Following a research initiative, energy models for the estimation of solar radiation in nine locations in North Carolina and Southern Spain developed by a team of researchers at the University of Cordoba.
Theoretically, measuring solar radiation is expensive, including various tasks related to the maintenance and calibration of most commonly used sensors: radiometers and pyranometers. This is hypothesized due to the scarcity of reliable data.
To validate this, a team of researchers at the University of Cordoba has developed and analyzed several ML models to predict solar radiation at nine locations across a range of various geo-climatic conditions. The research is published in the journal Applied Energy.
Importantly, the model needs only thermal data to predict solar radiations is a distinguished feature. Today, the measure and availability of air temperature data is extensively attributed to affordability of low-cost sensorization and IoT technologies. Meanwhile, most of the current meteorological stations worldwide have rainfall and thermal sensors, but only a handful of them measure solar radiation.
This brings to light that one of the key problems in current models based on AI are the configuration of internal parameters termed hyperparameters. These hyperparameters might be compared to the controls on the mixing board of a sound technician; they require adjustments, as the alteration of potentiometers must be constantly prevented to be a causative of sound problems, explains the lead author of the study.
To present a solution to this problem, the researchers used an automated algorithm called Bayesian Optimization. The function of the algorithm lies for efficient and quick searching for suitable parameters so that efficient and accurate results obtained from the models.