Explainable AI for Solar Energy: A Deep Learning-Based Approach for the Prediction of Key Performance Indicators for 1 MW Solar Plant Under Local Steppe Climate Conditions

Abstract

The paper presents the findings of the maiden study on the long-term seasonalperformance assessment of three PV module technologies—monocrystalline (m-Si),polycrystalline (p-Si), and amorphous silicon (a-Si) housed in a 1 MW solar plantinstalled at a local steppe climate in Gandhinagar, Gujarat, India using the keyperformance indicators (KPIs), daily power generation, final yield (Yf), reference yield(Yr), total energy loss (TEL) and performance ratio (PR). The am-Si PV modules turnout to be the best performing technology with the highest average PR of 71.26% andlowest average TEL of 631 hrs. for the monitored period. Moreover, PV systems havebecome more popular as a source of green, clean energy; yet, they have low base-load energy sustainability. Explainable AI (XAI) aids in understanding predictions madeby the machine learning models. The importance of XAI lies in the comprehension andtrust which is embed into the results of machine learning algorithms. Hence, thepresent work also builds a comprehensive XAI model for KPIs prediction and assessesthe performance of several learning algorithms using R-value, MAE, number ofiterations and execution time. The Levenberg-Marquardt algorithm predicts the KPIs forp-Si, am-Si, and m-Si with 98.63%, 98.58%, and 90.09% prediction accuracy,respectively.

Keywords: Photovoltaic, Polycrystalline, Monocrystalline, Amorphous Silicon, XAI, Final