With the wide application of power electronic devices and the large-scale access of distributed energy sources, the loads of the power system show the characteristics of low power factor, nonlinearity and time-varying nature. These changes bring great challenges to the traditional load modelling methods. Therefore, in this paper, a solution based on time-varying power function model and improved adaptive parameter identification method is proposed for the load modelling problem in low power factor scenarios. Firstly, considering that the composition of real distribution network loads changes with user behaviour, weather conditions and system operation mode, the relationship between the parameters of the load model and the external influencing factors is complex and nonlinear, and should change over time. Therefore, this paper proposes to use a power function model that considers time-varying properties to describe the load and voltage changes over time in distribution networks. Secondly, in order to improve the accuracy and robustness of model parameter identification, this paper proposes a hybrid algorithm combining Particle Swarm Optimization and Variable Forgetting Factor Recursive Least Squares (PSO-FFRLS). The method makes use of the global search capability of PSO to optimize the model parameters initially, avoiding the traditional recursive least squares method from falling into local optimum; meanwhile, it realizes the dynamic tracking of time-varying parameters through the adaptive mechanism of FFRLS. Finally, the PSO-FFRLS algorithm outperforms traditional methods in parameter identification across all error metrics, demonstrating superior tracking performance, noise immunity, and robustness.