Abstract: In order to improve the transmission performance of wireless power transfer (WPT) system, it is necessary to obtain key information such as load and coupling coefficient in real time during the control process. Currently, acquiring this information typically involves using wireless communication modules or adding additional communication coils, which adds complexity to the system. This complexity is further exacerbated in complex aquatic conditions and high-frequency electromagnetic environments, where abnormal communication during the process is highly likely to cause system paralysis. For this purpose, this paper proposes a novel Unscented Kalman filter (UKF) based online identification method for the critical parameters of the WPT system's transients and loads. This method only samples the instantaneous value of the voltage at the edge of origin, and can acquire the information of critical parameters such as transients and loads in real time. At the same time, to improve the discrimination accuracy and convergence speed, the offline neural network guided particle swarm optimization (PSO-NN) algorithm is used to establish a systematic noise covariance matrix. The experimental results show that the proposed online discrimination method achieves the maximum identification errors of 6.19% and 1.7% for load and mutual inductance, respectively, in the case of varying load, varying phase control angle and offset. The dynamic identification of load can be completed in about 2 ms, which has some engineering applications.