Abstract: As an important component of the smart grid energy storage system, high-precision state of health estimation of lithium-ion batteries is crucial for ensuring the power quality and supply capacity of the smart grid. To achieve this goal, improved integrated smart algorithms are proposed to estimate the SOH of Lithium-ion batteries. Kernel function parameters are used to simulate the update of particle position and speed, and genetic algorithm is introduced to select, cross and mutate particles. The improved particle swarm optimization is used to optimize the extreme value to improve prediction accuracy and model stability. The cycle data of different specifications are processed to construct the traditional high-dimensional health feature dataset and the low-dimensional fusion feature dataset, and each version of the constructed network is trained and tested separately. The results of the multi-indicator comparison show that the proposed algorithms can track the true value stably and accurately with satisfactory high accuracy and strong robustness, providing guarantees for the efficient and stable operation of the smart grid.