Aiming at the influence of the fast, large range and random change of dynamic load current on the electricity metering, this paper firstly establishes a non-stationary random process modulation model and bimodal modulation model for complex dynamic power signal, and derives model parameters such as quasi-steady term and dynamic term amplitude. Based on the first spline least squares empirical mode decomposition (FS-EMD) method, a parameter extraction method for the amplitude model of quasi-steady and dynamic terms is proposed, the correctness of the method is demonstrated through the decomposition of power signals from electrified railway traction substations and arc furnaces. By mapping the amplitude domain model parameters, the important characteristic parameters in amplitude domain representing complex dynamic electric energy signals are constructed, and four important features are extracted. Finally, the dynamic error tests conducted on electricity meters reveal that the four significant parameters identified herein act as critical factors triggering the meter′s performance to exceed tolerance thresholds.