Neuro-fuzzy systems for daily solar irradiance classification and PV efficiency forecasting

Considering the impact of photovoltaic installations and the fact that their performance depends on the type of day, this paper presents a classifier that makes use of fuzzy logic to classify daily irradiance profiles as a human would do. To do this, the system must be linguistically interpretable, so the classifier must be simple enough, but without losing accuracy. This is why the article combines the use of data mining and supervised learning algorithms to obtain an initial system and then exploits simplification techniques such as the concept of fuzzy classifiers with incomplete rule bases, as well as fuzzy tabular simplification of rules to obtain a compact and simple final system. The classifier obtained handles the ambiguity presented by the daily irradiance profiles with precision. Once the system has been obtained, a large number of days in southern Spain are classified, analysing the performances of a photovoltaic plant obtained in each of the classes. Then, a neuro-fuzzy system is designed to predict the performance of the photovoltaic installation, considering the type of day, the maximum ambient temperature reached during the day, and the degradation of the installation over time, proving its usefulness in alerting about anomalous behaviour of the system.

Automated summary

This paper proposes a classifier that uses fuzzy logic to classify daily irradiance profiles. The article combines data mining and supervised learning algorithms to obtain an initial system and then simplifies it using fuzzy classifiers and fuzzy tabular simplification techniques. The obtained classifier effectively handles the ambiguity in daily irradiance profiles. A neuro-fuzzy system is also designed to predict the performance of the photovoltaic installation, considering various factors such as weather type, ambient temperature, and installation degradation.