Virtualization of the Evolving Power Grid
- Written by Olufemi A. Omitaomu, Alexandre Sorokine and Varun Chandola
Power grid information processing is becoming increasingly important as we transition to smart grid architectures that collect and analyze massive data. Groups at Oak Ridge National Laboratory have developed tools to help utilities manage the greater flows of information. But the data collection and analysis center itself could become a failure point during major disruptions when situational awareness is most desperately needed. Ultimately, cloud-computing infrastructure will allow for secure virtual information and analysis centers that can distribute results to a large number of analysts, including operational personnel, regardless of their physical locations.
Smart grid technologies collect large amounts of data about power production and distribution, perform real-time analysis and deliver results to both decision makers and consumers. Compared to the legacy electric grid operation system, the smart grid provides new functionalities, such as support for decentralized power generation and storage capabilities, accommodation of plug-in electric vehicle charging stations, the integration of vehicle-to-grid peak shaving strategies and implementation of demand response decisions. The smart grid allows for customer participation in grid operation so that consumers will not only produce data but respond to data as decision makers within the context of demand response programs.
Thus, improved communication and data processing functionalities are essential for successful smart grid operations. Information systems supporting the smart grid are expected to be more reliable and able to accommodate a large number of sophisticated analysis and control algorithms. In addition, customer participation requires utilities to be able to distribute data and results of analyses to a large number of diverse users and third-party service providers.
Currently, power grid data collection and processing use a centralized approach in a client-server model of operation, with just a few servers running data collection, data storage, data access, and analysis and visualization applications. While this model has been found to be sufficient for the legacy electric grid operation system, the model has significant drawbacks in a smart grid system. These can be described in terms of three operational functions of the smart grid: wide-area situational awareness, intelligent smart grid data analysis, and demand response and demand-side management.
During the August 14, 2003 blackout and the days before, in the United States and Canada, information exchange between operation centers was limited to manual, paper-based processes. This practice has changed with the development by the Oak Ridge National Laboratory (ORNL), in Tennessee, of systems such as VERDE ( Visualizing Energy Resources Dynamically on Earth) and EARSS ( Energy Awareness and Resiliency Streaming Service). These toolkits allow utilities, their partners and other stakeholders to share information.
With the massive amounts of data and information expected to flow across utilities with the implementation of smart grid technologies, the data collection and analysis centers at the utilities themselves may become points of failure, resulting in loss of information system network connectivity. Yet, in the context of smart grid, it is now more critically important than ever for utilities to be able to share information for wide-area and real-time system analysis and visualization. Advanced distributed communications tools can provide the security needed for wide-area situational awareness of the grid network.
What utilities have to be able to do is intelligently analyze streaming data from operational sensors, smart meters and phasor measurements. Systems like GridEye, developed at ORNL, have a role to play here. Current operational strategies assume that all data is available at one location and that all data is available for analysis (but not in real-time). While there have been some recent advances in the area of analyzing streaming data from the Internet, most techniques—search query categorization, email spam detection, click-through rate estimation and so on — have little relevance to problems found in the power grid.
What is more, most data-driven event detection solutions are primarily based on time series anomaly detection and change detection methods (the analysis of sensor signatures for known and unknown changes in systems), and either ignore the spatial aspect or cannot scale well to massive data sizes. To promptly detect the kinds of changes that could lead to cascading power system failure, we need to develop advanced analytics for the so-called "big data" sets that cannot be captured, managed and processed using common database management tools. Spatiotemporal event detection algorithms that can analyze data within the allowable time frame of 2000 milliseconds are crucial and necessary. A distributed approach to data communication and infrastructure is equally needed, as transmission of all data to a central location for analyses would be too restrictive.
The increasing importance of demand response and demand-side management to electric utilities in today's environment has been widely discussed. There have also been some ideas about developing tools for customer engagement and energy efficiency. The tool, CoNNECT (Citizen Engagement for Energy Efficient Communities), developed by ORNL, uses geovisual analytics capabilities to help customers understand the patterns in their energy usage, correlate their consumption patterns to weather patterns, compare their consumption to that of their peers in the same geographic areas and monitor trends in their energy consumption, among other capabilities.
Modeling of demand response (DR) with a large number of participants is still a challenge. For smart grid implementations, there is a need to move the industry from "Manual DR" and "Auto DR" to "Optimized and Adaptive DR" by partitioning consumers into clusters for DR operation decisions modeling and alleviating privacy concerns. The use of distributed infrastructure can help achieve this vision.
The smart grid requirements highlighted above are better satisfied by advanced data analytics that implement functionality as a service in the cloud-computing infrastructure rather than using the traditional client-server model. Among the cloud computing models—Infrastructure-as-a-Service (IaaS), Platform-as-a-Service (PaaS), and Software-as-a-Service (SaaS)—the last seems to be most suited for the task of smart grid data processing and analysis. The development of a virtual cloud-based electric grid information and analysis center, using the Software-as-a-Service model, could demonstrate the suitability of cloud architecture for the electric grid.
However, most grid data collection and analysis applications have to undergo a deep redesign of architecture to be able to fully utilize SaaS advantages, such as elasticity, scalability and reliability in case of major disruptions. Given the diversity of the analytical applications, the new system has to be developed in a way that is able to accommodate multiple and diverse modeling and analysis applications. Continued research can play a vital role in the integration of cloud computing infrastructures into the power system.