Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Development of a system for maximizing the power production of photovoltaic arrays based on reinforcement learning

Bountoukos Theodoros

Full record

---

Summary

The global demand for energy is rising rapidly and conventional methods of producing energy contribute to global warming. Renewable energy sources are able to solve these problems. Solar energy produced by photovoltaic arrays is considered to be amongst the major renewable energy sources, abundantly available. The photovoltaic arrays however yield very low efficiency under non-uniform incident solar irradiance operating conditions. The subject of this thesis is the development of an electronic system for maximizing the power production of photovoltaic arrays. For that reason, a reinforcement learning-based global Maximum Power Point Tracking algorithm was developed. The PV system developed in this thesis consists of an MPPT control unit, a DC/DC Boost converter and two batteries as load. For the implementation of the MPPT control unit, a Q-learning algorithm, as well as a Particle Swarm Optimization (PSO) algorithm were developed. The Q-learning algorithm under study was used in multiple experiments for alternative shading patterns of the PV array and its performance was compared to that of the PSO algorithm. The experimental results demonstrated that the Q-learning MPPT algorithm converged faster and more accurately to the Global MPP than the PSO MPPT algorithm when an appropriate learning process was applied.

Available Files

Services