The short-term load forecasting study is characterized by an estimative of the consumption pattern ranging from a day to a few months ahead, related to the operation planning. In Smart grid concepts, this study also is important, but the quality of actual load forecasting methods must be improved, due new aspects like DG and Intermittent Loads. The objective of this work is develop a model to multi region short-term load forecasting, to next day, in hour basis, and for next 7 days. Will be considered the effects of the weighted average application on weather variables, to correct the large extension of Multi Regions, and a new approach of Power Demand Patterns Recognition, as an input variable of forecasting model. Will be proposed a model based on Artificial Neural Networks. Correlation results will be presented between the input variables, and a MAPE comparison to load forecasting.