Multiexpert Inference Model Based on Belief Rule Base Under Uncertainty for Complex System Performance Evaluation

Performance evaluation of complex system (CS) is critical for ensuring their operational efficiency. Under limited monitoring data, incorporating expert knowledge becomes essential to achieve reliable evaluation results. However, the multiple uncertainties from the interfering monitoring data, subjectivity of expert judgments, and difficulty of quantifying critical performance states pose challenges to developing models integrating data with knowledge. To address these challenges, this article proposes a multiexpert inference model based on belief rule base (BRB-ME) for CS performance evaluation under uncertainty. In the proposed framework, multiple belief rule bases are embedded within a mixture-of-experts architecture to fuse multisource monitoring data, mitigating data uncertainty and avoiding rule explosion. Expert cognitive strength and prior knowledge are encoded into the beta distribution's shape parameters to obtain the reference values, enabling the probabilistic modeling of expert knowledge uncertainty. Moreover, single-proposition outputs are extended to the power set to explicitly represent both partial and complete ignorance, enhancing uncertainty characterization. An alternating iteration optimization strategy is further designed to optimize parameters across all BRB-ME components. Finally, by the case study on rocket structure performance inference, the BRB-ME model achieves an average root mean square error of 0.0065, validating the effectiveness of the BRB-ME evaluation results.