Abstract:
To deal with the computational complexity problem of support vector data description (SVDD) during training large data set,a real -time SVDD algorithm based on samples-reduced is presented.Firstly,part of samples is sampled from the original samples by random sampling method to train SVDD.Secondly,the cluster center in feature space is estimated by the support vectors getting from the trained SVDD.Finally,the distances from all the original samples to the center are computed and the distances are ranked by descending order.The samples scale is reduced by extract proper proportion samples of longer distances for training SVDD.Experimental results show that the proposed algorithm cuts the complexity down and fulfills the real -time demand of SVDD fault detection.