Abstract: Distributed application failures in cloud data centers have the characteristics of complexity, randomness, which make applications operation and maintenance tasks difficult and inefficient. Therefore, this paper proposed an application failure prediction method for cloud data center, built an application fault prediction model of a cloud data center based on GRU (Gated Recurrent Unit) Recurrent Neural Network (RNN), analyzed and processed the application monitoring data of the cloud data center and predicted the application failures that will occur. The test results showed that the prediction accuracy of this method meets the relevant requirements of failure early detection and treatment in application operation and maintenance management, and has certain practical value in reducing the difficulty of application operation and maintenance management and improving operation and maintenance efficiency.