Comparative Analysis of Photovoltaic MPPT P&O Algorithm and Reinforcement Learning Agents Utilizing Fuzzy Logic Reward System

This work utilized a Reinforcement Learning (RL) agent to track for the Maximum Power Point (MPP) of a 200 kW Photovoltaic (PV) power system to supply power to a load of 150 kVA with a 0.8 lagging power factor at 120
$V_{\mathrm{r}ms}$
where the solar irradiance was changing randomly to a gaussian distribution in the range of 0 to 1000 W/m
²
. The implemented RL agent was trained using a Proximal Policy Optimization (PPO) algorithm that used a Fuzzy Logic reward system that was compared with the Perturb & Observe (P&O) algorithm, and another PPO agent trained with a simpler reward system. The implemented RL agent was able to outperform the P&O algorithm and PPO agent utilizing a simple reward system by outputting the highest mean PV power of 132.0 kW and regulated voltage of 117.6
$V_{rms}$
with little to no oscillations in MATLAB Simulink simulated for 10 seconds.