The rollout of smart meters and steadily increasing sample rates lead to a growing amount of raw data available for short-term load forecasting (STLF). While the
original motivation for high resolutions has been the enabling of non-intrusive load monitoring (NILM), so far their value for STLF has been limited. We propose a novel approach,
which allows the exploitation of high resolution data for STLF, by incorporating NILM and subsequent clustering of similarly behaving appliances as a preprocessing step.