Remaining Useful Life Prediction of High-Dimensional Kernel Density Estimation With Adaptive Relative Density Window Width Considering Multisource Information Fusion

For the remaining useful life (RUL) prediction of complex systems, some data in a large amount of multisensor monitoring data do not effectively characterize the degradation of complex systems, while there is redundancy between sensor data, which leads to low accuracy of prediction results. Therefore, a high-dimensional kernel density estimation (KDE) RUL prediction method was proposed with an adaptive relative density window width based on multisource information fusion, which contains a multiindicator sensor evaluation algorithm based on the entropy weight method and the maximum relevance–minimum redundancy (mRMR) sensor selection algorithm. First, the trendability, monotonicity, predictability, and robustness are proposed to evaluate the sensor data based on the mapping relationship between the sensor data and random degradation characteristics of the system. Moreover, a comprehensive evaluation indicator is constructed using the entropy weighting method to select sensors with higher comprehensive scores, which can better characterize the system degradation. Furthermore, an mRMR based on mutual information (MI) is proposed to select the sensor group that has the maximum relevance to the runtime and minimum redundancy of information between multisensor data. Moreover, a high-dimensional KDE RUL prediction model with adaptive relative density window width is established based on feature-level information fusion. Finally, the accuracy and effectiveness of the proposed method are verified by C-MAPSS data and N-MAPSS.