Smart Grid Communication Networks

Objective: To accelerate the development of scalable, reliable, secure, and interoperable communications and standards for smart grid applications; and to enable informed decision making by smart grid operators by developing measurement science-based guidelines and tools.

What is the new technical idea?
Traditionally, technology decisions have been dictated by offerings of system vendors, while business decisions are regulated by federal, state, and regional regulatory commissions and organizations. While there are many choices of communications and networking standards, most of these standards were not developed specifically for smart grid applications or technologies. The new technical idea is to work directly with the smart grid stakeholders (utilities, regulators, service providers and consumers) and the telecommunication industry (vendors, SDOs, service providers) to identify communication requirements for smart grid applications, evaluate and develop communication standards, and develop guidelines and recommendations on their use and deployment. Also, the introduction of new power distribution technologies will transform electrical networks to more closely resemble the behaviors of communications networks. This creates an opportunity to apply well-established analysis and optimization techniques from the telecommunications field to aid in the design of future electrical networks.

What is the research plan?
Our research plan is focused on understanding and modeling the power grid user and system behaviors and developing control and communication strategies for achieving the smart grid vision of a more efficient and dynamic electric grid.

In FY 2021, we plan to continue our evaluation of machine learning for smart CPS and IoT systems, such as Industrial Internet of Things (IIoT) systems and the Smart Grid. Our development of tools to analyze machine learning applications for IIoT systems will account for the stringent performance requirements imposed by IIoT applications, such as: connectivity, very low latency, resiliency, security, and accessibility. We will develop system modeling and simulation techniques to support our analysis. We will categorize distributed machine learning algorithms for IIoT systems and develop metrics and models to assess their performance in IIoT applications. We will conduct a literature survey that will include categorizing machine learning algorithms and architectures that can be used to improve the performance of IIoT systems. We will use the output of the literature survey to develop performance metrics and mathematical and simulation models to evaluate distributed machine learning algorithms, particularly federated learning algorithms, for IIoT applications. In addition, we will investigate the use of reinforcement machine learning to identify which IIoT operational parameters are coupled so that they can be adjusted jointly to achieve better system control, and we will study how to improve the machine learning training process by examining the tradeoff between exploration (more optimal, slower convergence) and exploitation (suboptimal, faster convergence) for co-designed IIoT systems. We will start this second task with a literature survey and identify which IIoT parameters are tightly coupled, and we will develop models and simulation tools to examine the exploration/exploitation tradeoff.

Our previous work evaluated machine learning techniques to support the creation of enhanced wireless networks that will enable next-generation CPS, the Internet of Things (IoT), and the Smart Grid, which will incorporate both CPS and IoT architectural elements. In FY 2020, we developed and applied performance measurement tools and metrics to evaluate machine learning architectures/models and algorithms in communication networks for Industrial IoT (I-IoT) systems, including the Smart Grid. We examined how control and networking system co-design can improve I-IoT performance, and we investigated how machine learning techniques can be used to make I-IoT systems more resilient. We also explored how machine learning techniques can be used to detect abnormal system activity, whether due to errors or malicious actors. Our work considered the accuracy of machine learning architectures and algorithms, and we evaluated the computational overhead associated with these algorithms. In FY 2020, we also studied resource allocation schemes that can meet the diversified QoS requirements of machine-type traffic, which is different from traditional Internet traffic, and we developed an understanding of telemetry data for supervised learning algorithms. We used this understanding to design low-cost supervised on-line learning algorithms to support dynamic resource allocation for I-IoT networks.

With the proliferation of Ethernet-based networks and widening PMU installations in massive numbers, future PMU characteristics will require synchronous data delivery with constant latency and minimum jitter. In addition, keeping up with the ever-increasing integration of multiple sub-networks to cope with the massive number of PMU devices, as well as other sensor and actuators, will require the support of a reliable high-speed network such as TT-Ethernet. TT-Ethernet is a deterministic, synchronized, and congestion-free network protocol, which is based on the IEEE 802.3 Ethernet. It is designed to fulfill the requirements of reliable, real-time data delivery in advanced integrated systems.

Having completed our software implementation of the TT-E network for synchrophasor communications, our plan in FY 21 and beyond is to develop 5G wireless communications that can remotely monitor PMU’s and other sensory devices, such as IED. 5G supports different types of services, including enhanced Mobile BroadBand (eMBB), massive Machine Type communications (mMTC), and Ultra-Reliable and Low Latency Communications (URLLC). This requires a wireless technology with high reliability and low delay. In view of the strict low latency requirements of synchrophasor communications, URLLC is the best 5G option for wireless support. This would require designing a wired/wireless gateway where TT-E, as the main bus, can provide wireless support using URLLC. The main challenge is how to design a gateway that can achieve over the air timing and synchronization between the 5G uRLLC and a TT-E switch.

In 2020, we also completed a special Issue on communications and data analysis, which has been published in the IEEE JSAC, June 2020. We are also completing the modal identification project. In this project two algorithms have been developed to identify and monitor low frequency oscillations; namely a fast subspace tracking algorithm and a gradient descent based fast recursive algorithm.

Finally, in FY ‘19 we completed our software implementation of a new two-stage processing approach that consists of a subspace-based estimation method to detect and identify all harmonic components, followed by a low-complexity fast tracking algorithm to monitor frequency variations of voltage and current signals in real-time with great accuracy. The outcome of this investigation has been published in the May, 2019 issue of the IEEE Transactions on Smart Grid. In this investigation we also studied PMU-based modal identification methods (such as recursive Fourier transform) to explore the possibility of expanding the application of our harmonic tracking approach for identifying power damping and oscillation disturbances for wide area monitoring.