Data-Driven Diagnosis of PV-Connected Batteries: Analysis of Two Years of Observed Irradiance

Abstract

The diagnosis and prognosis of PV-connected batteries are complicated because cells might never experience controlled conditions during operation as both the charge and discharge duty cycles are sporadic. This work presents the application of a new methodology that enables diagnosis without the need for any maintenance cycle. It uses a 1-dimensional convolutional neural network trained on the output from a clear sky irradiance model and validated on the observed irradiances for 720 days of synthetic battery data generated from pyranometer irradiance observations. The analysis was performed from three angles: the impact of sky conditions, degradation composition, and degradation extent. Our results indicate that for days with over 50% clear sky or with an average irradiance over 650 W/m2, diagnosis with an average RMSE of 1.75% is obtainable independent of the composition of the degradation and of its extent.