Modern textile industries require high-speed, energy-efficient, and precisely controlled motor drives to maintain consistent performance under frequent load variations and rapid start-stop operations. Conventional induction motors and permanent magnet synchronous motors, though widely used, are constrained by high costs, thermal stress, and complexity in coordinating multiple units. This paper presents a simulation-based control strategy for switched reluctance motor (SRM) drives tailored for textile applications, focusing on torque ripple minimization and regenerative energy recovery. A 6/4 SRM is modeled using an asymmetrical half-bridge converter to enable independent phase excitation and braking during deceleration. The core innovation lies in a hybrid deep learning architecture that integrates localized Multi-Layer Perceptrons (MLPs) for individual drive control with a Graph Neural Network (GNN) for inter-motor coordination. The MLPs dynamically modulate PWM duty cycles based on rotor position, phase current, and shaft speed, while the GNN captures cross-machine dependencies to optimize collective drive behavior. Simulation results demonstrate a torque ripple reduction of approximately 18.4%, energy efficiency exceeding 85%, acoustic noise below 70 dB, and up to 22% energy recovery during braking. This intelligent architecture eliminates the need for costly sensing hardware or microcontrollers, offering a scalable and cost-effective solution for multi-motor control in textile automation. The approach sets a strong foundation for future hardware implementation and sensorless control techniques in SRM-based systems.