Active vibration control based on neural network online identification for piezoelectric cantilever structure was studied. A online identification device and active vibration control device were designed based on NARMA-L2 neural network model. The neural network layers’ input and output structure was analyzed, and weights and thresholds adjustment formulas were developed. Then experiment of online identification and active vibration control with single-frequency and sweep inference excitation signal were made. Experimental results show some reduction of structural vibration and the control system has good robustness.