Application of Neural Models in Solar Radiation Measurements for a Better Data Gathering Quality

This document analyzes the results obtained from training a neural network with data from a meteorological station in Spain. A regression model has been trained with a multilayer perceptron using ADAM’s algorithm. Initially, a set of historic data was used. The model was then retrained by adding the recently measured data of the meteorological station. This was done to compare the predictions from both training sessions and to check for any variations between the two processes. The comparison shows that the one-to-one relationship between global radiation and the sum of diffuse and direct radiation is maintained. This study demonstrates the usefulness of neural models for detecting some errors that may occur in the meteorological station studied.