Technical Sessions

Session A1

Data Analytics for Demand Management and Grid Operations

Conference
3:10 PM — 4:00 PM UTC
Local
Nov 11 Wed, 10:10 AM — 11:00 AM EST

Accurate Disaggregation of Chiller Plant Loads Using Noisy Magnetic Field Measurements

ChongAih Hau (National University of Singapore, Singapore); Binbin Chen (Singapore University of Technology and Design, Singapore); Ziling Zhou and William G Temple (Ampotech, Singapore)

0
Chiller plants consume a significant amount of energy around the world. While there have been well established systems for monitoring their performance, those existing systems are expensive and difficult to install. In this work, we propose a low-cost and non-intrusive approach to achieve accurate monitoring of individual loads in a chiller plant. Our proposed system complements the in-situ aggregated power measurement with additional magnetic field measurements near individual branches using non-intrusive Anisotropic Magnetoresistive (AMR) sensors. AMR sensor measurements help disaggregate smaller pump loads in the system. However, they are subject to various noise from the environment, which can significantly lower the estimation accuracy. To overcome this, we design a novel hybrid solution that combines the AMR-based method with a multi-layer neural network (MLNN)-based method. Specifically, we use filtered outputs (under some quality check) from the AMR-based method to help train the MLNN, and then we combine the outputs from both the AMR-based method and the MLNN-based method to achieve better estimation results. We tested our proposed solution using real world power consumption traces collected from the chiller plant of a six-story commercial building, and under varying level of environmental noise. Our approach can robustly achieve low mean-absolute error (1% for the chiller component and 3% for individual water pump loads).

Demand-Side Scheduling Based on Multi-Agent Deep Actor-Critic Learning for Smart Grids

Joash Lee, Dusit Niyato and Wenbo Wang (Nanyang Technological University, Singapore)

3
We consider the problem of demand-side energy management, where each household is equipped with a smart meter that is able to schedule home appliances online. The goal is to minimise the overall cost under a real-time pricing scheme. While previous works have introduced centralised approaches in which the scheduling algorithm has full observability, we propose the formulation of a smart grid environment as a Markov game. Each household is a decentralised agent with partial observability, which allows scalability and privacy-preservation in a realistic setting. The grid operator produces a price signal that varies with the energy demand. We propose an extension to a multi-agent, deep actor-critic algorithm to address partial observability and the perceived non-stationarity of the environment from the agent's viewpoint. This algorithm learns a centralised critic that coordinates training of decentralised agents. Our approach thus uses centralised learning but decentralised execution. Simulation results show that our online deep reinforcement learning method can reduce both the peak-to-average ratio of total energy consumed and the cost of electricity for all households based purely on instantaneous observations and a price signal.

ShiftGuard: Towards Reliable 5G Network by Optimal Backup Power Allocation

Guoming Tang (Peng Cheng Laboratory, China); Yi Wang (Southern University of Science and Technology, China); Hongyu Lu (China Unicom, China)

0
The mobile network operators are shifting their network infrastructure from 4G/LTE to 5G in an unprecedented pace. The shift brings an explosive growth of 5G base stations. To guarantee the network reliability for 5G-enabled services, however, an extremely large number of backup batteries need to be deployed for the densely deployed base stations, leading to a huge investment for the mobile network operators. To cut down the cost and meanwhile keep the network reliability as it is expected, we propose ShiftGuard, an optimal backup power deployment framework. By leveraging the power demand differences among base stations from both spacial and temporal dimensions, ShiftGuard gives the most cost-efficient backup power allocation solution, also considering the constraints of network reliability and real-world deployment factors. Experiment results demonstrate that, compared to the strategy without backup power sharing or simply sharing with nearby base stations, ShiftGuard always achieves the best performance and cuts down the total cost by 27% ~ 40%.

DeepOPF+: A Deep Neural Network Approach for DC Optimal Power Flow for Ensuring Feasibility

Tianyu Zhao (The Chinese University of Hong Kong, Hong Kong); Xiang Pan (The Chinese University of Hong Kong, China); Minghua Chen (City University of Hong Kong, Hong Kong); Andreas Venzke (Technical University of Denmark (DTU), Denmark); Steven Low (California Institute of Technology, USA)

0
Deep Neural Networks approaches for the Optimal Power Flow (OPF) problem received considerable attention recently. A key challenge of these approaches lies in ensuring the feasibility of the predicted solutions to physical system constraints. Due to the inherent approximation errors, the solutions predicted by Deep Neural Networks (DNNs) may violate the operating constraints, e.g., the transmission line capacities, limiting their applicability in practice. To address this challenge, we develop DeepOPF+ as a DNN approach based on the so-called "preventive" framework. Specifically, we calibrate the generation and transmission line limits used in the DNN training, thereby anticipating approximation errors and ensuring that the resulting predicted solutions remain feasible. We theoretically characterize the calibration magnitude necessary for ensuring universal feasibility. Our DeepOPF+ approach improves over existing DNN-based schemes in that it ensures feasibility and achieves a consistent speed up performance in both light-load and heavy-load regimes. Detailed simulation results on a range of test instances show that the proposed DeepOPF+ generates 100% feasible solutions with minor optimality loss. Meanwhile, it achieves a computational speedup of two orders of magnitude compared to state-of-the-art solvers.

