Electricity price forecasting(EPF) is an essential task in all the deregulated markets of the world. EPF has gained even more importance after the COVID-19 Pandemic and the ensuing energy crisis. Since electricity cannot be stored in an economically efficient manner, the balance between supply and demand parties must be maintained regularly. The balance is tried to be ensured by the day-ahead which is followed by the intra-day and the balancing Market. Accordingly, all large companies and retailers that are electricity consumers, as well as electricity producers, will have a positive effect from making these electricity price forecasts with the least error. By accurate forecasts of electricity prices, the supply companies can plan their maintenance operation according to the low-price days and months as well as the demand-side companies can schedule their operations for the same periods.

Electricity price forecasting is a challenging task due to the nature of electricity prices. The seasonality in various frequencies jumps to both sides and high volatility are the most challenging features of electricity prices. The increasing penetration of renewable energy sources (RES) in today’s power systems makes electricity generation more volatile and the resulting electricity prices harder to predict.

Statistical and machine learning methods are used for EPF. Recently, because of the statistical methods’ limitations, various neural network models are applied to the electricity price forecasting problem. Deep neural networks (DNN), standard recurrent neural networks (RNN), and long-short term memory (LSTM) are well-known methods for EPF. The EPF can be divided into 3 categories where the main difference among these categories is how far in advance we predict. Short-term EPF represents the day-ahead and intra-day markets. In Medium-term EPF, we predict 3-15 days ahead, and long-term predictions are done a few months later.

SmartOpt utilizes its unique algorithms by improving the state-of-the-art methods in the EPF literature. We can produce reliable solutions for the desired market in the desired country. The evaluation of our methods done for the last year in the market by several metrics like MAE, RMSE, sMAPE and rMAE. In addition, we follow the real-time performance of our models and make the development process continuous.

Salih Gündüz, Data Scientist (Ph.D Candidate)