Because the power performance of the micro-grid battery will degenerate during the working process, its performance degradation has the characteristics of obvious nonlinear and undulation, and traditional mathematical modeling methods have poor universality, limited prediction under different working conditions, and lack of accuracy, so it is difficult to accurately assess their health status. In order to solve the above problems, the standard BP neural network and the BP neural network based on genetic algorithm optimization are constructed in this paper. The network is trained by the measurable parameters in each discharge process of the micro-grid battery to make the weights and thresholds of the neural network more accurately adjusted. The test set is used to test the neural network which has established. The results show that the BP neural network based on genetic algorithm optimization can effectively improve the accuracy of the evaluation results. The error results are controlled in the range of precision, the maximum error is within 5% and the average error is 2%. It is proved that the BP neural network optimized by genetic algorithm is effective and feasible for accurate estimation of battery SOH under different working conditions.