A Periodic Fractional Wiener Process for Remaining Useful Life Prediction of Photovoltaic Systems with Long-Range Dependence

Global warming concerns have spurred rapid growth in the photovoltaic industry, boosting academic interest in photovoltaic system degradation modeling. Although current studies have considered the natural environmental impact on photovoltaic systems, they often focus on single correlations, overlooking multiple correlations and their mutual influences. These multifaceted interactions affect both energy generation and degradation of photovoltaic systems, leading to long-range dependence in their degradation patterns. Moreover, the natural environmental impact on photovoltaic systems exhibits quasi-periodic characteristics driven by seasonal fluctuations. The coexistence of long-range dependence and quasi-periodicity presents significant challenges to the effectiveness of traditional models when applied to photovoltaic performance modeling. This study proposes the periodic fractional Wiener process (PFWP) to capture the long-range dependence and quasi-periodicity simultaneously. Building on this, we develop a PFWP-based degradation model to predict the remaining useful life (RUL) of photovoltaic systems, which is a crucial reliability metric reflecting system performance. Additionally, unknown parameters are predicted by a two-stage maximum likelihood estimation. Five open-source datasets from diverse photovoltaic technologies are adopted to verify that our model offers superior data fitting and RUL prediction accuracy. Particularly, we show that accurate and stable failure time estimations are guaranteed one year in advance.