On the Double Auction Mechanism Design for Electricity Market

Jiaman Wu (Tsinghua University, China); Chenye Wu (The Chinese University of Hong Kong, Shenzhen, China)

0
The electricity market, due to its complex underlying physical constraints, suffers from market manipulation. Most existing markets employ the double auction to organize the markets: the independent system operator collects the bidding information from both the supply sides and solves the Walrasian equilibrium to conduct dispatch. Market manipulation arises when market players strategically bid their information. Hence to contain the manipulation, one promising solution is to design the double auction mechanism to induce truthful bidding. In this work, we customize four double auction mechanisms (Walrasian equilibrium Mechanism, VCG mechanism, MUDA (Lottery) mechanism, and MUDA (VCG) mechanism) for the electricity market. After proposing key metrics to evaluate the performance of various mechanisms, we conduct extensive numerical studies based on the real market data for a thorough comparison between the four mechanisms and identify the unique features for each mechanism. This could serve as the theoretical guidance for the double auction mechanism design for the electricity market.

Session Chair

Chen Chen (Xi'an Jiaotong University)

Enter Zoom
Session N1

COMM1: Communications for Planning and Situational Awareness

Conference
3:10 PM — 4:00 PM UTC
Local
Nov 11 Wed, 10:10 AM — 11:00 AM EST

Joint Capacity Modeling for Electric Vehicles in V2I-enabled Wireless Charging Highways

Dimitrios Sikeridis and Michael Devetsikiotis (University of New Mexico, USA)

2
Wireless Charging Highways (WCHs) have been introduced by industry and academia to enable charging-while-driving for electric vehicles (EVs) and to combat range anxiety. While detailed planning and performance evaluation of such systems are crucial due to high cost and long life expectancy, most existing works assume a perfect communication environment. In this paper, we introduce a joint capacity model that takes into account both power and communication resources for WCH construction planning, and optimal day-to-day operation. The vehicle-to-infrastructure (V2I) communication and grid power capacities, along with the EV's average service rate are formulated following technology requirements, EV speed-density characteristics, and the EV's energy needs and consumption. In addition, a two-dimension Markov chain-based model is designed to capture the WCH power and connectivity dynamics. The proposed model can be used to calculate the system's Quality of Service (QoS) and profit, provide design insights, and assess the impact of speed regulation, or admission control on the WCH lane. Finally, the performance of the proposed model is evaluated using real US highway data with the results demonstrating its ability to accurately capture the service provision dynamics, and to identify trade-offs between system parameters.

Evaluation of LTE Based Communication for Fast State Estimation in Low Voltage Grids

Leonard Fisser, Hanko Ipach, Andreas Timm-Giel and Christian Becker (Hamburg University of Technology, Germany)

0
In this work we investigate the applicability of Long-Term-Evolution (LTE) based communication systems operating in the low Ultra High Frequency range for a fast State Estimation in Low Voltage (LV) grids at high update rates. To estimate the state of the LV grid, the Branch-Current based State Estimation algorithm is modified such that voltage measurements are additionally considered while keeping the complexity low. A detailed LV grid topology and simulation model is then designed and used to find the communication networks performance for operation on 450 MHz and 2100 MHz. The resulting delay and packet loss rate distributions are used to evaluated the state estimation accuracy for different update intervals for a LV grid during a period of high photovoltaic feed-in variations. Optimal update rates are identified and further verified by analytical LTE cell load factor calculations. In the spirit of open science, the simulation code as well as the result files are publicly available at the reference provided in the conclusion.

Cyber-Resilience Enhancement of PMU Networks Using Software-Defined Networking

Yanfeng Qu, Gong Chen, Xin Liu and Jiaqi Yan (Illinois Institute of Technology, USA); Bo Chen (ANL, USA); Dong Jin (Illinois Institute of Technology, USA)

0
Phasor measurement unit (PMU) networks are increasingly deployed to offer timely and high-precision measurement of today's highly interconnected electric power systems. To enhance the cyber-resilience of PMU networks against malicious attacks and system errors, we develop an optimization-based network management scheme based on the software-defined networking (SDN) communication infrastructure to recovery PMU network connectivity and restore power system observability. The scheme enables fast network recovery by optimizing the path generation and installation process, and moreover, compressing the SDN rules to be installed on the switches. We develop a prototype system and perform system evaluation in terms of power system observability, recovery speed, and rule compression using the IEEE 30-bus system and IEEE 118-bus system.

Ukko: Resilient DRES Management for Ancillary Services using 5G Service Orchestration

Charalampos Rotsos (Lancaster University, United Kingdom (Great Britain)); Angelos K. Marnerides (University of Glasgow, United Kingdom (Great Britain)); Abubakr Magzoub (Lancaster University, United Kingdom (Great Britain)); Anish Jindal (University of Essex, United Kingdom (Great Britain)); Paul McCherry, Martin C Bor and John Vidler (Lancaster University, United Kingdom (Great Britain)); David Hutchison (Lancaster University & InfoLab21, United Kingdom (Great Britain))

0
The modern Smart Grid (SG) requires the adequate exploitation of Ancillary Services (AS) in order to dynamically support evolving energy demands, ensure grid stability and optimize energy trading transactions. Distributed Renewable Energy Sources (DRES) are considered as the most promising avenue to underpin the dynamic composition of AS. Nonetheless, the inherent legacy, distributed and resource-constrained properties of DRES deployments trigger a plethora of challenges with respect to the underlying end-to-end (E2E) data communication performance with direct implications on AS orchestration. Hence, assuring resilient and scalable E2E AS provisioning is a highly challenging task requiring advanced networking mechanisms. This work goes beyond architectural, policy-level and theoretical suggestions of how 5G can be utilized in the SG context and provide a proof-of-concept, system for DRES management. We introduce Ukko; an open-source, 5G network service design facilitating the programmable, "in-network" orchestration of DRES management, supporting real-time AS application requirements. Experiments conducted over a UK-wide 5G testbed using real use-case scenarios demonstrate that the proposed solution can assure scalable and resilient DRES management at the likely occurrence of data communication challenges.

