In this paper, we present the winning solution to the ECML-PKDD Discovery Challenge 2017 on Multi-Plant Photovoltaic (PV) Energy Forecasting. The goal of the challenge is to utilize the historic data of three different PV plants in Italy regarding meteorological conditions and production in order to forecast their energy production. A major problem is that the data contains many missing value for most sensors and especially for the period of the year for which predictions shall be made in the subsequent year. We investigate two approaches: a regression tree ensemble whose hyperparameters where tuned via Bayesian optimization and a simple rule that just predicts the hourly averages that were observed during the previous year. The latter approach is the winning solution of the challenge.