In this paper an active defense mechanism for detecting cyber-attacks on a grid-tied PV system is discussed. The proposed defense mechanism introduces a unique random signal termed as “watermark” into the control inputs of the grid-tied inverter system. The watermark signal is then shown to propagate via the switching inverter and its components and appear in all of the measurement data employed to control the system. The proposed approach employs the measured data to construct a system identification model, followed by two statistical tests (Test-1 and Test-2) that are designed to accurately flag cyber-attacks. The proposed approach is shown to rapidly detect any malicious manipulation of measurement data such as false data injection, record and replay etc. from malicious actors that would like to gain control of the the grid tied inverter and/or disturb its operation to potentially cause greater damage to the grid. The robustness of the proposed active defense mechanism is tested and validated with several attack scenarios implemented on a 3kW laboratory grid tied PV system
6 September 2023 in IEEE Transactions on Industrial Electronics
Grid-connected inverter systems that are important for increased utilization of renewable energy are vulnerable to cyberattacks. Malicious agents can attack a system by providing false measurements of sensed signals, such as power available/generated, or the grid current to sabotage the inverter control system. To defend against such attacks, we propose a technique of dynamic watermarking that superimposes a small secret signal onto the actuator signal and raises an alarm when the purported measurement data returned from the sensors do not contain appropriate transformations of the secret watermark. In this article, a robust attack detector is introduced, which does not require either knowledge of the variance of the ambient sensor noise or a model of the system. We develop an additional test that is not only impervious to sensor noise variance but can also be used to distinguish malicious attacks from sensor unreliability due to changing levels of sensor noise. We also show that a data-driven system identification of the switching inverter system can be employed in the dynamic watermarking tests. We present both simulation and experimental results from a grid-connected photovoltaic (PV) system, which show that the resulting robust dynamic watermarking method can detect malicious attacks on inverter control systems.
In this article, we address the cyber-security problem of industrial control systems (ICSs) when their sensor measurements may be compromised due to an attacker who has intercepted those measurements via a network. We introduce a general-purpose method “Dynamic Watermarking (DW)” to detect potential cyber-intrusions on speed sensor measurements within industrial control systems, which deploy an adjustable speed drive (ASD) to control a critical process. The DW method is injecting a random private low-amplitude signal with a zero mean Gaussian distribution, “watermark”, into one of the input phase voltages powering the ASD system. The watermark signal propagates through the system including pulse width modulation (PWM) power conversion stage and motor, then ultimately appears in the speed sensor measurements. By deploying two statistical DW tests with two proper thresholds, the system can detect potential cyber-intrusions or unobservable cyber-attacks such as replay attacks and false data injection attacks (FDIA). The DW method tested on a laboratory-scale ASD system experimentally to protect the system against cyber-intrusions. This system, powered by a commercial PWM drive operating at 208 V, 3-phase, and 3.7 kW, served as our experimental platform.