This article is concerned with set-membership global estimation for a networked system under unknown-but-bounded process and measurement noises. First, a group of local set-membership estimators is deployed to obtain the local ellipsoidal estimate of the true system state. Each estimator is capable of communicating with its neighbors within its communication range. Second, a global estimation approach is proposed which generates a trace-maximal ellipsoid within the intersection of all the local estimation sets with an aim to improve the difference of the local estimate at each time instant. Sufficient conditions for providing a global estimate under both complete and incomplete measurement transmissions are derived. Third, as an application, a modified distributed photovoltaic grid-connected generation system is provided to verify the effectiveness of the developed set-membership global estimation approach. Furthermore, an islanding fault detection scheme is derived based on the calculated global ellipsoidal estimate. Finally, simulation verification of the obtained theoretical results on the distributed generation system is presented.