An Application of Artificial Intelligence Models for Predicting and
Controlling Solar Cell Output Power
Abstract
The study describes the application and comparison of five artificial
intelligence methods for PV system output power prediction. General
Regression Neural Network (GRNN), Radial Basis Function network (RBF),
Group Method of Data Handling (GMDH) network, Multilayer Perceptron
neural network (MLP) and Linear Regression (LR). Measured values of
temperature (T°C) and irradiance E (Kw/m^2) were used as inputs
(independent variables) and PV output power P (Kw) was used as output
(dependent variable). Predictive performances have been evaluated using
statistical metrics. Comparison of the results provided by the five
models has been conducted and commented. It was observed that predictive
accuracy depend of the nature of data set used and the optimization
parameters of each model. Response surfaces that represent the combined
impact of simultaneous variation in temperature and irradiance on PV
output power have been illustrated. Curves that showed how close were
validation predicted values and actual values have been plotted.
Relationship between output power and the two parameters have been
illustrated and it was found to be nonlinear. Importance of each ambient
parameter contribution to the PV output power has been demonstrated.