Designing an efficient photovoltaic (PV) system is a complex task that involves modelling and simulation of all of its subsystems. The simulation results are usually obtained by performing behavioural simulations, which is useful when simulating at the system level, though it is not possible to implement models of real electronic components due to an absence of adequate libraries. In this manuscript we will focus on a simulation of electronic circuits performing maximum power point tracking control (MPPT), which is one of the standalone PV subsystems, using circuit level model. Here we will introduce some modifications to perturb and observe method in order to facilitate circuit level simulation, and present a SPICE circuit implementing that MPPT algorithm. The importance of the contribution comes from the fact that the SPICE implementation of the algorithm is circuit (technology) independent.
Starting with the experience that the output voltage and the output current of a photovoltaic panel are not pure direct current constants due to the inevitable connection to a converter (or inverter) that is working as a switching system, we came to the conclusion that interest exists for the behavior of the solar cell at the frequencies of the harmonics of the converter’s switching frequency, which is subject to change according to the maximum power-point tracking. In other words, a need exists for frequency domain characterization of the solar cell, for which a linear small-signal model is necessary. To enable simulation for small signals, development of a linear reactive model was considered. Since a one-diode large-signal model already exists, it was used as a basis for the extraction of the parameters of the small-signal model. The new model was represented in the form of a parallel RC two-terminal circuit, the R and C being functions of the photocurrent (acting as a map of the illumination) and the diode voltage. Since the R and C of the model are quiescent-point dependent, their values as a function of the illumination and the diode voltages were approximated by artificial neural networks (ANNs). Separate ANNs were created for modeling R and C. To verify the model, two small-signal simulations were performed. The first one was done with the existing nonlinear model, while the second was done with the new linear model (running the ANNs). Excellent agreement was obtained.