An Optimal Wireless Resource Allocation of Machine-Type Communications in the 5G Network for Situation Awareness of Active Distribution Network

Qiyue Li (Hefei University of Technology, Hefei, China); Haochen Tang, Wei Sun, Weitao Li and Xiaobing Xu (Hefei University of Technology, China)

0
Advanced metering infrastructure is a key component of the active distribution network (ADN), aiming to enable massive multisource data in the ADN to be collected in real time. However, it is difficult to achieve optimal wireless resource allocation for massive devices, especially for scenarios where emergency faults occur randomly. In this paper, an optimal wireless resource allocation scheme based on the 5G network is proposed, which can optimally support collaborative scheduling and resource allocation for normal sampling data and emergency sampling data. The exhaustive simulation and experimental results show that with limited resource blocks, our proposed algorithm can maintain the dropping ratio of lower data packets while achieving optimal energy efficiency for massive smart meters, comparing with other typical counterparts.

Session Chair

Tan Thanh Le (Old Dominion Univ.)

Enter Zoom
Session S1

False Data Injection

Conference
3:10 PM — 4:00 PM UTC
Local
Nov 11 Wed, 10:10 AM — 11:00 AM EST

A Distributed Approach for Estimation of Information Matrix in Smart Grids and its Application for Anomaly Detection

Ramin Moslemi, Mohammadreza Davoodi and Javad Mohammadpour Velni (University of Georgia, USA)

0
Statistical structure learning (SSL)-based approaches have been employed in the recent years to detect different types of anomalies in a variety of cyber-physical systems (CPS). Although these approaches outperform conventional methods in the literature, their computational complexity, need for large number of measurements and centralized computations have limited their applicability to large-scale networks. In this work, we propose a distributed, multi-agent maximum likelihood (ML) approach to detect anomalies in smart grid applications aiming at reducing computational complexity, as well as preserving data privacy among different players in the network. The proposed multi-agent detector breaks the original ML problem into several local (smaller) ML optimization problems coupled by the alternating direction method of multipliers (ADMM). Then, these local ML problems are solved by their corresponding agents, eventually resulting in the construction of the global solution (network's information matrix). The numerical results obtained from two IEEE test (power transmission) systems confirm the accuracy and efficiency of the proposed approach for anomaly detection.

PMU and Communication Infrastructure Restoration for Post-Attack Observability Recovery of Power Grids

Shamsun Edib and Yuzhang Lin (University of Massachusetts, Lowell, USA); Vinod M. Vokkarane (University of Massachusetts Lowell, USA); Feng Qiu and Rui Yao (Argonne National Laboratory, USA); Dongbo Zhao (Georgia Institute of Technology, USA)

0
This paper is concerned about recovering the observability of cyber-physical power grids after massive cyber attacks, which helps to achieve the cyber-physical resilience of the grids. For recovering the observability of the grid, the measurability of the Phasor Measurement Units (PMUs) and the connectivity of the communication network are needed to be restored. The PMU and communication infrastructure restorations are jointly formulated as a Mixed Integer Linear Programming (MILP) problem to minimize the observability loss of the grid over time after a cyber attack while considering the constraint of limited resources. The efficacy of the proposed optimal restoration strategy is verified by comparing with heuristic methods on the IEEE 57-bus system.

Vulnerability Assessment of Large-Scale Power Systems to False Data Injection Attacks

Zhigang Chu (Arizona State University, USA); Jiazi Zhang (National Renewable Energy Laboratory, USA); Oliver Kosut and Lalitha Sankar (Arizona State University, USA)

0
This paper studies the vulnerability of large-scale power systems to false data injection (FDI) attacks through their physical consequences. An attacker-defender bi-level linear program (ADBLP) can be used to determine the worst-case consequences of FDI attacks aiming to maximize the physical power flow on a target line. This ADBLP can be transformed into a single-level mixed-integer linear program (MILP), but it is numerically intractable for power systems with a large number of buses and branches. In this paper, a modified Benders' decomposition algorithm is proposed to solve the ADBLP on large power systems without converting it to the MILP. Of more general interest, the proposed algorithm can be used to solve any ADBLP. Vulnerability of the IEEE 118-bus system and the Polish system with 2383 buses to FDI attacks is assessed using the proposed algorithm.

False Data Injection Cyber Range of Modernized Substation System

Muhammad M. Roomi (Illinois at Singapore Pte Ltd, Singapore); Partha P. Biswas and Daisuke Mashima (Advanced Digital Sciences Center, Singapore); Yuting Fan and Ee-Chien Chang (National University of Singapore, Singapore)

0
The extensive deployment of information and communication technologies in modern power grid makes the grid more vulnerable to cyber attacks, resulting in threats to the economy, social stability and sometimes human lives. While it is necessary to deploy cybersecurity measures to counter such threats, evaluation of the effectiveness and negative impact of the cybersecurity solutions through extensive experiments is challenging. For instance, it is often very restricted or even impossible to conduct experiments directly on the operating physical systems as the interruption of those critical systems will incur significant financial losses. In this paper, we discuss components and configurations that are necessary to implement a cyber range, a virtual environment for cybersecurity evaluation and experimentation, for emerging false data injection attacks against the smart grid. We also discuss the proof-of-concept implementation using open-source software and discuss case studies of attacks against modernized substation systems.

