The dynamic behavior and its optimization of a planar 3-link biped robot under randomly uncertain disturbance were investigated. Based on the biped kinematic model and hybrid dynamic model under ideal situation, the control objective function and HZD (Hybrid zero dynamics) feedback control were firstly established, from which the biped state parameters were optimized with the energy consumption function, and the robotic stable walking under the ideal situation was achieved. Secondly, the biped kinematic model and hybrid dynamic model under the randomly uncertain disturbance were established, from which the robotic dynamic behavior controlled by the HZD feedback control method was investigated under the randomly uncertain disturbance. Finally, taking the randomly uncertain disturbance into account, the disturbance compensator integrated into the HZD feedback control was constructed based on a back propagation neural network, and a novel BPNN-HZD (Back propagation neural network-hybrid zero dynamics) feedback control method was proposed to optimize the biped dynamic behavior. The dynamic simulation results and the stick figures showed that the planar 3-link biped robot would gradually appear divergent dynamic behavior if the randomly uncertain disturbance could not be well compensated, while the BPNN-HZD feedback control can effectively weaken the randomly uncertain disturbance effect on the dynamical response of robot, and the biped divergent dynamic behavior would be avoided.