Characterisation of hourly temperature of a thin-film module from weather conditions by artificial intelligence techniques

Abstract

The aim of this paper is the use and validation of artificial intelligence techniques to predict the temperature of a thin-film module based on tandem CdS/CdTe technology. The cell temperature of a module is usually tens of degrees above the air temperature, so that the greater the intensity of the received radiation, the greater the difference between these two temperature values. In practice, directly measuring the cell temperature is very complicated, since cells are encapsulated between insulation materials that do not allow direct access. In the literature there are several equations to obtain the cell temperature from the external conditions. However, these models use some coefficients which do not appear in the specification sheets and must be estimated experimentally. In this work, a support vector machine and a multilayer perceptron are proposed as alternative models to predict the cell temperature of a module. These methods allow us to achieve an automatic way to learn only from the underlying information extracted from the measured data, without proposing any previous equation. These proposed methods were validated through an experimental campaign of measurements. From the obtained results, it can be concluded that the proposed models can predict the cell temperature of a module with an error less than 1.5 °C.