The transition of the conventional energy supply scheme, typically characterized by integration of more small-scale renewable sources and reduction of greenhouse gas emissions, is an inevitable trend. At the same time, efficient operation of the existing distribution system, which hosts these decentralized generators, is equally important toward achieving a sustainable energy system. In order to keep track of the upcoming changes regarding system efficiency, an accurate determination of grid losses in the distribution system is the fundamental step. However, to evaluate the energy losses of power distribution grids at large scales is a difficult task. Currently, distribution system operators only have limited options for determining grid losses, especially in the low voltage sector. The growing installation rate and the fluctuating nature of distributed generations pose further difficulties and uncertainties for the loss evaluations. For this reason, a comprehensive and accurate evaluation on grid losses, which takes the grid information, the available metering data and the installed distributed generations into account, is highly relevant. In this work, a novel evaluation framework for energy losses for low voltage distribution grids is presented. Our evaluation framework consists of three building blocks: a fully automatic and parallelized generation of grid models, the determination of grid losses based on annual power flow simulations as well as a data-driven reference grid modeling method. We demonstrate and validate this framework based on a data-set of 5000 real distribution grids (including about 31,000 low voltage feeders), located in the service area of one of the largest German distribution system operators. The key contribution of the proposed framework is the automatic approach for grid modeling and feature evaluation at large scales and the implementation of the advanced, comprehensive reference grid modeling method for evaluating relevant grid features and estimating grid losses. This novel reference grid modeling approach and the complete framework are further validated by comparison with both exhaustive simulation results of grid losses and three state-of-the-art regression schemes. Among the case studies, this proposed evaluation framework shows a very good performance in the evaluation of grid losses. Moreover, this evaluation framework provides useful insights concerning the allocation of energy losses as well as the impact of distributed generations on grid losses. Accordingly, this novel evaluation framework gives distribution system operators a powerful tool considering the evaluation and improvement of system efficiency.