Optimisation of Run of River Production Forecasting Using Aiolos Forecast Studio

Abstract: Hydropower is simply the largest source of renewable energy in the Nordic countries, where It compose around 90% of power production in Iceland, 70% in Norway, 40% in Sweden and 20% in Finland. Mountainous terrain and abundance in the surface water is a significant contributing factor in hydropower production. Hydropower is also considered more reliable and continuous than other forms of renewable energy such as solar and wind. In deregulated power markets where large magnitudes of power are bid for and traded continuously, smart trading and accurate power production forecasts give the advantage to the companies participating in the bids. Aiolos Forecast Studio (AFS) is computer software that provides numerous services in power production forecasting; including hydropower production, which is forecasted by the Achelous model. Several power companies currently use this computer software, one of which is Fortum where it forecasts the power production of Småkraft’s hydropower stations. This Report explains how AFS forecasts hydropower as well as the shortcomings and possible improvements; also, it introduces the calibration procedure to the models to achiever better forecast accuracy. The Report starts with a comprehensive understanding of the model, the Nordic power market and the Hydropower stations taken as a case study; leading to the development of a recalibration procedure based on the understanding of the Achelous model and the nature of the watersheds. Seven hydropower stations were used for developing a calibration procedure, where two were used for calibration and five for validation. This study involved developing an excel sheet that analysis the accuracy of the forecasts to interpret the model forecasts results; forecasts of power production, runoff, snow cover and precipitation were compared for the year of 2019 with actual data using the excel sheet. By interpreting the statistical analysis results, several shortcomings of the model were highlighted that cause a decrease in the forecast accuracy. Most significant of which is the overestimation of the precipitation forecasts. Overall, the program showed promising results when compared with the previous method used for production forecasting, where it showed higher accuracies for all the seven power plants taken in the case study.