Price Forecasting for Day Ahead Energy Markets: A Difference Boosting Approach

As the energy sector is being restructured, the way electrical power is traded is changing across the globe and energy trading are becoming the focal points of the power grids. In developed energy markets, energy trading is pivotal and plays a crucial role in the modern economy, impacting both producers and consumers. Forecasts for the day ahead energy market prices play a critical role for the market participants, who formulate their bidding decision based on the forecasts. Reliable and accurate forecasts translate to better bidding decision and impacts the overall market efficiency. While machine learning based forecasting solutions have resulted in better market price forecasts, these models have an underlying assumption that the system under consideration is stationary, which does not hold for energy market prices since both the prices and its covariates have long term varying trend making them non-stationary systems. In this work, we propose a difference learning based machine learning approach for forecasting day ahead energy market price. Our approach uses gradient boosted trees to learn the deviations in prices from its corresponding covariates. To evaluate the performance of the proposed technique, two distinct markets are selected-the Midcontinent ISO electricity markets, and the Danish market from the DK2 region. We report improved forecasting accuracy with the proposed technique while compared with state of the art forecasting techniques.