing and system operations staff managing variability from wind and solar generation.
Data scientists building or improving machine learning models for renewable generation forecasting.
Asset managers of wind and solar farms responsible for curtailment and imbalance cost management.
Energy risk managers who need to quantify forecast uncertainty for hedging and balancing decisions.
Course Modules
Select any module to see its sessions and points.
01Weather Inputs and Physical Generation Models
2 sessions · 8 points
Session 1From Weather Fields to Site-Level Forecasts
- Explain how numerical weather prediction models generate wind speed and irradiance fields at grid-cell resolution.
- Downscale weather model output to a specific turbine or solar array location using terrain and historical bias correction.
- Identify the forecast horizons relevant to nowcasting, intraday trading and day-ahead market bidding.
- Assess the accuracy limits of raw weather model output before any site-specific correction is applied.
Session 2Turbine and Solar Power Curve Modelling
- Apply a turbine power curve to convert forecast wind speed into expected electrical output, including wake effects.
- Model photovoltaic output from irradiance, panel temperature and known derating factors.
- Account for curtailment, outages and availability when comparing forecast to achievable generation.
- Build a simple physical forecast as a baseline against which machine learning models will be compared.
02Machine Learning Forecasting Models
2 sessions · 8 points
Session 1Model Selection and Feature Engineering
- Engineer features from weather forecasts, historical generation and time-of-day patterns for model training.
- Train gradient boosting models to correct systematic bias in physical wind and solar forecasts.
- Apply recurrent neural network architectures where recent generation history improves short-horizon accuracy.
- Split training and validation data by time period to avoid leakage between correlated weather events.
Session 2Probabilistic and Ensemble Forecasting
- Produce quantile forecasts that express a range of plausible outcomes rather than a single expected value.
- Use weather ensemble members to generate a spread of generation scenarios for risk assessment.
- Interpret forecast intervals correctly when advising traders on the likelihood of extreme deviations.
- Combine deterministic and probabilistic outputs into a single forecast product for downstream users.
03Forecast Combination and Verification
2 sessions · 8 points
Session 1Blending Multiple Forecast Sources
- Combine forecasts from multiple weather providers using weighted averaging or a learned combination model.
- Incorporate satellite-derived irradiance nowcasts to improve very short-horizon solar forecasts.
- Detect when one weather provider's forecast degrades and adjust combination weights accordingly.
- Document the forecast combination methodology so its logic can be audited and reproduced.
Session 2Backtesting and Performance Verification
- Backtest a forecasting model against a historical period that includes both calm and extreme weather events.
- Calculate mean absolute error, RMSE and skill score to compare a new model against the current production forecast.
- Identify systematic error patterns, such as under-forecasting at high wind speeds, for targeted model improvement.
- Set criteria for promoting a new forecasting model from backtest into live production use.
04Applying Forecasts to Trading and Grid Operations
2 sessions · 8 points
Session 1Market Bidding and Imbalance Management
- Use day-ahead forecasts to prepare market bids that balance expected revenue against imbalance risk.
- Recalculate intraday positions as updated forecasts arrive closer to real time.
- Quantify the imbalance cost exposure created by forecast error at different horizons.
- Coordinate forecast updates with trading desks so bidding decisions reflect the latest available information.
Session 2Curtailment and Grid Balancing Applications
- Support curtailment decisions with forecasts that anticipate periods of excess generation relative to grid capacity.
- Share forecast uncertainty with grid operators to inform reserve and balancing capacity planning.
- Evaluate the cost and value trade-off of investing in improved forecasting against reduced imbalance exposure.
- Plan a forecasting system upgrade path as additional wind and solar capacity is added to a portfolio.
What the participant receives
4 course modules
A structured syllabus
8 training sessions
across 5 days
32 detailed points
Applied, detailed content
Accredited attendance certificate
On completing the programme
Complete your registration
We will contact you within one business day to confirm.
Ready to start?
Reserve your seat and start building the skill.