Aiming at the high wrong alarming rate caused by aero-engine work parameter time-varying feature, use self-adaptive strategy to improve kernel principal component analysis model for anomaly monitoring. The framework of anomaly monitoring based on kernel principal component analysis (KPCA) was presented. At first, use standardized normal data to establish off-line model, then carry on on-line monitoring for real time calculation statics. Because of time-varying feature of monitoring data, use moving window self-adaptive model to improve framework, and adopt real time normal data to renew model and monitoring data. Validate it by aero-engine anomaly monitoring simulation. Results of aero-engine simulation indicate that the proposed method can meet requirements of work parameter change, has higher fault detection rate and lower false alarm rate.