Solar Power Prediction: Over 3,000 Predictive Models Deployed Across 787 Plants

  • Predictive analysis
  • Energy

Challenge

Global implementation of incentive and penalty systems for renewable energy generation forecasting, including in Korea, encounters a significant obstacle: the unpredictable environmental conditions inherent in solar power generation, such as sunlight and weather changes. These conditions make accurately predicting power generation using rule-based methods difficult, underscoring the imperative to enhance prediction accuracy for effective power supply planning.

Approach

MakinaRocks adopted a diverse range of ensemble models to bolster solar power forecasting precision. By automating AI operations, we analyzed historical weather and time series data, enabling hourly generation forecasts tailored to plant locations. A specialized approach was developed to minimize forecast error for plants lacking generation history, strategically applied to minimize overall error.

Results

The implementation of an automated system (MLOps) facilitated the rapid retraining and deployment of over 3,000 models across 787 power plants, leveraging approximately two weeks of data. This resulted in forecasts boasting over 94% accuracy, thereby maximizing power trading income and ensuring the reliability of green power supply.

Want to learn more about this use case?
Get in touch with our industrial AI experts.


Talk to an AI Expert