Methods and software for anomalies searching in the telemetry data of a solar power plant based on the artificial neuron network – autoencoder

A new method and software tool for identifying anomalies in the operation of solar panels have been developed based on an artificial neural network of the autoencoder type, trained using solar power plant telemetry data. The method is based on statistical studies of deviations measured from the values of current and voltage of all solar panels of the power plant restored by the neural network. A criterion for assessing the presence of a malfunction in the operation of a solar panel based on statistical studies is introduced. Using the developed methodology and software for searching for anomalies in telemetry data over six months of observations, 14 to 45 anomalies were detected in 33 solar panels under different evaluation criteria. All the cases were analyzed for the causes of anomalies in the operation of solar panels. It has been established that the use of four standard deviations for average daily measured values of current $\Delta I$ and voltage $\Delta U$ as anomaly detection criterion in the analysis of the results of the artificial neural network operation makes it possible to detect faulty solar panels. And the use of three and two standard deviations as anomaly detection criterion can help to detect a decrease in the efficiency of solar panels associated with degradation, excessive shading and other factors.