Improved ANN Method Based on Explicit Model for Characterization and Power Prediction of Photovoltaic Module

Abstract

With the continuous increase of solar penetration rate, the volatility and randomness of solar irradiance have brought difficulties to grid management. Therefore, accurate characterization and maximum power point prediction of photovoltaic (PV) modules is important for maximum power point tracking and ensuring grid stability. The single-diode model is that traditional characterization method commonly used but is inconvenient and complex because it employs an implicit equation. This paper proposes a novel model, based on an artificial neural network (ANN) with 4-input and an explicit analytical model, for estimating the maximum power point and current-voltage (I-V) characteristics of PV modules under different operating conditions. Based on irradiance and temperature, the model considers the solar zenith angle and the solar azimuth angle, reducing the loss of accuracy of the solar spectrum and the solar incident angle. The input layer of the neural network has four neurons, which are solar radiation, module temperature, zenith and azimuth. The outputs are the four shape parameters in the explicit analytical model. Once the neural network is trained using historical measured data, it can build a shape model based on the temperature, solar radiation, zenith angle and azimuth angle at a certain moment, and estimate the maximum power point and I-V characteristics without solving any nonlinearity formula equation. Through experimental verification, it can be concluded that the 4-input neural network after increasing the zenith angle and the azimuth angle is more accurate than the 2-input neural network. Finally, the experimental data of four types of photovoltaic modules are used to verify the reliability and accuracy of the proposed model. Since the zenith angle and azimuth angle are considered, the accuracy of the model is improved and the calculation time is reduced.