This paper presents a novel parameter identification method for fractional-order (FO) autonomous Magnetic Coupling Wireless Power Transfer (MCWPT) systems using Back Propagation (BP) neural networks. The method simultaneously identifies three key parameters: load, self-inductance, and mutual inductance. By operating the system at constant apparent power and collecting data at three fractional orders (1.02, 1.03, and 1.04), the proposed approach achieves model-free parameter identification without requiring communication between transmitter and receiver. Using 540 Simulink-sampled data samples for training, the method achieves high accuracy: maximum relative errors of 4.1% for load, 1.5% for self-inductance, and 4% for mutual inductance, with self-inductance averaging <0.5% relative error. This offers a practical real-time solution for WPT system parameter identification.