By Subhonmesh Bose and Dileep Kalathil
Under the auspices of the DIMACS Special Program: Mathematics of Planet Earth (MPE) 2013+’ and the Special Focus on Energy and Algorithms, the workshop on Data-aware Energy Use brought in a diverse group of researchers to University of California, San Diego on Sep 29-Oct 1, 2014.
As the power system evolves into a smart grid, considerable research effort is underway to make use of information technology for making energy consumption more efficient. Large scale AMI deployment by the utility companies, scalable and low cost methods of sensor deployment in the new and old buildings, closer system monitoring via PMUs and newdata available (often at sub-hourly time scales) through these infrastructures offer an exciting opportunity for a dramatic change in the way we analyze and optimize power networks. Rather than adapting the classical approach of treating demands as inflexible and inelastic, a new approach - namely, a “data driven and data aware approach to energy use” - has to be adapted to reap the full benefit. Since such an approach will be significantly different from current practices, there is an urgent need to come up with high-fidelity models, tractable algorithms and validation through pilot deployments that can suitably exploit this (ever growing and possibly real time) data for various applications - e.g., load forecasting, flexibility estimation, resource matching, optimal control, pricing. Powerful techniques from machine learning and data analytics can be used effectively to tackle these emerging challenges as well as the security and privacy issues that come along with it. This effort is multi-pronged and the workshop featured presentations on different facets of this research.
Under the auspices of the DIMACS Special Program: Mathematics of Planet Earth (MPE) 2013+’ and the Special Focus on Energy and Algorithms, the workshop on Data-aware Energy Use brought in a diverse group of researchers to University of California, San Diego on Sep 29-Oct 1, 2014.
As the power system evolves into a smart grid, considerable research effort is underway to make use of information technology for making energy consumption more efficient. Large scale AMI deployment by the utility companies, scalable and low cost methods of sensor deployment in the new and old buildings, closer system monitoring via PMUs and newdata available (often at sub-hourly time scales) through these infrastructures offer an exciting opportunity for a dramatic change in the way we analyze and optimize power networks. Rather than adapting the classical approach of treating demands as inflexible and inelastic, a new approach - namely, a “data driven and data aware approach to energy use” - has to be adapted to reap the full benefit. Since such an approach will be significantly different from current practices, there is an urgent need to come up with high-fidelity models, tractable algorithms and validation through pilot deployments that can suitably exploit this (ever growing and possibly real time) data for various applications - e.g., load forecasting, flexibility estimation, resource matching, optimal control, pricing. Powerful techniques from machine learning and data analytics can be used effectively to tackle these emerging challenges as well as the security and privacy issues that come along with it. This effort is multi-pronged and the workshop featured presentations on different facets of this research.
Ram Rajagopal (Stanford) talked about a new approach based on stochastic optimization and machine learning techniques which use the large scale data available by AMI deployments for modeling consumer behavior and using this model for demand side resource matching. He presented statistics based on real data which shows a substantial gain in the customer enrolling for demand response program when compared to the classical method. Brewster McCracken (Pecan Street) presented some interesting insights gained from the data, like, (i) over 80% discretionary electric use was for air conditioning, (ii) EVs are unlikely to impact utility system reliability even at a fairly high adoption level, (iii) west facing solar panels can generate more electricity during summer peak hours than typical south facing system.
Promoting energy-efficiency in the grid can be thought about at various different scales — from a single appliance level to the bulk transmission system. Although actions at the different scales are interrelated, each level has its own opportunities and challenges. The first step in observing/ controlling appliance behavior inside a home/office requires the knowledge of all available appliances. One approach is to use machine learning and signal processing techniques on the aggregate power profiles and disaggregate them using discernible signatures of different appliances. Mario Berges (CMU) and David Irwin (U Mass, Amherst) presented their work in this area. Balakrishnan Narayanaswamy (UCSD) talked about using similar techniques to control energy consumption and identify faults in HVAC systems.
Abstracting away from the appliance level, next unit to monitor and control is a building. The idea is to develop building analytics and control algorithms to make building-level energy-use more efficient. This was a major area addressed in this workshop. Yuvaraj Agarwal (CMU) talked about using data from the wireless sensors or even from the WiFi access point to estimate occupancy and to reduce the energy consumption in computing systems and HVAC systems. They have developed a new open source architecture called BuildingDepot which is a platform for data storage and management for building related data (and one can develop apps for ‘smart buildings’ over this platform). A. J. Brush (Microsoft Research) talked about another open source platform - Lab of Things (LoT) - developed by MSR, for home sensing, data collection and actuation. She also talked about some very interesting research projects using LoT. David Culler (UC Berkeley) gave an overview of the development of the Building Operating System and Services (BOSS) and using this to extend the energy efficiency techniques to commercial buildings in a large scale. Mary Ann Piette (LBNL) talked about using data management system to exploit the flexible and responsive building loads and Rahul Mangharam (U Penn) talked about the design aspects of real time control for such energy efficient infrastructure.
Mani Srivastava (UCLA) gave a nice presentation about the opportunities and challenges in deploying extensive sensing infrastructure for building energy use, based on the actual deployment data in UCLA and in IIIT-Delhi. One major issue he addressed was the privacy and security challenges associated with such systems. Residential energy-use data is sensitive. It can be potentially used to infer a number of facts about a consumer, like when is a person usually at home, which appliances he/she is more likely to run and so on. Such data can cause security risks and may not be desirable to share. He gave insights into how one can use information-theoretic tools to selectively share some part of the data and obfuscate the rest. A related talk was by Bruno Sinopoli (CMU), where he addressed the security of such sensor networks from the point of view of control of cyber-physical systems. He argued that the traditional design paradigm of control systems which mainly emphasize the stability of the system should be changed to incorporate the security aspect of such systems.
