Abstract:Artificial neural networks (BP network, i.e. BackPropagation network) can be used to forecast the Length of Day (LOD) for its advantage of parallel disposal data and strong nonlinear mapping ability, but it can be easily affected by local minimum and resulting in slow convergence speed. Aiming at the shortage of the BP neural network, the wavelet basis function is used instead of the activation function of the BP neural network to improve its weight and threshold, then come into being optimized wavelet neural network which could effectually avoid local minimum and has swift convergence speed. As it is used to forecast LOD, good effects have been achieved.