<p>To fully describe the relationship between degradation and vibration signals and improve the accuracy of rolling bearing residual life prediction, this paper proposes a scientific machine learning method that mixes Gamma process and neural network. Firstly, in order to solve the instability problem of the parrot optimizer in global search, the exploration stage is added to improve the global search ability of the algorithm while ensuring the randomness of the algorithm, and the ability of the algorithm to jump out of the local optimal solution is enhanced by dynamically adjusting the diversity threshold and jump factor. Secondly, the graded reliability of bearings under live conditions was evaluated based on Gamma degradation model and stress-strength interference model, data were pre-processed based on 3sigma criteria and LR technology, Spearman correlation coefficient and minimum redundancy maximum relevance (mRMR) were used to reduce the dimensions of bearing high Wett recruitment, and a health index (HI) was constructed using the random forest model to calculate the weight of features. Finally, Parrot optimizer for jump strategy improvements (JPO) algorithm was used to optimize the parameters of BiLSTM prediction model, and reliability was added to the model as an input to predict the remaining life of rolling bearings. The experimental results show that the proposed prediction scheme can predict RUL with high precision. </p>