I am a third year Computer Science major Ph.D. student from Florida International University in Miami. I am working in Dr. Bogdan Carbunar's Cyber Security and Privacy Research (CaSPR) Lab. My researches mainly focus on data-driven computer system security and data privacy.
Distributed solar energy resources (DSERs) in smart grid are rapidly increasing due to the steep decline in solar module prices. is DSERs penetration has been playing pressure on the utilities to balance electricity’s real-time supply and demand. Recently, there is a rising interest to develop a cost-effective approach virtual power plants (VPPs) that enable solar generated energy trading to mitigate the impacts of the internment distributed DSERs and take advantage of distributed generation from DSERs for more reliable and profitable grid management. However, the existing trading approaches in residential VPPs do not actually allow DSER users to trade their surplus solar energy concurrently to achieve their maximum benefits, and typically require a trusted third-party to play the role of middleman online. In addition, due to a lack of fair trading algorithms, these approaches do not necessarily result in “fair” solar energy saving among all the VPP users for long term.
To address these problems, we design a new solar energy trading system – SolarTrader that enables unsupervised, distributed, and long term fair solar energy trading in residential VPPs. In essence, SolarTrader leverages a new multiple-agent deep reinforcement learning approach that enables Peer-to-Peer solar energy trading among different DSERs to ensure both the DSER users and the VPPs to achieve maximum benefits equally and simultaneously. We implement our SolarTrader and evaluate it using both synthetic and real smart meter data from 4 U.S. residential VPP communities that are comprised of ∼229 residential DSERs in total. Our results show that SolarTrader can reduce the aggregated VPP energy consumption by 83.8% than the non-trading approach. In addition, we show that SolarTrader achieves ∼105% average saving in VPP residents’ monthly electricity cost. We also nd that SolarTrader— enabled VPPs can achieve a Gini Coecient (a standard measure of an approach’s fairness performance) as 0.05, which is the same as the best fairness Round-Robin approach. We will release all the datasets and source code of SolarTrader to the research community.
Smart cities, utilities, third-parties, and government agencies are having pressure on managing stochastic power generation from distributed rooftop solar photovoltaic arrays, such as accurately predicting solar generation capacity and react to the variations in the electric grid. Recently, there is a rising interest in automatically collecting solar installation information that are critical to manage this stochastic solar generation. Given a geospatial region, the information may include the quantity and locations of solar deployments within the region, and also the profiling information for each deployment such as orientation, size, inverter inefficiency, etc. Traditional approaches such as online assessment and utilities interconnection filings are time-consuming and costly, and also limited in geospatial resolution and thus do not scale up to every location. Significant recent work focuses on using aerial imagery to train machine learning or deep learning models to automatically detect solar arrays. Unfortunately, these approaches all require training data that includes Very High Resolution (VHR) images and human handcrafted image templates, which have a minimum cost of \$15 per km^2 and are not always available at every location.
To address the problem, we design a new system—SolarFinder that can automatically detect distributed solar photovoltaic arrays in a given geospatial region without any extra cost. SolarFinder first automatically fetches low or regular resolution satellite images within the region using publicly-available maps APIs. Then, SolarFinder leverages multi-dimensional K-means algorithm to automatically segment solar arrays on rooftop images. Eventually, SolarFinder employs hybrid linear regression approach that integrates support vectors machine (SVMs-RBF) modeling with a deep convolutional neural network (CNN) approach to accurately identify rooftop solar arrays and also learn the detailed installation information for each solar array simultaneously. We evaluate SolarFinder using 41,683 public satellite images that include 180,833 contours from 11 geospatial regions in the U.S. We find that pre-trained (or unsupervised) SolarFinder yields a MCC of 0.17, which is 3 times better than the most recent pre-trained CNN approach and is the same as a supervised CNN approach.
The Internet of Things (IoT) has been erupting the world widely over the decade. Smart homeowners and smart building managers are increasingly deploying IoT devices to monitor and control their environments due to the rapid decline in the price of IoT devices. The network traffic data produced by these IoT devices are collected by Internet Service Providers (ISPs) and telecom providers, and often shared with third-parties to maintain and promote user services. Such network traffic data is considered “anonymous” if it is not associated with identifying device information, e.g., MAC address and DHCP negotiation. Extensive prior work has shown that IoT devices are vulnerable to multiple cyber attacks. However, people do not believe that these attacks can be launched successfully without the knowledge of what IoT devices are deployed in their houses. Our key insight is that the network traffic data is not anonymous: IoT devices have unique network traffic patterns, and they embedded detailed device information. To explore the severity and extent of this privacy threat, we design IoTSpot to identify the IoT devices using their “anonymous” network traffic data. We evaluate IoTSpot on publicly-available network traffic data from 3 homes. We find that IoTSpot is able to identify 19 IoT devices with F1 accuracy of 0.984. More importantly, our approach only requires very limited data for training, as few as 40 minutes. IoTSpot paves the way for operators of smart homes and smart buildings to monitor the functionality, security and privacy threat without requiring any additional devices.
Yuzhou Feng, Qi Li, Dong Chen, Raju Rangaswami. “SolarTrader: Enabling Distributed Solar Energy Trading in Residential Virtual Power Plants”, ACM BuildSys Yokohama, 2020. Acceptance Rate = 24.3%.
Qi Li, Yuzhou Feng, Yuyang Leng, Dong Chen. “SolarFinder: Automatic Detection of Solar Photovoltaic Arrays”, IPSN Sydney, 2020. Acceptance Rate = 21.77%.
Liangdong Deng, Yuzhou Feng, Dong Chen, Naphtali Rishe. “IoTSpot: Identifying the IoT Devices Using their Anonymous Network Traffic Data”, MilCom Norfolk, 2019.
Yuzhou Feng, Liangdong Deng, Dong Chen. “Poster: IoT Devices Discovery and Identification Using Network Traffic Data”, presented in ACM WiSec Miami, 2019.
Doing data science and system research in online service fraud and abuse, mobile authentication, social media validation and cryptocurrencies.
Worked on data science and system research in new energy, IoT fields. Developed SolarTrader, SolarFinder, IoTSpot researching projects, published three conference papers.
Developed BreazeHome.com real estate project. Managed Scrum team as a Scrum master with 6 -10 team members by using JIRA, Mingle.
Maecenas tempus, tellus eget condimentum rhoncus, sem quam semper libero, sit amet adipiscing sem neque sed ipsum.
Implemented a RPG Game Engine by C++ and MFC framework. Developed a commercial mobile game project "The Journey to the West" on Brew mobile system.