Information Theoretic Data Injection Attacks with Sparsity Constraints

Xiuzhen Ye and Iñaki Esnaola (University of Sheffield, United Kingdom (Great Britain)); Samir M. Perlaza (INRIA, France); Robert Harrison (University of Sheffield, United Kingdom (Great Britain))

0
Information theoretic sparse attacks that minimize simultaneously the information obtained by the operator and the probability of detection are studied in a Bayesian state estimation setting. The attack construction is formulated as an optimization problem that aims to minimize the mutual information between the state variables and the observations while guaranteeing the stealth of the attack. Stealth is described in terms of the Kullback-Leibler (KL) divergence between the distributions of the observations under attack and without attack. To overcome the difficulty posed by the combinatorial nature of a sparse attack construction, the attack case in which only one sensor is compromised is analytically solved first. The insight generated in this case is then used to propose a greedy algorithm that constructs random sparse attacks. The performance of the proposed attack is evaluated in the IEEE 30 Bus Test Case.

Session Chair

Enter Zoom
Session C1

Learning from Data

Conference
4:10 PM — 5:00 PM UTC
Local
Nov 11 Wed, 11:10 AM — 12:00 PM EST

Contract-Based Time-of-Use Pricing for Energy Storage Investment

Dongwei Zhao (The Chiniese University of Hong Kong, Hong Kong); Hao Wang (Monash University, Australia); Jianwei Huang (The Chinese University of Hong Kong, Hong Kong); Xiaojun Lin (Purdue University, USA)

0
Time-of-use (ToU) pricing is widely used by the electricity utility.
A carefully designed ToU pricing can incentivize end-users' energy storage deployment, which helps shave the system peak load and reduce the system social cost. However, the optimization of ToU pricing is highly non-trivial, and an improperly designed ToU pricing may lead to storage investments that are far from the social optimum. In this paper, we aim at designing the optimal ToU pricing, jointly considering the social cost of the utility and the storage investment decisions of users. Since the storage investment costs are users' private information, we design low-complexity contracts to elicit the necessary information and induce the proper behavior of users' storage investment. The proposed contracts only specify three contract items, which guides users of arbitrarily many different storage-cost types to invest in full, partial, or no storage capacity with respect to their peak demands. Our contracts can achieve the social optimum when the utility knows the aggregate demand of each storage-cost type (but not the individual user's type). When the utility only knows the distribution of each storage-cost type's demand, our contracts can lead to a near-optimal solution. The gap with the social optimum is as small as 1.5% based on the simulations using realistic data. We also show that the proposed contracts can reduce the system social cost by over 30%, compared with no storage investment benchmark.

A Multi-Agent Deep Reinforcement Learning Approach for a Distributed Energy Marketplace in Smart Grids

Arman Ghasemi, Amin Shojaeighadikolaei, Kailani Jones, Morteza Hashemi, Alexandru G. Bardas and Reza Ahmadi (University of Kansas, USA)

3
This paper presents a Reinforcement Learning (RL) based energy market for a prosumer dominated microgrid. The proposed market model facilitates a real-time and demand-dependent dynamic pricing environment, which reduces grid costs and improves the economic benefits for prosumers. Furthermore, this market model enables the grid operator to leverage prosumers' storage capacity as a dispatchable asset for grid support applications. Simulation results based on the Deep Q-Network (DQN) framework demonstrate significant improvements of the 24-hour accumulative profit for both prosumers and the grid operator, as well as major reductions in grid reserve power utilization.

Learning a Distributed Control Scheme for Demand Flexibility in Thermostatically Controlled Loads

Bingqing Chen and Weiran Yao (Carnegie Mellon University, USA); Jonathan Francis (Carnegie Mellon University & Bosch Research and Technology Center, USA); Mario Berges (Carnegie Mellon University, USA)

0
Demand flexibility is increasingly important for power grids, in light of growing penetration of renewable generation. Careful coordination of thermostatically controlled loads (TCLs) can potentially modulate energy demand, decrease operating costs, and increase grid resiliency. However, it is challenging to control a heterogeneous population of TCLs: the control problem has a large state action space; each TCL has unique and complex dynamics; and multiple system-level objectives need to be optimized simultaneously. To address these challenges, we propose a distributed control solution, which consists of a central load aggregator that optimizes system-level objectives and building-level controllers that track the load profiles planned by the aggregator. To optimize our agents' policies, we draw inspirations from both reinforcement learning (RL) and model predictive control. Specifically, the aggregator is updated with an evolutionary strategy, which was recently demonstrated to be a competitive and scalable alternative to more sophisticated RL algorithms and enables policy updates independent of the building-level controllers. We evaluate our proposed approach across four climate zones in four nine-building clusters, using the newly-introduced CityLearn simulation environment. Our approach achieved an average reduction of 16.8% in the environment cost compared to the benchmark rule-based controller.

