By analyzing influence factors of the satellite solar array output current， a solar array output current prediction method based on artificial bee colony（ABC）-BP neural network is proposed. Sunlight incident angle， solar array working temperature， satellite time are used as the input of neural network to predict the output current. An improved ABC algorithm is used to optimize initial parameters considering neural network’s sensitivity to the initial weights and bias. The trained model can be used for output current analyzing， detecting and alarming abnormality of the solar array. The results show that the trained model can achieve high prediction accuracy. The mean squared error （MSE） is 0.10 A for the same satellite and 0.12 A for different satellites， which are obviously better than those of the traditional data fitting method. By using this model and the proposed alarm method， there is no false alarm for the normal satellite data of 7 years and 5 months， and the abnormal satellite data can be alarmed timely.