Forecasting Electricity Price With KNN and SARIMA Models

Abstract

Over the past few decades, the combination of market deregulation and renewable energy integration into the electricity grid has made the Finnish electricity market more volatile and unpredictable than ever. This situation has raised the demand for accurate electricity price forecasts. Despite some studies on Finnish day-ahead electricity market prices forecasting, there is a notable gap in research regarding forecasting other Finnish electricity markets, especially the reserve markets. A classical SARIMA model and a machine learning KNN model are preliminarily chosen to study the reserve market prices forecasting. An exploratory example of two models on Finnish Automatic Frequency Restoration Reserve Upregulation (aFRR Up) reveals the complexity of forecasting reserve market prices and concludes that such simple models like SARIMA and KNN are not sufficient for the forecasting objectives. In addition, the example also provides some insights for the performance comparison between KNN and SARIMA in time-series forecasting: while SARIMA is more consistent over different time periods, KNN is much faster to train and can sometimes yield much better accuracy.