In order to accurately predict the remaining useful life of bearings,a bearing remaining life prediction method based on feature fusion and hunter-prey optimization (HPO) algorithm optimal relevance vector machine was proposed.Time domain,frequency domain and time-frequency domain features were extracted to accurately describe the degradation state of the bearing,and the extracted features were screened by using comprehensive evaluation indexes to obtain the sensitive feature set;kernel entropy component analysis was used to adaptively fuse the sensitive features to obtain the degradation characteristics of the bearing;the hybrid kernel function was constructed as the kernel function of the relevance vector machine to improve the model prediction performance;finally,the parameters of the hybrid kernel function were obtained by using HPO algorithm,and the parameters obtained from the optimization search were used for the training of the life prediction model.By performing the experiments on the bearing-accelerated degradation dataset,the experimental results show that the proposed life prediction model is superior to BP,ELM and SVM models,the hybrid kernel model is superior to Gaussian kernel model,and the optimization algorithm is superior to particle swarm optimization and genetic algorithm.