Abstract: In this paper, we propose a method that combines the improved particle swarm optimization algorithm (IPSO) and back-propagation neural network (BP-NN) for improving the quasi-stability of the state-of-charge (SOC) prediction of power batteries. First, the data on current, voltage, power, and charging/discharging efficiency of the battery during long-time charging/discharging are preprocessed using the pandas library, including data cleaning and min-max normalization. Then, two improvements are made to the particle swarm algorithm: one is to propose an adaptive inertia weight, and the other is to consider the optimal positions of all particles instead of a single particle during the update iteration process to enhance the global search capability of the algorithm. Through these improvements, the IPSO algorithm is used to optimize the key parameters of the BP neural network, such as connection weights, bias terms and network structure. Finally, the battery dataset provided by the University of Oxford is used for experimental validation, and the results show that compared with the traditional PSO-BP algorithm, the proposed IPSO-BP algorithm significantly reduces the SOC prediction error from 6.0973 \times 10^ - 7 to 2.8925 \times 10^ - 7, which is up to 52.56\% , and makes the prediction curves smoother and more stable, which effectively enhances the reliability and accuracy of SOC prediction.