Reinforcement Learning with Variable Fractional Order Approach for MPPT Control of PV Systems for the Real Operating Climatic Condition

Abstract

The designing of maximum power point tracking (MPPT) controller is an integral part of the PV array system to ensure a continuous supply of energy in dynamic environmental conditions. The most challenging part here is to design a model that can track the maximum point irrespective of variations in environmental conditions and its parametric variations. The model designed in this article combats both the challenges as it is based on reinforcement learning with fractional-order. The application of Deep Q-learning makes the model parametric free and once the model trained can be implanted in a different scenario and run effectively. The amalgamation of fractional-order aids in the process by reducing the tracking time, oscillation around the peak, and total harmonic distortions. The model is well tested on standard conditions and has successfully achieved the desired results. Also, the proposed design is compared against various existing comparative algorithms to showcase its effectiveness in tracking time, THD, and maximum power. The design is also tested on the real data set, from the solcast where the test region is New Delhi, the capital of India. This region is taken as it faces one of extreme climatic condition and also being the second-highest most populated state faces an acute shortage of power throughout the year. The results have demonstrated that the model can produce maximum power even in the least solar irradiance conditions.