Abstract: A combined pattern recognition model that comprises a localization method for the domains and space coordinates of acoustic emission sources is presented. First, to reduce the irrelevant domain searching, we decompose the monitoring data of the whole structure health into an array of multiple submonitoring acoustic emission sensors, and formulate a control equation for the rapid identification of the array and subdivision of the acoustic emission source area. Then, a strategy for nonlinear variations and the Lévy random flight step is adopted in the calculation of the control parameters to improve global searching ability and avoid falling into local optimizations in application of the Gray Wolf Optimizer（GWO） algorithm. Wave speed is regarded as a variable and the improved GWO is used to localize those acoustic emission sources with unknown wave speeds in the source sub-domains aforementioned. The results of pencil lead broken test on a concrete slab show that this method has a positioning accuracy higher than that of the Least Square, LS-Geiger, or GWO localization algorithm, and it is effective in detecting damage in hydraulic concrete slab structures.