Solar Irradiance Nowcasting using IoT with LSTM-RNN

Photovoltaic (PV) power stations are dependent on weather factors that influence their yield. Among the array of meteorological sensors required to determine power production is irradiance sensing instrumentation such as thermopile pyranometers and PV reference cells. Given the relatively high cost of these sensors some alternatives are entailed. This paper follows the development of a low cost solar irradiance sensor or a PV reference cell, using ubiquitous hardware – a Raspberry Pi Zero W, an INA219 current sensor, and a 300 milliwatt [mW] solar cell, calibrated with the help of a pyranometer. Deep learning techniques are used to reconstruct the GHI from values read by the current sensor. The univariate time series values read from the reference cell's current sensor are used as input to encode information into the long short-term memory recurrent neural networks (LSTM-RNN), with univariate time series values of GHI from the pyranometer as output. A signal translator is obtained with the role of predicting univariate time series of GHI that can be later used in PV applications.