A Poisoning Attack for Data-Driven Strategies in Power Systems

With the expansion of system scale and data size in power systems, data-driven methods are gradually becoming widely used. However, compared with conventional methods, data-driven methods encounter greater risks in terms of data security and algorithm security. This paper suggests a method of poisoning attacks on data-driven strategies in power systems, which considers the success rate of the attack on each node and the bad data detection method. The proposed attack method can generate backdoors during the training of data-driven algorithms, which can be activated not only by information methods such as injecting poisoned data, but also by physical techniques such as artificially creating short-circuit faults. The effectiveness of the proposed poisoning attack is demonstrated with the experiment on an online transient stability application.