A method for estimating and monitoring the power generated by a photovoltaic module based on supervised adaptive neural networks

The estimate and the subsequent monitoring of the energy production of a photovoltaic system is a difficult issue because of the many variables involved, such as weather conditions and construction parameters. The mathematical models usually used do not describe, in an optimal way, the actual behavior of the photovoltaic module as they do not consider all the possible variables involved. This approach leads to an estimation error, from which arises the need to improve the mathematical model used. Given the extreme difficulty in identifying and measuring variables other than solar radiation and ambient temperature, we proposed to optimize the mathematical model using the theory of adaptive neural networks. We aim to create a better estimation method that pursues the real behavior of the PV module based on experimental data. We used a SART (Supervised Adaptive Resonance Theory) neural network to correct the power estimates of the one diode model (ODM). For this purpose, we presented an Estimation Model (EM) for estimating and monitoring the maximum power output of a photovoltaic panel that can take into account the non-linear characteristics of the system. We implemented this system via Matlab and evaluated the performance on a significant sample of actual data for a specific type of PV module. The experimental results show that we can improve the estimation and that this method can then be also used in the monitoring process of the PV system in order to identify specific faults. Finally we proposed a scheme of a possible system for estimating and monitoring the output power of a PV module.