Abstract: With the development of coal-fired power transformation, there is an increasing need for large-scale scheduling and other macro decision-making in the power grid. Previous reliability prediction methods for single units cannot meet the power grid's demand for reliability information of unit sets. In order to accurately and efficiently predict the reliability of coal-fired power unit sets, a reliability prediction model of unit sets based on Apriori algorithm optimized back propagation (BP) neural network input (Apriori-BP) was proposed. The Apriori algorithm measured the correlation between the influencing factors and the prediction target based on their impact, which could more accurately measure the overall correlation between the composite influencing factors and the prediction target, thereby better optimizing the model input. Apply the Apriori-BP prediction model to the actual power grid, predict the reliability of multiple different sets of units, and compare the prediction errors with the grey relational analysis (GRA)-BP model and principal components analysis (PCA)-BP model. Results indicate that the Apriori-BP model has the highest prediction accuracy, with root mean square error and mean absolute error of 0.021 4 and 0.009 7, and better generalization ability and robustness.