Artificial intelligence based maximum power point tracking

Photovoltaic (PV) power systems stand out as a clean energy in our age. The fact that the efficiency of these power systems is not yet at the desired level has increased the importance of efforts to increase efficiency in this field. In efficiency, these four parameters, namely voltage, current, power coming from the PV panels, and the D value, which is the occupancy-space ratio of the boost converter used for the inverter, are quite determinant. In this thesis study, since the classical P&O algorithm is frequently used in the literature, the P&O algorithm was carried out for 2 seconds on the grid-connected 100 kW PV system model in MATLAB / SIMULINK, and the D value, which is the occupancy-space ratio of 20 different brands of solar panels, was recorded. realized the learning with 0.00626 mean square error in 428 eppoints by using the data set created in (20 x 54609 x 3) dimensions. For LSTM, an architecture with 250 epochs with "ADAM" algorithm was selected as a learning option in an architecture with 2 inputs, single outputs and 80 hidden layers. With this study, it is thought that artificial intelligence-based MPPT systems designed using deep learning in PV systems are more efficient than classical MPPT algorithms and can guide studies in this area.