A robust experimental-based artificial neural network approach for photovoltaic maximum power point identification considering electrical, thermal and meteorological impact
Publisher version (Check access options)
Check access options
MetadataShow full item record
This paper aims to develop a robust and practical photovoltaic (PV) Maximum Power Point (MPP) identification tool developed using reliable experimental data sets. The correlations between the voltage and the current (Vmp and Imp) at maximum power from one side, and the irradiance information, electrical parameters, thermal parameters and weather parameters from another side, are investigated and compared. A comparative study between a number of input scenarios is conducted to minimize the MPP estimation error. Four scenarios based on a combination of various PV parameters using various Artificial Neural Network (ANN)-based MPP identifiers are presented, evaluated using the most common regression measure (Mean Squared Error (MSE)), improved in terms of the accuracy of the identification of MPP, and then compared. The first scenario is divided into two parts I(a) and I(b) and considers the irradiance information in addition to the highest correlated parameters with Imp and Vmp, which are circuit current (Isc) and open-circuit voltage (Voc), respectively. The second scenario considers irradiance information and the electrical parameters only. The irradiance information, in addition to the electrical, thermal, and weather parameters, are considered in the third scenario using a single layer network, while the irradiance information, in addition to the electrical, thermal, and weather parameters, are considered in the fourth scenario using a two-layer ANN network. Although the correlation study shows that the Vmp and Imp have the best correlation with the open-circuit voltage and the short circuit current (scenario I), respectively. Nonetheless, the consideration of irradiance, electrical, thermal, and weather parameters (scenario IV) yielded higher identification accuracy. The results showed a decrease in the MSE of Vmp by 74.3% (from 1.6 V to 0.411 V), and in the MSE of Imp by 95% (from 4.4e−6 A to 2.16e−7 A), respectively. In comparison to the conventional methods, the proposed concept outperforms their performances and dynamic responses. Moreover, it has the potential to eliminate the oscillations around the MPP in cloudy days. The MPP prediction performance is 99.6%, and the dynamic response is 276 ms.