Short-term transmission system losses forecast based on supervised machine learning

Although active power losses in transmission networks are not significant in percentage, especially compared to the distribution networks, they constitute a major expense for the system operators. Predicting these losses and procuring them in a most feasible way becomes of out-most importance. The paper discusses the importance of short-term active power losses forecasting of different scales and proposes a model based on supervised machine learning to tackle the issue. Support vector regression method with weather forecasts as input data is validated on Croatian Transmission System Operators (HOPS) data, showing significant improvements as compared to business-as-usual approach. The developed model is integrated into a software tool and deployed at HOPS.