The problem of short term load forecasting (STLF) for power grids using the dynamic mode decomposition with control (DMDc) is considered. A forecasting model is discovered from time-series data based on the dynamic mode decomposition algorithm in which the effect of climatic factors on electric power consumption is considered. An input selection method is also proposed to provide more informative dataset that efficiently reflects the load pattern changes. The meteorological data are processed through a hierarchical clustering method and is used by the DMDc algorithm as the inputs. The forecasting results with three datasets from Electric Reliability Council of Texas, ISO New England, and Australian Energy Market Operator show the effective performance of the proposed method compared to several other well-known forecasting methods within the literature of STLF such as ARIMAX, SVR, and DMD. Specifically, the average daily load forecasting errors are 4.78%, 7.6%, and 3.94% for the load datasets of three companies which indicates an improvement of 21.64%, 15.55% and 10.45%, respectively, compared to the DMD method without considering the effect of the climatic factors.
