Abstract: Soft magnetic composites, due to the superior high-frequency characteristics, have become the preferred core materials in the design of transformers and motors. To enhance the efficiency and power density of these devices under high-efficiency magnetization, it is imperative to refine the accuracy of iron loss calculations during the product design phase. This paper presents a dynamic hysteresis model based on trapezoidal equivalent circuit combined with neural network, which can be used to calculate the iron loss of high frequency soft magnetic composites. This model computes static hysteresis loss, eddy current loss, and anomalous loss using non-ideal inductance, constant resistance, and nonlinear resistance, respectively. A neural network algorithm, which accurately characterizes the magnetization process, is employed to enhance the simulation accuracy of static hysteresis loops at low magnetic densities. The model also accounts for the skin effect's impact on iron losses when calculating eddy current and anomalous losses using the trapezoidal equivalent circuit. Furthermore, a magnetic characteristic testing system for soft magnetic materials under high-frequency sinusoidal excitation is developed. Experiments measuring hysteresis loops and iron losses of soft magnetic composites are conducted in the 1Hz to 10kHz frequency range. The calculated iron losses are compared with experimental data, confirming the model's accuracy in estimating losses under high-frequency sinusoidal excitation. This research offers a loss calculation method with a straightforward model structure, high precision, and practical applicability for the optimized design of transformers and motors.