Safe-Guarding Artificial Intelligence in Power Systems

保障电力系统中的人工智能安全

• 批准号: 458030766
• 负责人: Professor Dr.-Ing. Matthias Althoff
• 金额: --
• 依托单位: Lehrstuhl für Informatik VI: Robotik, Künstliche Intelligenz und Echtzeitsysteme
• 依托单位国家: 德国
• 项目类别: Research Grants
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/458030766?language=en
• 关键词: Safe; Guarding; Artificial; Intelligence; Power

To cope with the challenges of a sustainable energy supply, methods from artificial intelligence are increasing deployed in power systems. Often, artificial intelligence is encoded in the form of neural networks, which are very difficult to analyze and verify. Artificial intelligence will have an increasing role in power systems, whose correct behavior has to be a top priority. However, despite the rise of artificial intelligence in power systems, current solutions do not provide any correctness guarantees. Thus, the main objective of this proposal is to safeguard methods from artificial intelligence in power systems using methods from formal verification.To safeguard artificial intelligence and in particular techniques from machine learning, we propose to build a formally-correct safety layer. Our safety layer will verify just-in-time whether the proposed actions generated from an artificial intelligence fulfill the system specification. If a proposed action is not safe, the safety layer will execute a fail-safe controller that brings the power system into a safe operating condition. Our proposed architecture has the big advantage that artificial intelligence techniques do not have to be certified---only the safety layer requires certification. Even when changing the artificial intelligence module, all its decisions stay safe due to our proposed safety layer. Our safety layer is composed of 1) a module for automatic conformance checking, 2) a module for scalable reachability analysis, 3) a fail-safe controller, and 4) a just-in-time verification module.The outcome of this project will be the first formally-correct safety layer for artificial intelligence in power systems. We will demonstrate the usefulness of our methods within the Center for Combined Smart Energy Systems (CoSES) at the Technical University of Munich. CoSES is not only a research facility, but also an open, interdisciplinary research group covering expertise from several chairs and departments, ranging from energy management to communication technology. In particular, we will investigates microgrids that combine electrical as well as the thermal energy systems.

为了应对可持续能源供应的挑战，人工智能方法越来越多地应用于电力系统。通常，人工智能以神经网络的形式进行编码，这非常难以分析和验证。人工智能将在电力系统中发挥越来越重要的作用，电力系统的正确行为必须成为重中之重。然而，尽管人工智能在电力系统中兴起，但当前的解决方案并不能提供任何正确性保证。因此，该提案的主要目标是使用形式验证的方法来保护电力系统中的人工智能方法。为了保护人工智能，特别是机器学习技术，我们建议建立一个形式正确的安全层。我们的安全层将及时验证人工智能生成的建议操作是否满足系统规范。如果提议的操作不安全，安全层将执行故障安全控制器，使电力系统进入安全运行状态。我们提出的架构有一个很大的优势，那就是人工智能技术不需要经过认证——只有安全层需要认证。即使更改人工智能模块，由于我们提出的安全层，其所有决策仍然安全。我们的安全层由 1) 用于自动一致性检查的模块、2) 用于可扩展可达性分析的模块、3) 故障安全控制器和 4) 即时验证模块组成。该项目的成果将是电力系统中人工智能的第一个形式正确的安全层。我们将在慕尼黑工业大学联合智能能源系统中心 (CoSES) 展示我们的方法的实用性。 CoSES 不仅是一个研究机构，而且还是一个开放的跨学科研究小组，涵盖从能源管理到通信技术等多个主席和部门的专业知识。特别是，我们将研究结合电力和热能系统的微电网。