In this paper, we address input load/generation uncertainty in the stochastic economic dispatch and unit commitment for forecasting and then scenario generation purposes. We propose a constrained multi-variate linear regression formulation to model the joint spatial-temporal distribution of forecasts of loads, and renewable energy productions such as solar cells and wind turbines at various locations. In this model, l 0 and l 1 constraints are added to simulate LASSO and Ridge regression models. In order to generate robust solutions for stochastic economic dispatch and unit commitment, we eliminate heavy-tails from forecasted errors using extreme value theory. We then employ a Monte-Carlo sampling technique to generate scenarios. Finally, we propose a new exact model for scenario reduction. For this paper, we used data published at NYISO and ERCOT websites. Our numerical results show how energy loads and renewable energies can be forecasted using our methodologies.