Fast online transfer learning for photovoltaic power prediction

Abstract

Since photovoltaic (PV) power generation is affected by various factors such as weather conditions and solar radiation, it is difficult to establish an accurate model by using the data directly from newly established PV power station. In response to this challenge, an algorithm named Fast Online Transfer Regression Learning (FR-OTL) for PV power prediction is proposed in this paper. FR-OTL utilizes the concept of online transfer learning, which is able to use the knowledge from historical prediction models and mitigates the effects of insufficient amount of samples, high correlation of data features, high data redundancy, etc. Benefiting from an online training approach, the models trained by FR-OTL are more flexible and can be quickly applied to real industrial environments. Through algorithm comparison and ablation experiments, it has been verified that FR-OTL reduces model training cost while maintaining the accuracy, which is more efficient than other comparable algorithms.