The electricity generation from photovoltaic systems (PV) is becoming more relevant from year to year in the context of global power generation, with the newly installed capacity exceeding all other forms of generation for the first time in 2016. With its triumphal sweep around the globe, it is becoming increasingly cheap, more competitive in more places and thus an increasingly important pillar of our energy system.This is accompanied by increasing demands on the quality of solar energy yield forecasts, as these form the basis for every investment decision. Furthermore, yield forecasts form an important basis for grid operators and actors in the energy market, on which the power generated must be brought into line with the power consumed at all times. At present, major energy yield forecast uncertainties are caused by the input data – time series of global irradiance which are mostly available in one-hour resolution – and by the models which convert these data into the irradiance at the PV module level. To reduce these uncertainties, this work introduces two new algorithms and validates many existing ones with a very large set of measurement data. The validation data-set comprises high-quality measurement data of the Baseline Surface Radiation Network (BSRN), covering a large part of the Earth's climate zones.The first algorithm synthesizes time series of global irradiance of one minute resolution from time series with one hour resolution. Thanks to this algorithm it is possible to simulate PV systems with statistically representative, synthetic input data with a resolution of one minute even at locations where only hourly measured data is available. Compared to existing algorithms the new approach is capable of producing substantially more natural frequency distributions of the global irradiance, of the irradiance gradients and of the clear-sky index. The root mean squared deviation (RMSD) of the global irradiance distribution is reduced by 61%, the RMSD of the gradients by 52% and the RMSD of the clear-sky index by 71%.In addition, a new model for calculating the diffuse fraction of the global irradiance is presented and compared with a selection of existing models. The new approach realizes a reduction by 50% of the deviations of the modelled from measured diffuse irradiation per year, the RMSD is reduced by 18%. In contrast to existing models, the annual deviation of the diffuse irradiation is smaller than 20% in all cases, while it is smaller than 10% in 80% of the analyzed test cases.It is a complex and time consuming task to implement these two algorithms. In order to be usable by fellow researchers, they are publicly available on http://www.pvmodelling.org.A comprehensive matrix simulation analysis forms the third part of the thesis. A wide range of available irradiance models, different simulation time steps and orientations of the PV modules are combined with each other to analyze the effects of the different models on the irradiance on the inclined module surface and finally the PV yield. Thanks to this data, it is possible to evaluate the interaction of the models and the time step for different PV systems with regard to energy yield simulations.The thesis is rounded off by a detailed validation study of models that calculate the global irradiance on tilted surfaces. A validation dataset of long-term irradiance measurements at two locations and 19 different PV module orientations with one minute resolution is used to evaluate the performance of five transposition models. The study helps to answer important questions about the model uncertainties for calculating the irradiance for differently oriented PV module. Recommendations for locations with mostly cloudless or overcast skies are developed as well.