Taking the voltage, current, temperature and internal resistance of the power battery as input and the state of charge as output, a neural network simulation model with four inputs and one output was established to predict the state of charge of the battery. Then, based on the state of charge, the health state of the battery is estimated by the improved capacity method, the improved internal resistance method and the voltage method, respectively. The health state of the battery is estimated by the genetic neural network algorithm. By Simulink simulation and experimental study were carried out on four 12 V series lithium ion battery packs. The charge state and healthy state of the battery were tested by collecting voltage, current, temperature, internal resistance and discharge quantity data during charging and discharging. The experimental results show that the prediction accuracy of the battery state is 1.6%. The results of simulation and experiment show that the maximum error of the combined method is 1.5%, which is higher than the other three methods. In this paper, the health state prediction method is proposed, which omits the complex steps of finding health factors by traditional neural network algorithm, and avoids the limitation of single parameter judgment method for battery health state estimation.