The effect of mixed orientation on the accuracy of a forecast model for building integrated photovoltaic systems

Abstract

BIPVs (Building Integrated Photovoltaics) are expected to dominate the solar industry in urban areas, especially in connection with zero-emission buildings. They allow the use of mixed orientations which result in multiple generation peaks throughout the day that can precisely match the consumption profile of the building. Multiple generation peaks however make the design of an accurate PV (Photovoltaic) output power forecasting tool a complex task. This paper, therefore, aims to quantify the effect of these mixed orientations on the accuracy of such prediction models for a 181.15 kWp BIPV located in Trondheim, Norway. The results show that a forecast model that has a complete perception of all the mixed orientations reduced the RMSE (Root Mean Squared Error) forecast error by 34%. The findings in this work have important implications for developing a practical energy management system that is common in the operation of BIPVs.

Keywords

• Photovoltaic power forecasting
• Building integrated photovoltaics
• Zero emission building
• Mixed orientations
• Deep learning