Computationally Efficient Learning of Large Scale Dynamical Systems: A Koopman Theoretic Approach

Subhrajit Sinha (PNNL, USA); Sai Pushpak Nandanoori (Pacific Northwest National Laboratory, USA); Enoch Yeung (University of California Santa Barbara, USA)

0
In recent years there has been a considerable drive towards data-driven analysis, discovery and control of dynamical systems. To this end, operator theoretic methods, namely, Koopman operator methods have gained a lot of interest. In general, the Koopman operator is obtained as a solution to a least-squares problem, and as such, the Koopman operator can be expressed as a closed-form solution that involves the computation of Moore-Penrose inverse of a matrix. For high dimensional systems and also if the size of the obtained data-set is large, the computation of the Moore-Penrose inverse becomes computationally challenging. In this paper, we provide an algorithm for computing the Koopman operator for high dimensional systems in a time-efficient manner. We further demonstrate the efficacy of the proposed approach on two different systems, namely a network of coupled oscillators (with state-space dimension up to 2500) and IEEE 68 bus system (with state-space dimension 204 and up to 24,000 time-points).

Session Chair

Xiaojun Lin (Purdue Univ.)

Enter Zoom
Session IS-1

Energy Consumption

Conference
4:10 PM — 5:00 PM UTC
Local
Nov 11 Wed, 11:10 AM — 12:00 PM EST

A Cross-Domain Approach to Analyzing the Short-Run Impact of COVID-19 on the U.S. Electricity Sector

Le Xie (TAMU)

0
The novel coronavirus disease (COVID-19) has rapidly spread around the globe in 2020, with the U.S. becoming the epicenter of COVID-19 cases since late March. As the U.S. begins to gradually resume economic activity, it is imperative for policymakers and power system operators to take a scientific approach to understanding and predicting the impact on the electricity sector. Here, we release a first-of-its-kind cross-domain open-access data hub, integrating data from across all existing U.S. wholesale electricity markets with COVID-19 case, weather, mobile device location, and satellite imaging data. Leveraging cross-domain insights from public health and mobility data, we rigorously uncover a significant reduction in electricity consumption that is strongly correlated with the number of COVID-19 cases, degree of social distancing, and level of commercial activity.

A Locational Marginal CO2 Emissions Sensitivity to Help Manage Data Centers Carbon Footprint

Bernie Leseuitre (UW-Madison)

0
Data centers are large valuable flexible loads that are almost unique in their ability to allocate and shift computing tasks both temporally and geographically. Furthermore, many data center operators are environmentally conscious and have expressed goals to manage their carbon footprint. There are many ways to approach this. In this presentation we discuss how a locational marginal CO2 emissions sensitivity can be calculated and then used to guide load allocation among geographically distributed data centers to reduce overall carbon emissions.

Session Chair

Enter Zoom
Session S2

Smart Meter Data and Privacy

Conference
4:10 PM — 5:00 PM UTC
Local
Nov 11 Wed, 11:10 AM — 12:00 PM EST

Side Channel Security of Smart Meter Data Compression Techniques

Marcell Fehér and Niloofar Yazdani (Aarhus University, Denmark); Diego F Aranha (Aarhus University, Denmark & University of Campinas, Brazil); Daniel E. Lucani (Aarhus University, Denmark); Morten Tranberg Hansen and Flemming Enevold Vester (Kamstrup A/S, Denmark)

0
Given the large and sustained growth in the number of smart meters for different applications, e.g., electricity, water or heat, effective data compression has become increasingly important. Although smart meters tend to encrypt payloads using state-of-the-art solutions, the packet length variability introduced by compression of the data can be exploited in a side channel attack to gain knowledge about the consumption of individual meters. For example, a meter reporting constant (e.g. zero) consumption can be compressed more than one reporting more erratic usage. An attacker may gain knowledge of behavioral patterns of a household, e.g., when is no one home, or company, e.g., active periods of production. This paper analyzes the correlation between compressed packet length and reported consumption of multiple signals and practical reporting periods for the DLMS standard using real (anonymized) smart meter measurements.
We consider various built-in compressors and also propose new techniques that can both increase the compression and reduce this correlation. Our proposed schemes are particularly well suited for the increasingly popular case of high frequency reporting, e.g., reporting each measurement as it becomes available.

SearchFromFree: Adversarial Measurements for Machine Learning-based Energy Theft Detection

Jiangnan Li (University of Tennessee, USA); Yingyuan Yang (University of Illinois, USA); Jinyuan Sun (University of Tennessee, USA)

0
Energy theft causes large economic losses to utility companies around the world. In recent years, energy theft detection approaches based on machine learning (ML) techniques, especially neural networks, are becoming popular in the research community and shown to achieve state-of-the-art detection performance. However, in this work, we demonstrate that the well-trained ML models for energy theft detection are highly vulnerable to adversarial attacks. In particular, we design an adversarial measurement generation approach that enables the attacker to report extremely low power consumption measurements to utilities while bypassing the ML energy theft detection. We evaluate our approach with three kinds of neural networks based on a real-world smart meter dataset. The evaluation results demonstrate that our approach is able to significantly decrease the ML models' detection accuracy, even for black-box attackers.

