Reputation Dynamics in Networks: Application to Cyber Security of Wind Farms

Abstract

Wind power is becoming an increasingly important technology for electric power generation in many power systems around the world. A typical wind farm consists of a relatively large number of turbines under some form of decentralized control. In this context, a cyber-attack to wind power infrastructure may not only cause immediate physical damage to the facility in question, but may also threaten the stability of the power system in which this facility operates. In this paper, we present a novel distributed algorithm aimed at detecting cyber-attacks in a wind farm. The cyber-attacks considered take the form of adversarial manipulation of different turbine-specific control logic parameters. The proposed algorithm relies on the operational dynamics specific to wind farms namely that turbines that are relatively close are subject to similar wind patterns and hence must be in similar control states most of the time. This can be seen as a wisdom of the crowd effect: the true is the of the majority of turbines in physical proximity when only a few have been compromised. In the proposed algorithm, each turbine in the farm periodically shares information regarding their own control state with neighboring turbines through a dedicated wireless channel. We test the performance of the proposed algorithm in a simulation test-bed based on Denmark's Horns Rev wind farm. The results indicate that iterative reputation or trust re-allocation enables a distributed identification of affected turbines under a wide array of cyber-attack scenarios.