Dr. Rhythm Singh, Indian Institute of Technology Roorkee (IIT Roorkee) delivered an interesting invited lecture titled Explorations in Machine Learning-based Solar PV Forecasting at the Institute of Technical Sciences of SASA).
After an introduction about the research at IIT Roorkee, he explained the importance of the hour-ahead and day-ahead hourly forecasting of photovoltaic power. A brief overview of the typology of forecasting methods was presented, followed by methods for decomposing global solar irradiation and its transposition to the plane of array. The presented methods are very accurate, with a mean squared error of up to 10% in selected regions for hour-ahead forecasting, and up to 20% for day-ahead hourly forecasting. In conclusion, the algorithms presented enable improvements in accuracy within a 15-minute forecast horizon.
The discussion focused on the practical applications of these results for India's electricity generation needs, the openness of software code and data, and forecasting tools that are not in the public domain.
