On the contribution of pre-trained models to accuracy and utility in modeling distributed energy resources