Abstract-Introducing a wavelet neural network model (WNNM)at a route maintenance stage to predict the position of nodes in ad hoc networks, a new weighted clustering algorithm (WCA) is presented. Using the nodes' position information from GPS, this algorithm predicts the position of nodes at next time by WNNM, then calculates the time the neighbor nodes spend moving out of the coverage of cluster heads by the predicted values and takes it as a measurement of aggregate holding time. If the cluster structure tends to be unstable, a pre-repair mechanism will function before the link fails, thus avoiding frequent break of links and improving the network performance. Simulation results show that compared to the Lowest-ID WCA and Location-based WCA, the algorithm proposed can increase the packet delivery rate by 9% and 6%, respectively, and decrease the numbers of link break by about 70% and 50%, respectively.