Adam Wierman (Caltech) and Zhenhua Liu (LBNL) brought in a novel direction to the table. They saw data centers as valuable resources for demand response. They argued that with significantly larger demands than individual residential customers and flexible workloads, data centers are ideal candidates for providing demand flexibility and acting as grid-level storage resources. Other than data centers, a few others presented on grid-level technologies such as frequency-controlled demand-response, placement of storage technologies, etc.
The program also brought in many postdocs and graduate students working in various aspects of renewable energy integration. Enrique Mallada (Caltech) talked about challenges in frequency regulation arising from the distributed and uncertain nature of renewable power generation. While Subhonmesh Bose (Cornell) discussed the placement and control energy storage in the grid, Borhan Sanandaji described modeling air conditioners as an energy storage and exploiting the resulting flexibility for demand response. Baosen Zhang’s (Stanford) talk was on analyzing the strategic behavior of renewable energy producers using coalitional game theory. Also, there were many interesting talks on various approaches to solar power prediction.
As Fred Roberts (Rutgers) pointed out, the event took a considerable effort from the organizers, but it was a great success. Many including Rajesh Gupta (UCSD) and others corroborated in organizing similar workshops in the near future. Also, there are exciting follow-up plans, like initiating a website for sharing various energy data available. Overall, it was a learning experience for all us who participated. And what better way to share and gain knowledge than with plenty of great food, abundant sunshine and hospitable hosts in the beautiful Atkinson Hall at UC, San Diego.
Videos of most of the talks are available online now. Also see, Adam Wierman's note about this workshop in his blog.
Promoting energy-efficiency in the grid can be thought about at various different scales — from a single appliance level to the bulk transmission system. Although actions at the different scales are interrelated, each level has its own opportunities and challenges. The first step in observing/ controlling appliance behavior inside a home/office requires the knowledge of all available appliances. One approach is to use machine learning and signal processing techniques on the aggregate power profiles and disaggregate them using discernible signatures of different appliances. Mario Berges (CMU) and David Irwin (U Mass, Amherst) presented their work in this area. Balakrishnan Narayanaswamy (UCSD) talked about using similar techniques to control energy consumption and identify faults in HVAC systems.
Abstracting away from the appliance level, next unit to monitor and control is a building. The idea is to develop building analytics and control algorithms to make building-level energy-use more efficient. This was a major area addressed in this workshop. Yuvaraj Agarwal (CMU) talked about using data from the wireless sensors or even from the WiFi access point to estimate occupancy and to reduce the energy consumption in computing systems and HVAC systems. They have developed a new open source architecture called BuildingDepot which is a platform for data storage and management for building related data (and one can develop apps for ‘smart buildings’ over this platform). A. J. Brush (Microsoft Research) talked about another open source platform - Lab of Things (LoT) - developed by MSR, for home sensing, data collection and actuation. She also talked about some very interesting research projects using LoT. David Culler (UC Berkeley) gave an overview of the development of the Building Operating System and Services (BOSS) and using this to extend the energy efficiency techniques to commercial buildings in a large scale. Mary Ann Piette (LBNL) talked about using data management system to exploit the flexible and responsive building loads and Rahul Mangharam (U Penn) talked about the design aspects of real time control for such energy efficient infrastructure.
Mani Srivastava (UCLA) gave a nice presentation about the opportunities and challenges in deploying extensive sensing infrastructure for building energy use, based on the actual deployment data in UCLA and in IIIT-Delhi. One major issue he addressed was the privacy and security challenges associated with such systems. Residential energy-use data is sensitive. It can be potentially used to infer a number of facts about a consumer, like when is a person usually at home, which appliances he/she is more likely to run and so on. Such data can cause security risks and may not be desirable to share. He gave insights into how one can use information-theoretic tools to selectively share some part of the data and obfuscate the rest. A related talk was by Bruno Sinopoli (CMU), where he addressed the security of such sensor networks from the point of view of control of cyber-physical systems. He argued that the traditional design paradigm of control systems which mainly emphasize the stability of the system should be changed to incorporate the security aspect of such systems.
Adam Wierman (Caltech) and Zhenhua Liu (LBNL) brought in a novel direction to the table. They saw data centers as valuable resources for demand response. They argued that with significantly larger demands than individual residential customers and flexible workloads, data centers are ideal candidates for providing demand flexibility and acting as grid-level storage resources. Other than data centers, a few others presented on grid-level technologies such as frequency-controlled demand-response, placement of storage technologies, etc.
The program also brought in many postdocs and graduate students working in various aspects of renewable energy integration. Enrique Mallada (Caltech) talked about challenges in frequency regulation arising from the distributed and uncertain nature of renewable power generation. While Subhonmesh Bose (Cornell) discussed the placement and control energy storage in the grid, Borhan Sanandaji described modeling air conditioners as an energy storage and exploiting the resulting flexibility for demand response. Baosen Zhang’s (Stanford) talk was on analyzing the strategic behavior of renewable energy producers using coalitional game theory. Also, there were many interesting talks on various approaches to solar power prediction.
As Fred Roberts (Rutgers) pointed out, the event took a considerable effort from the organizers, but it was a great success. Many including Rajesh Gupta (UCSD) and others corroborated in organizing similar workshops in the near future. Also, there are exciting follow-up plans, like initiating a website for sharing various energy data available. Overall, it was a learning experience for all us who participated. And what better way to share and gain knowledge than with plenty of great food, abundant sunshine and hospitable hosts in the beautiful Atkinson Hall at UC, San Diego.
Videos of most of the talks are available online now. Also see, Adam Wierman's note about this workshop in his blog.