Solar Power Forecasting to Solve the Duck Curve Problem

Abstract

Solar power is widely regarded as the power of a green future. However, excessive generation of solar energy can cause damage to the existing power sources, if it’s not balanced properly. Herein comes the concept of the duck curve, a load curve that shows how much power is required to be produced by non-solar sources when solar power generation is in full swing. In order to calculate the duck curve, it is necessary to calculate the solar power generated. There are several methods to achieve this, and this project focuses on the machine learning side of it. Using weather data with parameters like temperature, humidity and wind temperature of a region, the solar output is predicted. This allows easier allocation of power demand to the non-solar sources. As expected, solar power peaks during the daytime and this results in a sharp drop in demand for non-solar sources. This is followed by a steady increase as the sunsets. This load curve essentially shows the impact of solar power on the load demand and gives valuable information on how the other sources have to be adjusted for efficient power generation. Using the dataset obtained from NSRDB, we were able to predict the per hour GHI of VIT Chennai and corroborate with official sources. We were able to plot the duck curve of VIT and then use the model to observe the GHI of other locations in India.