Matlab has many excellent tools for prbabilistic forecasting. A baseline of knowledge is needed, in the area of time seris analysis and probability and statistics. If you have a good data set of past energy use, then you can use it for future predicting. Perhaps a lognormal distribution, with μ and σ growing linearly with time, is useful. The timescale of interest and the data itslf will determine what model is appropriate. The Statistics and Mchine Learning Toolbox and the System Identification Toolbox are likely to be very useful. Use the links above to dive into the documentation.