Fault diagnosis for photovoltaic inverters using a multi-layer neural network

In this paper, an algorithm for diagnosing and predicting solar inverter failures was studied. A multi-layer neural network failure diagnosis model that can diagnose failures using inverter failure data was designed. The data were acquired from the field, and there are 307,200 items such as watt-hour meter reading value, inverter meter reading value, meter failure status, and inverter fault status. And using this data, simulations were performed to optimize parameters, and the size and input/output, activation function, loss function, optimization function, and batch size and number of times of the neural network were determined. The final simulation was performed using the determined parameters and the failure of the inverter was diagnosed. As a result, it was confirmed that the failure of the inverter was predicted with an accuracy of up to 97 [%]. Inverter failure was predicted when the operation was completely stopped, when the error between the amount of power generation and the inverter instruction value increased, and when the efficiency of the inverter changed abruptly.