Using Smart Computer Networks to Better Predict Wind Farm Energy Production

What is it about?

This study tackled a major challenge in renewable energy: accurately predicting how much electricity a wind farm will produce 24 hours in advance. The researchers used artificial intelligence, specifically deep learning neural networks, to improve predictions at a wind farm in Tenerife, Canary Islands. Currently, the wind farm uses a simple mathematical formula based on weather forecasts to predict energy output, but this method has error rates of 20-60%, which costs money in penalties and requires backup fossil fuel plants. The researchers trained computer models using three years of data (2014-2016) that included weather conditions like wind speed, direction, humidity, temperature, and actual power generation. The key finding was that a relatively simple neural network with just one hidden layer of about 20 neurons performed better than both the current polynomial method and standard reference predictions. Surprisingly, adding more layers or complexity didn't improve results - sometimes simpler is better. The study found that while deep learning techniques improved prediction accuracy (achieving better results than both the current polynomial model and reference estimators), relatively simple neural networks with just one hidden layer of around 20 neurons were sufficient - truly "deep" networks with many layers didn't provide additional benefits. The researchers concluded that for this specific wind power forecasting problem, complex deep learning architectures aren't necessary, but the tools and techniques from deep learning (like TensorFlow) still provide valuable advantages for developing, training, and deploying these predictive models.

Featured Image

Photo by Raychel Sanner on Unsplash

Photo by Raychel Sanner on Unsplash

Why is it important?

This research addresses a critical need for renewable energy integration. Accurate wind power forecasting is essential for: - Grid stability: Electricity grids need to balance supply and demand in real-time - Economic efficiency: Better predictions reduce costly penalties and minimize reliance on backup fossil fuel plants - Renewable energy adoption: Improved forecasting makes wind power more reliable and attractive to grid operators The study is particularly valuable for isolated electrical systems like islands, where unpredictable renewable sources can't rely on connections to other grids for backup. The researchers demonstrated that modern AI tools can improve renewable energy predictions without requiring overly complex "deep" networks. Their sensitivity analysis also revealed that wind direction forecasts are more critical than wind speed for accurate predictions, providing practical insights for meteorologists and wind farm operators.

Perspectives