Reinforcement Learning-Based Sensorless Vector Control of High-Gain Boost Inverter for Solar Water Pumps

The integration of an innovative inverter design represents a critical solution with advanced control techniques. It is crucial for the development of next-generation, sustainable, and efficient solar-powered water pumping systems (WPS). The designed inverter eliminates the need for a cascade stage or bulky transformer, effectively regulating the low voltage of the photovoltaic system and the high-voltage demand of induction motors. Implementing a sensor-less technique reduces the overall system cost and complexity, making it feasible for large-scale adoption in remote and rural areas where energy access and water scarcity are critical issues. The DC-link current estimates the motor phase current without the use of a current sensor, and an extended SPWM is employed to generate a PWM signal for the voltage-source inverter (VSI). An extended Kalman filter (EKF) and a generalized integrator-based flux adaptation method were employed to estimate the rotor speed and enhance drive performance accurately. Peak power point tracking was achieved using a whale optimization based Perturb and Observe technique, considering the effects of partial shading and abrupt irradiance changes. The deep deterministic policy gradient (DDPG) algorithm of reinforcement learning (RL) is utilized in conjunction with field-oriented (FOC) controllers to minimize torque and speed ripples, which cause vibration in the motor, thereby enhancing the efficiency and reliability of motor operation.