A methodology for short-term forecasting of solar generation to manage power quality in distribution networks

Abstract

Purpose of research. With the increasing share of generation from solar plants in distribution networks, the requirements for the quality of supplied electricity are growing, as the instability of solar generation directly affects the network load and the voltage parameters of the electricity supplied to the consumer. The purpose of the research is experimental validation of methods for forecasting the output of a PV plant for the next day with hourly planning, which makes it possible to link the magnitude of the forecast error with the electric power quality indicators according to GOST 32144-2013. Methods . To solve this problem, a comparative study of three approaches was carried out: ARIMA/ARIMAX models, the Random Forest ensemble, and the hybrid STL+RF scheme. In all cases, the influence of meteorological factors on the accuracy of solar generation forecasting was analyzed. Results . It was found that the inclusion of weather data significantly improves the forecast quality compared to models using only the history of consumption. The best results were shown by the hybrid STL+RF model with meteorological features, which provided an optimal compromise between forecast accuracy and stability. This assessment shows that in the tasks of network operation management, the use of hybrid models is more justified than the use of statistical models alone. Conclusion . The proposed approach can be used as a basis for early warning systems for the risks of electric power quality deterioration in networks with a high share of RES. The practical value of the work lies in linking the results of solar generation forecasting with the regulatory requirements of GOST 32144-2013 and the tasks of dispatch control.