Abstract: In high-voltage circuit breaker health management technology, existing health status assessment and remaining life prediction methods suffer from issues such as single-source data and failure to account for model uncertainties. By applying principal component analysis to reduce multi-source data dimensions and constructing a health index model using weighted Mardorf distance, we significantly enhance the accuracy of health status evaluation, enabling earlier identification of equipment performance changes. Additionally, we propose an approximate Bayesian inference-based remaining life prediction method that integrates Monte Carlo Dropout and bidirectional long-short-term memory networks to quantify prediction uncertainties, yielding accurate and reliable results. This approach overcomes the limitations of traditional single- data assessments, providing scientific support for predictive maintenance of high-voltage circuit breaker mechanical operating mechanisms, thereby improving power system operational reliability and demonstrating significant engineering application potential.