On the Impact of Side Information on Smart Meter Privacy-Preserving Methods

Mohammadhadi Shateri and Francisco Messina (McGill University, Canada); Pablo Piantanida (CentraleSupélec-CNRS-Université Paris-Sud, France); Fabrice Labeau (McGill University, Canada)

0
Smart meters (SMs) can pose privacy threats for consumers, an issue that has received significant attention in recent years. This paper studies the impact of Side Information (SI) on the performance of distortion-based real-time privacy-preserving algorithms for SMs. In particular, we consider a deep adversarial learning framework, in which the desired releaser (a recurrent neural network) is trained by fighting against an adversary network until convergence. To define the loss functions, two different approaches are considered: the Causal Adversarial Learning (CAL) and the Directed Information (DI)-based learning. The main difference between these approaches is in how the privacy term is measured during the training process. On the one hand, the releaser in the CAL method, by getting supervision from the actual values of the private variables and feedback from the adversary performance, tries to minimize the adversary log-likelihood. On the other hand, the releaser in the DI approach completely relies on the feedback received from the adversary and is optimized to maximize its uncertainty. The performance of these two algorithms is evaluated empirically using real-world SMs data, considering an attacker with access to SI (e.g., the day of the week) that tries to infer the occupancy status from the released SMs data. The results show that, although they perform similarly when the attacker does not exploit the SI, in general, the CAL method is less sensitive to the inclusion of SI. However, in both cases, privacy levels are significantly affected, particularly when multiple sources of SI are included.

Online Energy Management Strategy Design for Smart Meter Privacy Against FHMM-Based NILM

Yang You and Tobias J. Oechtering (KTH Royal Institute of Technology, Sweden)

0
We consider the privacy-preserving problem for smart grid consumers where the adversary employs a factorial hidden Markov model based inference for load disaggregation. An online convex optimization framework is further proposed for the privacy-preserving energy management strategy design. With certain specific assumptions, the derived online energy management strategy is shown to have a sublinear dynamic regret and a sublinear dynamic fit, which means our proposed online algorithm has the asymptotic performance with the optimal offline dynamic benchmark. The performance of the design approach is finally illustrated in numerical experiments.

Session Chair

Enter Zoom
Session IS-2

Grid Resilience

Conference
5:10 PM — 6:00 PM UTC
Local
Nov 11 Wed, 12:10 PM — 1:00 PM EST

Extracting Resilience Metrics from Utility Data

Ian Dobson (Iowa State)

0
We show some new ways to extract resilience metrics from historical data that is routinely collected by utilities. For transmission systems, we use utility data that describes outcomes of resilience processes to drive a high-level probabilistic model that evaluates the probability distributions of resilience metrics, including high-impact rare events. This is joint work with Molly Rose Kelly-Gorham and Paul Hines of the University of Vermont. For distribution systems, we show how to decompose utility outage data into outage and restoring processes that can be described with standard resilience metrics. We quantify the uncertainty in the restoration time so that an upper bound on the restoration time can be estimated with 95% confidence. This is joint work with Nichelle’Le Carrington and Zhaoyu Wang of Iowa State University.

Coordinating DERs to Provide Ancillary Services without Hurting the Distribution Network

Johanna Matthieu (U. Michigan)

0
The power consumption of distributed energy resources (DERs) such as storage and flexible loads can be controlled to provide a variety of services to the electric grid to improve grid reliability, economics, and environmental impact. However, DER control schemes developed to provide transmission-level ancillary services can impact distribution network operation, causing under/overvoltages and transformer overheating. I will describe a variety of DER control architectures and strategies that can be used to manage distribution network impacts. In particular, I will discuss utility-aggregator coordination, where a load aggregator wishes to control electric loads to provide ancillary services but does not know the topology/parameters of the distribution network and the utility wishes to protect their network from negative impacts resulting from the actions of the load aggregator.

Session Chair

Enter Zoom
Session WAEG1

W-AEG1

Conference
5:10 PM — 6:00 PM UTC
Local
Nov 11 Wed, 12:10 PM — 1:00 PM EST

Matrix Completion Using Alternating Minimization for Distribution System State Estimation

Yajing Liu (National Renewable Energy Laboratory (NREL), USA); April Sagan (Rensselaer Polytechnic Institute, USA); Andrey Bernstein, Rui Yang and Xinyang Zhou (National Renewable Energy Laboratory, USA); Yingchen Zhang (UTK, USA)

0
This paper examines the problem of state estimation in power distribution systems under low-observability conditions. The recently proposed constrained matrix completion method which combines the standard matrix completion method and power flow constraints has been shown to be effective in estimating voltage phasors under low-observability conditions using single-snapshot information. However, the method requires solving a semidefinite programming (SDP) problem, which becomes computationally infeasible for large systems and if multiple-snapshot (time-series) information is used. This paper proposes an efficient algorithm to solve the constrained matrix completion problem with time-series data. This algorithm is based on reformulating the matrix completion problem as a bilinear (non-convex) optimization problem, and applying the alternating minimization algorithm to solve this problem. This paper proves the summable convergence of the proposed algorithm, and demonstrates its efficacy and scalability via IEEE 123-bus system and a real utility feeder system. This paper also explores the value of adding more data from the history in terms of computation time and estimation accuracy.

Computation-Efficient Algorithm for Distributed Feedback Optimization of Distribution Grids

Chin-Yao Chang, Xinyang Zhou and Andrey Bernstein (National Renewable Energy Laboratory, USA)

0
Feedback-based optimization algorithms use real-time measurements to update the optimal control for the underlying system which may not be fully identified. Recently, we have developed a distributed feedback-based algorithm that avoids the requirement of fast communication between central computing and local actuator/sensor agents. This paper extends the work by greatly reducing the number of copies of variables involved in the distributed feedback-based algorithm, which results in faster convergence and lower communication requirement. The main idea is to leverage the specific structural properties of the admittance matrix for distribution systems with tree network topology. We also show the effectiveness of the proposed algorithm in simulations.

