In the coming years, the electrical grid will have to deal with deep changes in the way we consume and produce electricity. These changes result from the democratization of the electric vehicle, the rise of household production of photovoltaic energy, and the implementation of building retrofitting policies to name but a few. To help the grid to face these evolutions, the development of prediction tools for planning studies becomes vital for stakeholders and distribution system operators. It is in that context that the load curve simulator MOSAIC, developed by ENEDIS and Mines Paris, comes into play. This bottom-up approach simulates the current and future electric consumption of a wide range of uses (e.g. electric heating, cooking) for residential and tertiary customers. MOSAIC has been validated in the scope of residential customers during peak and average demand. Today the objective is to improve the accuracy of the simulator by providing a better calibration of the different uses. To this end, we propose in this paper a calibration process based on simplified versions of the models implemented in MOSAIC. Then an optimization approach is conducted to find the best parameters to fit observations. Scores improvements are observed not only with traditional error criteria, but also with new metrics proposed in this study.