Abstract: Aiming at the problem that the traditional extreme learning machine neural network cannot get the best classification performance when processing complex data, an improved method of integrated extreme learning machine based on feature selection is proposed. And it is used to calculate to constitute a high-dimensional data set. Then the high-dimensional data set is attribute reduced by the neighborhood rough set algorithm, and different feature subsets are generated by using different neighborhood radii to reduce the data setach feature subset is divided into training set and test set and input to the extreme learning machine for pattern recognition. Finally, the prediction results multiple extreme learning machines are integratedfinal recognition results the relative majority voting method. It is proved thatcomparing with traditional extreme learning machine this method can improve the recognition accuracy of rolling bearing fault categories and make the fault classification results more accurate and effective.

Key words: fuzzy approximate entropy, feature selection, classifier ensemble, extreme learning machine