用差分算法对在线惯序极限学习机的输入权值和偏置进行优化（本程序只限激励函数为sig），其学习时间比极限学习机分批进行学习总共的时间有所提升，诊断精度也提高了。-Algorithm using differential input weights and the bias line Extreme Learning Machine sequencer used to optimize (this program only excitation function for sig), its learning time than the Extreme Learning Machine Learning batches total time has improved diagnostic accuracy also increased.