Electricity is a commodity that is traded in markets provided by the independent system operators or Energy Exchanges. Its prices in the day-ahead market are generated by an optimal grid solver and in the real-time market is driven by actual demand at that instant. Predicting these prices are useful to market participants to improve their understanding of the impact to their positions to meet their financial objectives. Price predictions also help in developing a strategy to capitalize opportunities and mitigate risks.
Electricity prices are driven by factors that impact supply of electricity, transmission of electricity throughout the electrical grid and factors that impact the consumption of electricity on the demand side. Solving the electricity network is complex and data representing the grid's latest conditions may not be accurate.
In this case study we will attempt to predict electricity prices using a machine learning approach.
We explore how using factors like system load, wind generation capacity, weather factors impact electricity prices. We well automatically identify the most important feature variables as well as the best among a variety of machine learning models (step-wise linear, decision-tree-based, SVM, GRP, ensemble, etc.). We will quantify the model's accuracy and explore strategies to scale up the analysis for a large number of nodes to cloud based infrastructure.
We will show how more variables, engineered features and weather-based variables indeed improve the model's accuracy and learn how to deploy the algorithms to the enterprise scale solutions.