Self-updating continual learning classification method based on artificial immune system

Currently, major classification methods belong to batch learning methods, which need to obtain all data once before learning. However, in practice, it is usually difficult to handle all samples at once for samples are obtained gradually at different periods. Unfortunately, research on continual learning study is deficiency and remains to be improved. In this work, a self-updating continual learning classification method (SU-CLCM) is proposed, according to the sophisticated continual learning mechanism of biological immune system. In SU-CLCM, the self-updating cell division strategy overcomes the irrational cell strategies; SU-CLCM takes the advantage of super memory cells and totipotent stem cells with self-updating cell weight so that different types of cell edge are more distinguishable; SU-CLCM can improve the result of high-dimensional data processing. Experiments demonstrated on 20 UCI repository datasets compared with intelligent classification methods and artificial immune methods with continual learning ability to corroborate the highlights of SU-CLCM. Ultimately, a dataset of actual compressor valve fault is employed to verify the effectiveness and superiority of the proposed method.