SMART INFORMATION FLOW TECHNOLOGIES LLC — Department of Energy SBIR Phase I: 12b

Description

The SolarGuard project will address the problem of detecting malicious data injection (MDI) attacks on power transmission systems handling solar power. Such attacks could cause the power transmission system to become unstable, leading to blackouts and cascade failures. MDI can also be used to perfom, high-level application attacks disrupting applications in Energy Control Centers, from Energy Management Systems, such as frequency control, optimal power flow, adaptive protection, remedial action systems (RAS), congestion mitigation, real-time pricing, energy balancing of renewable variability, demand response, storage, etc.MDI attacks could thereby cause major blackouts and brownouts. The SolarGuard Phase I project, will establish the boundaries of the problem by identifying the relevant attack surface, and likely MDI attacks. We will explore attacks that originate from corrupting Internet of Things (IoT) devices of prosumers, attacks on substation software, and attacks that are enabled by physical access to substation and to distribution hardware (which is typically not carefully protected from access). We will also identify the data measurements available to estimate the state of the system and detect MDI. Based on this preliminary study, we will develop detector systems that combine first principles models of the electrical power grid with machine learning models based on empirical data. The former will identify malicious data that violates physical constraints given other measurements, and the latter will help handle model mismatch, and realistic noise models. In Phase I we will conduct a small test evaluation of the method, based on scenarios identified from the Northwest power grid, in order to establish the feasibility of the SolarGuard techniques. Evaluation will be based on measurements of precision and recall, and receiver operating characteristics evaluation of the tradeoffs between precision and recall.