In response to the problems of large sample size, high demand for fault samples, need for fault feature extraction, need for manual setting of fault thresholds, and need to clarify the correlation between faults and fault data in current methods for predicting degradation faults in shipborne radar, a degradation fault prediction method for shipborne radar based on combined intelligence is proposed. Considering the real-time characteristics of shipborne radar data acquisition, a Dynamic Updated-Long Short Term Memory (DU-LSTM) data prediction method is proposed. When using this method to predict peak power, the root-mean-square error is reduced by 0.31541kW compared with the traditional static data prediction method, which effectively improves the data prediction accuracy. On the basis of high-precision data prediction, in order to address the limitation of difficult acquisition of shipborne radar fault samples, the multivariate Gaussian unsupervised model and the DU-LSTM model are combined to detect faults and anomalies in the predicted data. The feasibility and portability of the combined intelligence method are verified by setting different degradation fault prediction experiments. The experimental results show that the proposed method can predict and alarm degradation faults at least 16 time sampling intervals in advance, and this method is superior to the existing degradation fault prediction methods in terms of sample size, fault threshold setting, fault feature extraction and alarm time.