The impact of inhabitant traits and house types on the electricity load forecasting of private households

The propagation of digital infrastructure and intelligence within the electricity grids opens the door to new methods of residential electricity load forecasting: away from the usage of aggregated and standardized load profiles to an individual load prediction at the single-household level. Given the highly individual nature of residential load data, the performance of state-of-the-art forecasting techniques, such as the employment of artificial neural networks (ANNs), is expected to differ among certain household types. In this work, we study the impact of different household types, characterized by inhabitant traits and household composition, on the predictive power of a multilayer perceptron (MLP) trained to forecast the household’s electricity load in 15-minute intervals. To investigate the influence of large electric installations such as photovoltaic systems or heatpumps on the MLP’s forecasting performance, we perform the same analysis for different groups of house types. We find the number of household inhabitants to have the largest effect on the predictive power of the studied MLP. Among the investigated categories, only the impact of the household’s average age on the MLP is not significant.