Abstract:
The output of a photovoltaic (PV) solar panel under varying irradiance and temperature values, is non-linear and poses a challenge when harvesting maximum energy from the panels. The Power Voltage curves from a solar panel exhibit different behavior under changes in ambient conditions. The maximum power point is a point on this curve on which the maximum power values resides. This point changes in some cases, rapidly and in others very steadily, to make sure our system extracts the highest possible power from the panel at different conditions, we must operate/charge at this very value, for which a Maximum Power Point Tracking (MPPT) device is needed. An MPPT optimizes the power being generated from a PV panel under varying environmental conditions by virtue of algorithms controlling the duty cycle or the PWM of the DC-DC converter.
Conventionally hill-climbing and open circuit voltage methods have been deployed for this purpose which provide a cheap and easy way of tracking, however their drawbacks are low accuracy, slow operation and periodic data logging. Artificial Neural Networks (ANN), however, can quickly and accurately estimate the output based on different data sets without falling trap to local maxima.
In this thesis a Solar PV panel is modelled and later used to create data set to train a Neural Network. The trained neural network is then incorporated in an MPPT and subjected to varying environmental conditions to check for its efficiency. Lastly a comparative analysis between three MPPT techniques; Perturb and Observe (P&O), Fractional open circuit voltage (FOCV) (the models of which are also simulated) and the proposed ANN based technique is presented.
