Aiming at the problem of dynamic error testing of electricity meter, firstly, this paper adopts wavelet analysis to decompose the collected signals of three actual working conditions, such as heating furnace load, extracts feature in the time domain and in the time-frequency domain. Four kinds of test signals, including amplitude variations, low frequency steady state, high frequency steady state and high frequency transient, are summarized from the two aspects of amplitude and frequency, which can comprehensively characterize the characteristics of various dynamic signals. And the thin plate spline interpolation is used to establish seven single test signal models, including the envelope signal, amplitude swell, sag, interruption, harmonic, interharmonic and impulse. The influence of parameters of the single test signal on the error is analyzed on the established dynamic signal testing platform. The results show that the amplitude variation, envelope and interharmonic have great influence on the metering error of electricity meter. Then, the test signal model of composite signal is further established by parameter adjustment of single test signal and compound of multiple signals. Various working modes and conditions of dynamic load can be approximated to carry out error prediction and compensation for the load of laser cutting machine. The results of model prediction error verify the correctness of the composite test signal model, which can provide a basis for the study of dynamic error of electricity meter.