Conversion of Solar Irradiance to Photovoltaic Power with Hybrid Model Chains

Abstract

Accurate solar resourcing and forecasting depend upon the ability to convert weather forecasts to photovoltaic (PV) power forecasts, which remains challenging till this day. This study explores the fusion of physical model chains and machine learning, to achieve improved irradiance-to-power conversion. The outcomes of some well-tested steps of a model chain are used as input features of machine learning models, so as to form a hybrid model with high precision and wide applicability. Within this framework, a comparative analysis is conducted among three potential machine-learning models, including the long short-term memory (LSTM) network, k-nearest neighbors, and gradient boosting regressor. The results indicate that the physical-LSTM hybrid model exhibits superior performance to other options, reaching a correlation coefficient of 0.997. In cases where specific modeling parameters are unavailable, the hybrid model can mitigate the reliance on PV design parameters while gaining a notable increase in irradiance-to-power conversion accuracy, thereby substantiating a robust underpinning for PV grid connection.