AutoPV: Automated photovoltaic forecasts with limited information using an ensemble of pre-trained models

Abstract

Forecasting the power generation of locally distributed PhotoVoltaic plants is vital for the efficient operation of Smart Grids. The automated design of such models for PV plants includes two challenges: First, information about the PV mounting configuration (i.e. tilt and azimuth angles) is often missing. Second, for new PV plants, the amount of historical data available to train a PV model is limited (cold-start problem). Therefore, we aim to address these two problems while reaching an accuracy comparable to methods that require such information and historical data, and propose AutoPV. AutoPV is a weighted ensemble of models that represent different PV mounting configurations. This representation is achieved by pre-training each model on data of a separate PV plant scaled by its peak power rating. To tackle the cold-start problem, we initially weight each model in the ensemble equally. To tackle the problem of missing information about the PV mounting configuration, we use new data that become available during operation to adapt the ensemble weights to minimize the error. AutoPV is advantageous since the unknown PV mounting configuration is implicitly reflected in the ensemble weights, and only the PV plant’s peak power rating is required to re-scale the ensemble’s output. The ensemble approach also allows the representation of mixed-oriented PV plants, as the multiple mounting configurations can be reflected proportionally in the weighting. AutoPV’s automated weight adaptation and cold-start capability are essential for real-world applications to keep pace with the expansion of the PV power generation capacity in future energy systems. It is shown that for a real-world data set, the accuracy of AutoPV is comparable to a non-cold-start model and outperforms (quasi-)cold-start capable models.