Short-term load forecasting model based on smart metering data: Daily energy prediction using physically based component model structure

Performance of smart grids and energy markets depends on the accuracy of forecasted power balances and power flows. This document describes the following approach to predict daily energy consumption of large groups of small customers that have electrical heating and cooling. The model is divided into parallel submodels, such as transfer function models, for differently behaving load types. Each linear transfer function has also physically based input nonlinearities such as saturation defining the heating and cooling ranges, or heat pump coefficient of performance. The submodels and their input nonlinearities were identified one after another in decreasing size order. 13 months of hourly metered data from about 6672 houses were used in the model development and verification. The model was identified from 2664 randomly selected houses. The model is described and its simulations are compared with measured loads. Future verification and development steps are briefly discussed.