Abstract: By analyzing distribution law of localization error in underground wireless network localization area based on ZigBee technology, namely localization nodes will offset from outside to the inside of the network at the boundary, so the localization error is great, but the positive and negative directions are fixed, the boundary effect was defined and a optimization method for the boundary effects was put forward. The least-squares method was used to fit quadratic functions of average error in the boundary area, improved BP neural network was used to predict compensation function of the boundary area for network model corresponding to different node spacing. The experimental results show that the compensation function of BP neural network can significantly improve localization accuracy of boundary area, and the localization error of boundary area is one meter or less.