Deep Learning for Reactive Power Control of Smart Inverters under Communication Constraints

Sarthak Gupta, Vassilis Kekatos and Ming Jin (Virginia Tech, USA)

0
Aiming for the median solution between cyber-intensive optimal power flow (OPF) solutions and subpar local control, this work advocates deciding inverter injection setpoints using deep neural networks (DNNs). Instead of fitting OPF solutions in a black-box manner, inverter DNNs are naturally integrated with the feeder model and trained to minimize a grid-wide objective subject to inverter and network constraints enforced on the average over uncertain grid conditions. Learning occurs in a quasi-stationary fashion and is posed as a stochastic OPF, handled via stochastic primal-dual updates acting on grid data scenarios. Although trained as a whole, the proposed DNN is operated in a master-slave architecture. Its master part is run at the utility to output a condensed control signal broadcast to all inverters. Its slave parts are implemented by inverters and are driven by the utility signal along with local inverter readings. This novel DNN structure uniquely addresses the small-big data conundrum where utilities collect detailed smart meter readings yet on an hourly basis, while in real time inverters should be driven by local inputs and minimal utility coordination to save on communication. Numerical tests corroborate the efficacy of this physics-aware DNN-based inverter solution over an optimal control policy.

Characteristics of Electric Vehicle Charging Sessions and its Benefits in Managing Peak Demands of a Commercial Parking Garage

Rongxin Yin and Doug Black (Lawrence Berkeley National Laboratory, USA); Bin Wang (Lawrence Berkeley National Lab, USA)

1
In this study, we use 28,262 charging sessions data collected from over 40 plug-in electric vehicle (PEV) charging stations during a period from March 2013 to September 2016. Following metrics are calculated for each charging session: (1) arrival and departure times; (2) plug-in and charging duration; (3) number of charging sessions; and (4) charging load flexibility. We develop two options in the discretized charging control algorithm: Option A changes only the charging start time and optimally shifts the entire period when the vehicle is actively charging within the connected charging session, and Option B splits the active charging period into 15-min segments and optimally spreads them over the connected charging session. Overall, the monthly bills savings are 21.3% and 23.5% in the summer season, 11.4% and 13.7% in the winter season, and 17.9% and 20.2% over the year for the control option 1-A and 1-B, respectively. Additionally, we develop a control algorithm in which all current charging sessions are re-optimized each time a new vehicle arrives and initiates a charging session. Greater cost savings can be achieved, e.g. 25.1% reduction for the summer season.

Deep Policy Gradient for Reactive Power Control in Distribution Systems

Qiuling Yang (Beijing Institute of Technology, China); Alireza Sadeghi (University of Minnesota, USA); Gang Wang (Beijing Institute of Technology, China); Georgios B. Giannakis (University of Minnesota, USA); Jian Sun (Beijing Institute of Technology, China)

0
Pronounced variability due to the growth of renewable energy sources, flexible loads, and distributed generation is challenging residential distribution systems. This context, motivates well fast, efficient, and robust reactive power control. Real-time optimal reactive power control is possible in theory by solving a non-convex optimization problem based on the exact model of distribution flow. However, lack of high-precision instrumentation and reliable communications, as well as the heavy computational burden of non-convex optimization solvers render computing and implementing the optimal control challenging in practice. Taking a statistical learning viewpoint, the input-output relationship between each grid state and the corresponding optimal reactive power control is parameterized in the present work by a deep neural network, whose unknown weights are learned offline by minimizing the power loss over a number of historical and simulated training pairs. In the inference phase, one just feeds the real-time state vector into the learned neural network to obtain the `optimal' reactive power control with only several matrix-vector multiplications. The merits of this novel statistical learning approach are computational efficiency as well as robustness to random input perturbations. Numerical tests on a 47-bus distribution network using real data corroborate these practical merits.

Session Chair

Enter Zoom
Session WMLPTS

W-ML-PTS

Conference
5:10 PM — 6:00 PM UTC
Local
Nov 11 Wed, 12:10 PM — 1:00 PM EST

Quantifying Load Uncertainty Using Real Smart Meter Data

Fankun Bu, Kaveh Dehghanpour, Yuxuan Yuan and Zhaoyu Wang (Iowa State University, USA)

0
As we get closer to customers in distribution systems, load stochasticity increases. In the past, due to lack of real-time data, the comprehensive knowledge of load behavior was limited, and simplistic assumptions had to be made for distribution system modeling and analysis, especially in the processes of network design and expansion. With the deployment of Advanced Metering Infrastructure (AMI), ample real-time smart meter data has become available to utilities. In this paper, using real hourly smart meter data, we have quantified load uncertainty in terms of average, maximum and maximum noncoincident
demands on a daily basis, as well as load factor and diversity factor. These uncertainty metrics are examined for individual residential, commercial and industrial customers, as well as distribution transformers serving residential customers. This paper provides a benchmark on load uncertainty quantification
for practicing engineers and researchers.

