Study of Learning of Power Grid Defense Strategy in Adversarial Stage Game
Study of Learning of Power Grid Defense Strategy in Adversarial Stage Game
复制标题
DOI：
10.1109/eit.2019.8834202
复制
发表时间：
2019-05
期刊：
影响因子：
--
通讯作者：
S. Paul;Z. Ni
中科院分区：
文献类型：
--
作者：
S. Paul;Z. Ni
输电和配电系统的安全性是目前最具挑战性的问题之一，因为人们越来越担心能源部门的网络攻击增加。在智能电力输配电系统中，网络攻击者能够造成大规模的破坏（包括停电）。为了应对能源领域的这些攻击，不同的基于机器学习的博弈论方法被用来模拟智能电力系统中对手（攻击者和防御者）之间的复杂交互。现有的大多数工作无法通过验证所识别的突发事件的临界性或通过反映攻击对电力系统的影响来复制实时交互。本文采用价值迭代对抗阶段博弈的方法，对输配电系统的临界突发事件进行识别。我们调整的防御策略，从攻击者的学习攻击行动（最终减少发电损失），并提供替代行动的选择，在有限的情况下访问系统。然后，我们分析了学习攻击策略的影响，在一个模拟的电力系统中使用的PowerWorld模拟器在两个案例研究。所有的实验都是在两个标准的电力系统测试用例（W & W 6节点系统和IEEE 39节点系统）上进行的。通过根据攻击者的学习策略调整防御者的策略来验证学习策略的有效性。仿真结果成功地证明了所提出的研究在学习关键突发事件，提供防御策略，并复制攻击对电力系统的影响的效率。
Security of electric power transmission and distribution systems is currently one of the most challenging issues due to rising concerns regarding increased cyber-attacks in the energy sector. In the smart electric power transmission and distribution system, cyber-attackers are capable of causing large-scale damage (including blackout). In response to these attacks in the energy sector, different machine learning based game theory approaches are used to mimic the complex interactions between adversaries (the attacker and defender) in a smart electric power system. Most of the existing works fail to replicate the real-time interactions by verifying the criticality of the identified contingencies or by reflecting the attack impacts on the power system. In this paper, we identify the critical contingencies of an electric power transmission and distribution system adopting an adversarial stage game with value iteration. We adjust the defense strategy from attacker’s learned attack action (eventually reduces the generation loss) and provide alternative action choices in case of limited access to the system. Then, we analyze the impact of the learned attack policies in a simulated power system using the PowerWorld simulator in two case studies. All the experiments are conducted on two standard power system test cases (W & W 6 bus system and IEEE 39 bus system). The effectiveness of the learned policy is verified by adjusting the defender’s policy according to the attacker’s learned policy. The simulation results successfully prove the efficiency of the proposed research in learning critical contingencies, providing defense strategies, and replicating the attack impacts on power systems.