论文信息 - Data driven prognostics using a Kalman filter ensemble of neural network models

Data driven prognostics using a Kalman filter ensemble of neural network models

This paper details the winning method in the IEEE GOLD category of the PHM psila08 Data Challenge. The task was to estimate the remaining useable life left of an unspecified complex system using a purely data driven approach. The method involves the construction of Multi-Layer Perceptron and Radial Basis Function networks for regression. A suitable selection of these networks has been successfully combined in an ensemble using a Kalman filter. The Kalman filter provides a mechanism for fusing multiple neural network model predictions over time. The essential initial stages of pre-processing and data exploration are also discussed.

L. Peel | Leto Peel | L. Peel

[1] Kurt Hornik,et al. Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[2] John W. Sammon,et al. A Nonlinear Mapping for Data Structure Analysis , 1969, IEEE Transactions on Computers.

[3] R. E. Kalman,et al. A New Approach to Linear Filtering and Prediction Problems , 2002 .

[4] A. Krogh,et al. Statistical mechanics of ensemble learning , 1997 .

[5] T. Başar,et al. A New Approach to Linear Filtering and Prediction Problems , 2001 .

[6] Teuvo Kohonen,et al. Self-organized formation of topologically correct feature maps , 2004, Biological Cybernetics.

[7] Michael E. Tipping,et al. Feed-forward neural networks and topographic mappings for exploratory data analysis , 1996, Neural Computing & Applications.