This paper aims to improve the control performance for the air-sea heterogeneous system consisting of multi-unmanned aerial vehicles (UAVs) and an unmanned surface vehicle (USV) by means of solving unknown faults on position transmissions and actuators. Specifically, the transmission fault-resilient and actuator fault-tolerant issues are transformed into distributed observer and zero-sum game problems, respectively. Based on the local velocities and unreliable positions, a distributed observer with adaptive gains is designed for each vehicle to estimate the virtual leader's states in the presence of position transmission faults. As for the actuator faults, the bias fault and efficiency loss are considered. In particular, a zero-sum game is set up for control inputs and bias faults, where the best-response inputs and worst-case bias faults are derived by actor-critic neural networks (NNs). Adaptive parameters are also designed to compensate the unknown efficiency losses. Moreover, identifier NNs are introduced to handle the vehicles' uncertainties. The stability of USV-UAVs closed-loop system is discussed through Lyapunov theory. Finally, two simulation cases and a practical experiment are illustrated to verify the effectiveness, superiorities and potential engineering application values of the proposed control scheme.