Prediction of spot prices in Nord Pool's Day-ahead market using machine and deep learning

Aim of this paper is to describe and compare the machine learning and deep learning based forecasting models that predict Spot prices in Nord Pool’s Day-ahead market in Finland with open-source software. The liberalization of electricity markets has launched an interest in forecasting future prices and developing models on how the prices will develop. Due to the improvements in computing capabilities, more and more complex machine learning models and neural networks can be trained faster as well as the growing amount of open data enables to collect of the large and relevant dataset. The dataset consist of multiple different features ranging from weather data to production plans was constructed. Different statistical models generated forecasts from Spot price history and machine learning models were trained on the constructed dataset. The forecasts were compared to a baseline model using three different error metrics. The result was an ensemble of statistical and machine learning models, where the models’ forecasts were combined and given weights by a neural network acting as a metalearner. The results also prove that the model is able to forecast the trend and seasonality of Spot prices but unable to predict sudden price spikes