| Abstract [eng] |
After the Baltic countries disconnected from the BRELL ring, balancing the energy system became a relevant and complex task. To balance the energy system, transmission system operators allow network users to provide balancing services. One of the balancing services provided by network users is reducing energy consumption. To fairly compensate network users who have reduced their consumption, it is necessary to determine how much energy they would have consumed if the balancing service had not been provided. This reference point is the baseline energy consumption. To encourage broader participation of network users in energy system balancing, the methodology for determining baseline energy consumption must be accurate, fair, simple, and applicable. The main types of methodologies are historical data-based methods; machine learning-based methods; meter before and after; analogous user groups; and declarative consumption. This study compares historical data-based methodologies. These methods evaluate data up to approximately two weeks old, divided into working and non-working days. The forecast may be adjusted based on the accuracy of previous forecasts for the calculation day – this is the same-day adjustment. In Lithuania, a historical data-based methodology with same-day adjustment is applied. The mean absolute percentage error (MAPE) of this method across yearly data is 4.31 %. This methodology is fair—the forecasted energy consumption over the year is only 0,17 % higher than the actual consumption. The forecasts are least accurate in March, April, and from October to December. A significant contributor to forecast inaccuracies are prosumers—excluding them improves the monthly MAPE by up to 1,01 %. The methodology is less accurate when evaluating a smaller number of households. Forecast accuracy decreases when assessing the lowest energy-consuming households. In the United Kingdom, a historical data-based methodology without same-day adjustment is used. The MAPE of this method is significantly higher than the Lithuanian one—7,01 %. According to the UK methodology, forecasts for non-working days are calculated without considering the days with the highest and lowest consumption. This change has no significant impact on the method’s accuracy. Considering day-to-day weather changes it becomes clear that energy consumption of recent days is likely more similar than consumption two weeks prior. Acknowledging this a weighted average methodology was developed. According to this method, the influence of each past day decreases by 60 % the further back it is from the day being calculated. This adjustment unequivocally improves the results of the methodology used in Lithuania. The MAPE across yearly data with this improved method is 3,85 %. |
