Dynamic Threshold-Based Anomaly Detection in Photovoltaic Generation Time Series Using Statistical Methods
Michelle Melo Cavalcante, João Lucas de Souza Silva, Tárcio André dos Santos Barros · 2024
The efficient operation of photovoltaic (PV) plants requires continuous monitoring to identify and correct anomalies that may affect the performance and lifespan of the equipment. However, some challenges include defining which data can be collected and useful from the installation to identify anomalies, as well as which models can be applied. Therefore, this paper proposes an approach for anomaly detection in PV generation using dynamic threshold techniques based on descriptive statistics. The methodology involves monthly analysis of AC power characteristic curves and irradiance normalization, applying interquartile ranges and standard deviations to identify anomalies. AC power was chosen as it is the inverter output and, therefore, more easily obtained data. The methodology’s implementation is validated using real PV inverter data to identify anomalous behaviors. Additionally, the data used are generally available in PV plants without the need for additional sensors. Therefore, this approach provides an effective tool for predictive maintenance and optimization of PV systems.