Abstract: Aiming at the problem that the strength limit state of wind turbine blade is difficult to be defined under the mutual influence of each basic random variable,a wind turbine blade strength reliability analysis method combining forward and reverse prediction of generalized regression neural network was proposed. The states of each random variable at the time of blade failure were estimated by the reverse prediction model of the neural network,and then used as reinforcement samples for the training of the forward prediction model after calibrated by the finite element analysis method. The neural network model constructed by the above method was compared with that constructed by more random samples. The results show that the number of learning samples of the former is reduced by 26%,and the mean square error and mean absolute percentage error of the test set are reduced by 48.19% and 58.24%,respectively. Therefore,the neural network model constructed by this method has better prediction performance in the blade failure boundary region. Finally the strength reliability of the blade was calculated using the model,which further verified the effectiveness of the method.