Abstract: Under the goal of carbon peaking and carbon neutrality, the development of large-scale electric vehicles will have a great impact on the distribution power system. Accurate prediction of electric vehicle charging load is the basis of regulation. Aiming at problems such as the subjective description of charging behavior and the lack of utilization of real-time measurement data in current load forecasting methods, a charging load online forecasting method based on finite mixture model (FMM) for large-scale electric vehicles is proposed, where FMM is a statistical method that weights and mixes a limited number of random distributions. Starting from the key random distribution that affects charging load, this paper proposes to use FMM to describe the random distribution of electric vehicle initial charging time, establishes a charging load probability model for large-scale electric vehicles. On this basis, a charging load online prediction method driven by measured data is proposed. This method uses a two-stage algorithm to identify load model parameters for different types of days, and updates the load model parameters online in a time-driven manner to achieve charging load forecasting. Finally, taking the measured charging data of a certain charging operator as an example, the charging load model parameters in the operating area were continuously identified. The charging load of electric vehicles in the region was predicted based on the identified model. By analyzing the model parameter identification effect and the charging load prediction results, the effectiveness and feasibility of the proposed prediction method were verified.