Abstract
This technical report provides detailed analyses of the accuracy performance of a commercial photovoltaic production forecasting solution implemented for a building in Oslo, Norway. These analyses are based on forecasting and monitoring data collected from January 2025 to June 2025. The evaluation shows that the normalized mean absolute error (NMAE) is 3.5% on average for PV power forecasts and 2% for energy forecasts. Power forecasting errors range from 1.5% in winter to 5% in summer, increasing slightly with the forecasting horizon. Energy forecasting errors benefit from temporal aggregation, which smooths short-term inaccuracies, thus decreasing the summed energy forecasting error with increasing prediction horizon. The forecasting service tends to slightly underestimate production during high-output periods and overestimate during low-output periods. While absolute errors remain moderate, normalized metrics can appear inflated during low-production periods. Four service outages occurred during the evaluation period, resulting in 4% missing data and no automated fault alerts. Future studies are recommended to compare this commercial solution with simple data-driven models using the same inputs to better understand the sources of forecasting error.