关键词: 自适应步长, 布谷鸟搜索算法, 神经网络, 故障诊断

Abstract:  In order to solve the problem of slow convergent speed and poor local search ability of CS(Cuckoo
Search) algorithm, a self-adaptive step is introduced in CS, which balances the global and local search
ability and improves the convergent speed. Experiments have been conducted on functions. The results
indicate the good performance of the new algorithm. Initial weights and threshold values are optimized by the
modified CS. Compared with the conventional BP ( Back Propagation) algorithm, the condition of slow
training speed and premature result is released. The new neural network is used in pump units蒺 fault
diagnosis. The result shows that CS-BP enhances the convergence rate and robust of neural network, and the
accuracy of the fault diagnosis.

Key words: neural network,  self-adaptive step, fault diagnosis, cuckoo search algorithm