The rapid growth of solar generation technology has become a boon in the energy sector. Smart grids have replaced the conventional Grids due to upcoming various distributed energy sources feeding the grid. The correct estimation of solar intensity according to geographical features will help in determining the capacity of smart Grids. In real-time, most smart grids are compelled to change their renewable energy production process according to the real-time availability of energy resources like wind and solar during the day. Thus, to assuage this problem, the possibility is investigated by using readily available weather data on the NSRDB website to predict solar forecasts 48 hours ahead in the future by using various Machine Learning(ML) algorithms. In this paper, a new day-night model has been designed to limit the uncertainty of solar power generation and reduce the dependability of power grids on non-renewable energy sources like fossil fuels. Further, to improve the solar forecasting prediction, multiple weather observations were taken from preceding time intervals to establish a new data set in linear regression and its subtypes (Ridge and Lasso regression).