The degradation mechanism analysis and prediction of the maritime-based system performance is vital for improving the survivability of the maritime-based systems. However, traditional methods can't deal with uncertainty well since many types of information under uncertainty need to be taken into consideration. As the knowledge base of the evidential reasoning approach, the belief rule base(BRB) can not optimize the modeling complexity and modeling accuracy at the same time. In this study, the structure and parameters of the BRB are jointly optimized to construct a bi-level maritime-based system BRB model. In the bi-level model, the Akaike information criterion(AIC) and mean square error(MSE) are used as the upper-level and lower-level optimization objectives, respectively. By comparing the model outputs with actual outputs, the results of the experiment using the support vector machine(SVM) show that, the optimized BRB can not only reduce the number of rules, but also achieve the highest modeling accuracy, and the effectiveness of the proposed method is verified.