This paper addresses the concept of a set-theoretic framework for the detection of bias injection cyber-attacks on the load frequency control loop of a networked power system. The proposed attack detection mechanism is based on the use of convex and compact polyhedral robust invariant sets. An alarm signal is triggered whenever the state vector exits the invariant sets, indicating a potential security breach. The attack scenario studied involves the transmission of corrupted frequency sensor measurements to the automatic generation control unit of a compromised control area. Simulation studies demonstrate the ability of a set-theoretic detector to disclose intermittent attack patterns even in the presence of disturbances.