A Review on Artificial Intelligence for Grid Stability Assessment

Shutang You, Yinfeng Zhao and Mirka Mandich (University of Tennessee, USA); Yi Cui (University of Queensland, Australia); Hongyu Li and Huangqing Xiao (University of Tennessee, USA); Summer Fabus (EPRI, USA); Yu Su and Yilu Liu (University of Tennessee, USA); Haoyu Yuan (NREL, USA); Huaiguang Jiang (National Renewable Energy Laboratory, USA); Jin Tan (NREL, USA)

0
Artificial intelligence provides a fast approach for power grid stability assessment. Compared with simulation-based approaches, using artificial intelligence to assess stability can potentially save time on model development and numerical computation. This paper first reviewed existing literature on using artificial intelligence for power grid stability assessment. Then a unified machine leaning approach is proposed to assess power grid transient stability, frequency stability, and small signals stability simultaneously. Test results verified the accuracy and effectiveness of the proposed approach for power grid stability assessment.

Decision Trees for Voltage Stability Assessment

V. S. Narasimham Arava (GE Digital, United Kingdom (Great Britain)); Luigi Vanfretti (Rensselaer Polytechnic Institute, USA)

1
This paper proposes two different methods to train the DTs for voltage stability assessment, which in turn can aid in deriving preventive actions that can be given as recommendations to system operators or automatic load shedding schemes. In the voltage stability indices method, the DTs are trained on contingency cases that are classified based on voltage stability indices. In the region classification method, the DTs are trained on a new classification criterion that enlarges and generalizes the existing security boundary method of "stable" and "unstable" regions to a more granular operating space based on the distance from the nearest Saddle-Node Bifurcation. Case studies were performed using the Nordic 32 system for different contingency cases, several operating conditions and different network configurations. The ability to classify the degree of voltage stability of a multitude of operation conditions could be useful to aid operators in selecting and applying preventive measures to steer away the system from unstable conditions or conditions that are close to breaching operational requirements w.r.t. voltage stability.

Synthetic Training Data Generation for ML-based Small-Signal Stability Assessment

Sergio Dorado-Rojas, Marcelo de Castro Fernandes and Luigi Vanfretti (Rensselaer Polytechnic Institute, USA)

1
This article presents a simulation-based massive data generation procedure with applications in training machine learning (ML) solutions to automatically assess the small-signal stability condition of a power system subjected to contingencies. This method of scenario generation for employs a Monte Carlo two-stage sampling procedure to set up a contingency condition while considering the likelihood of a given combination of line outages. The generated data is pre-processed and then used to train several ML models (logistic and softmax regression, support vector machines, k-nearest Neighbors, Na•ve Bayes and decision trees), and a deep learning neural network. The performance of the ML algorithms shows the potential to be deployed in efficient real-time solutions to assist power system operators.

ML-based Data Anomaly Mitigation and Cyber-Power Transmission Resiliency Analysis

Anshuman Anshuman, Zhijie Nie, Sajan Sadanandan and Anurag. Srivastava (Washington State University, USA)

0
In recent years, the cyber and physical extreme events have increased and impacted the power system operations. Although there are multiple work reported for improving the resiliency of the power grid systems, there are a limited number of resiliency management tools available to the grid operators. Addressing the data quality issue is critical before feeding the measurements for situational awareness and decision-making using resiliency management tools. In this work, we describe an automated ML-based measurement data anomaly mitigation technique that uses regression, clustering, deep learning techniques as a base detector. Maximum Likelihood Criterion (MLE) based ensemble of these base detectors helps in anomaly detection and mitigation using SyncAED tool and feeding data for enhanced resiliency using a tool: Cyber-Physical Transmission Resiliency Assessment Metric (CP-TRAM). CP-TRAM utilizes real-time power grid data and aims to assist operators in measuring resiliency and ensuring the energy supply to critical loads given a cyber-attack or a natural disaster. This paper discusses the multiple ML algorithms for data anomaly detection, the basis of software design considerations, open-source software components, and use cases for the prototype developed tools.

Machine-Learning-Based Online Transient Analysis via Iterative Computation of Generator Dynamics

Jiaming Li (Stony Brook University, USA); Meng Yue (Brookhaven National Lab, USA); Yue Zhao (Stony Brook University, USA); Guang Lin (Purdue University, USA)

0
Transient analysis is vital to the planning and operation of electric power systems. Traditional transient analysis utilizes numerical methods to solve the differential-algebraic equations (DAEs) to compute the trajectories of quantities in the grid. For this, various numerical integration methods have been developed and used for decades. On the other hand, solving the DAEs for a relatively large system such as power grids is computationally intensive. and is particularly challenging to perform online. In this paper, a novel machine learning (ML) based approach is proposed and developed to predict post-contingency trajectories of a generator in the time domain. The training data are generated by using an off-line simulation platform considering random disturbance occurrences and clearing times in the New York/New England 16-machine 86-bus power system. As a proof-of-concept study, the proposed ML-based approach is applied to a single generator. A Long Short Term Memory (LSTM) network representation of the selected generator is successfully trained to capture the dependencies of its dynamics across a sufficiently long time span. In the online assessment stage, the LSTM network predicts the entire post-contingency transient trajectories given initial conditions of the power system triggered by system changes due to fault scenarios. The results show that the trained LSTM network reliably and accurately predicts the generator's transient trajectories. Compared to existing numerical integration methods, the post-disturbance trajectories of generator's dynamic states are computed much faster using the trained predictor, offering great promises for greatly accelerating both offline and online transient studies.

Session Chair

Enter Zoom

Made with in Toronto · Privacy Policy · © 2020 Duetone Corp.