Problem Statement:
A wind farm in Europe was facing high penalties due to a high margin of error in their power output prediction, caused by a 30% error rate in their wind and weather prediction model. They needed an improved model to reduce the error rate and avoid these penalties.
Solution:
To address this issue, the data scientists decided to use multiple models in combination with hyper-local weather sensors and other climate data to develop an ensemble of models that would deliver higher accuracy. They used machine learning techniques on the collected data to make multiple base models, which were then combined to create a hybrid model that reduced the variability of the individual models to arrive at a final prediction of wind speed and direction with greater accuracy. Historical power output data from the farm was then used to predict the power output, and the machine learning continuously learned from past data to assign appropriate weight to each model, improving the prediction accuracy each time.
The techniques, technologies, and tools used in the Power-Gen Forecasting Engine solution include ARIMA, Neural Networks, Time Series Forecasting, and Deep Learning. Multiple base models were created using machine learning techniques, and these models were combined to create a hybrid model to reduce variability and arrive at a final prediction of wind speed and direction with greater accuracy. The machine learning continuously learns from past data and assigns appropriate weight to each model, improving the prediction accuracy each time.
Benefits:
The forecasting engine developed through this approach was able to predict power generation with greater accuracy, bringing down the error rate from 30% to 16%. This resulted in a 70% reduction in potential penalties, saving the wind farm significant costs. The improved accuracy also helped to ensure a stable supply of power to the electric grid, benefitting the wider community.
