Decision support in a volatile electricity market: forecasting and cost optimization

Abstract: Given the increase in electricity prices in recent years due to two reasons; the rebound effect after the initial corona outbreak and the Russian invasion of Ukraine, the burden of paying the rising monthly expense for electricity has become an unwelcome reality for a significant part of society. The electricity trades on the open market called Nord Pool for the Nordic countries, among others, where buyers and sellers come together to find a market price for electricity each day. By forecasting future electricity prices using the machine-learning model XGBoost and a select number of features, the use of electricity can be optimized in the near future with respect to cost. Two different XGBoost models were constructed and evaluated on their ability to predict future prices. Each model was trained on a unique dataset, where the datasets are of different characteristics in terms of volatility. The first model, trained on historical electricity prices with less volatility showed a much more reliable forecasting ability than the second model, trained on historical electricity prices with much more inherent volatility. The optimizations were executed in Matlab with two different optimization solvers. The cost-optimization with the forecasted electricity prices is compared to other charging patterns, in order to determine if the forecasted prices are accurate enough to save cost. Each optimization problem had a number of defined objectives and a number of constraints assigned to it. The result of this thesis showed that the charging protocol incorporating the forecasted electricity prices while minimizing the cost produced a more cost-efficient solution in comparison to the other charging protocols brought up.