<p>This paper concerns the predefined-time control for a class of cyber-physical Euler-Lagrange systems under deception attacks. Firstly, a particular high-gain observer is constructed to achieve output feedback control so that the sensors do not need to measure all states. Due to the deception attacks on the sensors, only compromised outputs can be obtained for the observer. The cooperation of a novel speed function and barrier Lyapunov function guarantees the practical predefined-time convergence. Meanwhile, the Nussbaum gain method and fuzzy universal approximation compensate for the effects of attacks and estimation biases. Then, to solve the problem of limited communication resources in the networks, an improved dynamic event-triggered mechanism is introduced into the controller design to save more communication resources and lower the probability of being attacked than static ones. Compared to the existing results, the assumption of deception attack boundary is relaxed, allowing the controller to handle more cases of attacks. Finally, the efficiency of the proposed control approach is validated through robotic arm simulation results.</p>