Assuming that the number of PV installations in Austria will further increase, a well performed area forecast of PV electricity generation will gain in importance especially for the grid operating companies. To create the forecast as precise as possible contributes to a successful integration of higher amount of PV into the electricity system. The main focus of this master thesis is to compile the best 3D regional clustering model for forecasting of PV electricity generation based on the geographical position of the PV installations and reflecting the micro-climatological regional specifics, and to evaluate the potential for increasing the accuracy of PV electricity generation forecast in the future. This is applied on the example of the federal state of Styria. Due to its mountainous landscape, defining the clusters by the height above mean sea level, is assumed to be the key-criteria. Particularly, this is connected to various micro-climatic zones and reflects the important input factor characteristics, which influence the PV electricity output, like the intensity of global horizontal irradiance, ambient temperature, frequency of cloud cover, fog and inversion. To be able to evaluate the influencing factors and their interactions, various prediction models are defined and proved. The forecast is based on the regressions analysis and the statistical MARS-model (multivariate adaptive regression splines). The criteria of a normalised Root Mean Square Error (nRMSE) referring to installed capacity (kWp) is used for verification of the accuracy of the prediction models. Finally, the best results are achieved if the region is divided in two (poss. three) clusters according to the position of the PV systems above sea level defined by 500m (poss. by 1.000m in addition) and overall 6 representing weather measuring stations (providing the forecasted values of global irradiance and ambient temperature) are taken into account. The results confirmed that the idea of finer clustering, respecting the geographical and climatological specifics of the 3D location of PV installations, opens the door to an improvement of the forecast accuracy in the future. With an increasing number of clusters, necessarily the number of representative weather measuring stations has to be increased as well. Due to the additional meteorological information, the local variations are reflected more precisely, which leads to more exact forecasts results.