State of health estimation combining robust deep feature learning with support vector regression

Abstract

Combining Stacked Contractive Auto-Encoders (SCAE) with Support Vector Regression (SVR) method based on mass of data, a novel state of health estimation method is proposed in this paper. With the development of SCAE-SVR, SCAE could learn features automatically for SVR instead of extracting hand-designed features. SCAE is a deep machine learning method of unsupervised statistical algorithm that makes the learned features more robust and efficient. Then Support Vector Regression machine is used to estimate quantitative values dealing with the new feature representations. The composite structure of network not only remedies not enough features abstracted by a simplex shallow machine learning net, but also effectively avoid over-fitting in data regression. State of health estimation for Fuel cell systems from Prognostics and Health Management (PHM) 2014 Data Challenge demonstrates that the proposed method outperforms than other state of health estimation methods based on data-driven.