2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)

Optimal Smart Grid Operation and Control

Enhancement by Edge Computing
Yuan Liao, Jiangbiao He
Department of Electrical and Computer Engineering
University of Kentucky
Lexington, USA
Emails: [email protected], [email protected]

Abstract—The national electric power grid is being transformed provide an accurate model of the power network for enhanced
into a smart grid through deploying a huge number of distributed planning and operation. These devices are placed at different
sensors across the network, a two-way communication system, locations capturing different quantities at various sampling
intelligent control and optimization algorithms, and advanced frequencies. For example, RTU usually performs a scan every
hardware components. An increasing volume of data are collected
four seconds, PMU every 1/30 second, and AMI every 15
by the sensing system, and transfer of these data to a central location
for centralized processing poses burden to the communication minutes. The intention to transform the national grid into a smart
system and the central computing system. Edge computing, a grid speeds up deployment of more sensing devices across the
distributed computing paradigm, processes data and makes proper network.
decisions locally, stores data locally and provides selected, processed At the transmission level, the system operator runs
data to a higher level, and thus may significantly relieve contingency constrained unit commitment and economic
communication burden and reduce response time of certain control dispatch programs to clear the power market considering the
applications. This paper explores possible applications of edge bidding from generators and distribution companies. The
computing to enhance distributed optimization and control of smart objective is usually to minimize the costs to secure enough
grid, including power system asset management, distributed
generation to meet the demand, while satisfying operating
charging scheme and microgrid protection.
constraints. In a market based active distribution system,
customers have generation resources, controllable load and
storage devices and actively participate in the distribution
operation. Participation can be through individual household, a
I. INTRODUCTION commercial building, or a microgrid (MG).
Electric power grid has been equipped with an increasing Recent rapid adoption and integration of renewable energy
number of measurement devices that have been continuously resources into power grid creates new paradigms of operation
capturing and producing a huge volume of time series analog and and control. More microgrids emerge as a feasible way for
status data at various spatial and temporal scales. Devices include increasing resiliency of power supply, utilization of renewable
protection and control devices (e.g., digital relays, reclosers, resources, and economy of operation. A MG is a small power
Remote Terminal Units (RTU)), smart meters with Advanced system that includes generation, load and storage system and can
Metering Infrastructure (AMI), general monitoring devices (e.g., operate either grid connected or standalone. The transition
digital fault recorders (DFR), power quality meters), and between the two states also requires control and protection
synchrophasor measurement devices Phasor Measurement Units considerations. MGs have different structures, including single
(PMU). The Supervisory Control and Data Acquisition MG or networked MGs, which have its own MG controller for
(SCADA) system periodically polls data from the power network managing the operation of MGs including power purchase and
or receives data when defined thresholds are met. The Network sale between the Area Power System (APS), optimal generation
Management System (NMS), including Energy Management resource dispatch in the MGs and load management. Protection
System (EMS) for transmission system and Distribution coordination of the MG and the feeder to which the MG is
Management System (DMS) for distribution system, connected is necessary.
encompasses a set of functions such as unit commitment, At a finer level, smart home energy management system has
economic dispatch, and voltage and var control for operating and also emerged, which determines the optimal schedule of the
controlling the system. Based on real time network information, operation of smart appliances and charging and discharging
the topology processor/state estimator (TP/SE) is utilized to scheduling of the battery storage system which may be installed
standalone or together with a home PV system.

978-1-7281-6127-3/20/$31.00 ©2020 IEEE

978-1-7281-6127-3/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on May 17,2022 at 12:12:59 UTC from IEEE Xplore. Restrictions apply.
 2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)

Electric vehicle (EV) charging/discharging management is

another feature of active distribution system. Charging stations
may be installed at individual residential homes or at bigger
commercial parking lots. A smart charge management (SCM)
system can be deployed to optimize the charging/discharging
schedule to achieve desired system performance. The system
consists of a central SCM system managed by the utility and
networked, distributed autonomous SCM agents deployed across
the distribution grid.
As shown above, there are a number of applications that
require or evolve to a computing paradigm consisting of a
centralized computing engine for processing a central data source
that concentrates data from across the network and distributed
computing engines for processing local data close to the edge
devices. Such a coordinated, distributed optimization, control
and protection scheme has become more prevalent in a smart
grid, especially with more abundant data produced by and more
processing power possessed by the edge devices. Moving a huge
amount of data over a network consumes a great deal of network
bandwidth and causes inherent response delays. On the other
hand, coordinated control on the higher hierarchy still plays an
important role. Therefore, it is natural to exploit the computing
power of edge devices and services to process raw data and make
necessary and timely decision locally and move selected,
processed data to the central location to be utilized by the central
controller. Reduction of transferring bulk data over network will
also mitigate vulnerability to cyber intrusion and enhance data
security. Processing and storing some data at the edge may Fig. 1. Transmission level operation and control.
provide a feasible solution to data privacy concerns with properly
designed policy. Consequently, such paradigm will improve
utilization of network bandwidth, improve response time, and
enhance system cybersecurity.
In the rest of the paper, Section II presents the overall power
system operation and control framework as well as the edge
computing structure. Then Section III describes specific smart
grid optimization and control applications that may take
advantage of edge computing. Conclusions are provided at the
end.
II. OVERALL FRAMEWORK FOR POWER SYSTEM OPERATION AND
CONTROL WITH EDGE COMPUTING

A. Smart grid operation and control overview

Fig. 1 and Fig. 2 present the overall framework for smart

grid operation and control at the transmission and distribution
level, respectively. Transmission and distribution SCADA, EMS
and DMS are the cornerstones for operating and controlling the
power grid. Emerging renewable energy resources, microgrids,
and sensors prompt advent of additional controllers and data-
driven analytics based on advanced algorithms including
artificial intelligence. The smart grid control paradigm will be a
combination of centralized and distributed control structure.
Application of the edge computing is promising for various Fig. 2. Distribution level operation and control [1].
applications in the smart grid, which would significantly
alleviate the concerns on data latency, response time, data
privacy and security from the conventional cloud-centric
structure by pushing the data processing tasks to the numerous
edge devices of the grid.

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on May 17,2022 at 12:12:59 UTC from IEEE Xplore. Restrictions apply.
 2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)

B. Overview of edge computing for smart grid operation and datacenter. Accordingly, there might be at least three types of
control communication paths in such a smart grid system, as listed
below:
(a) Edge device to edge node.
(b) Edge node to edge node.
(c) Edge node to cloud datacenter.

The potential communication protocols to interconnect the

edge devices, edge nodes, and the cloud datacenter can be
classified into two categories summarized as follows:
(a) Wired communication protocol: optical communications,
Ethernet, digital subscriber line (DSL), and power line
communications (PLC).
(b) Wireless communication protocol: ZigBee, Wi-Fi (IEEE
802.11), Bluetooth, 3G cellular, 4G LTE/LTE-A, and the like.
Wireless communications improve grid distribution
automation for the smart grid network, but they may possess
lower bandwidth and lower transmission range than the wired
communications. Unlike other general industrial applications,
the edge computing based smart grid typically requires the
communication and network to be secure, reliable and
sustainable. In [2], a system architecture is presented which
allows smart grid applications to operate at the mobile network
edge, precisely at Long-Term Evolution (LTE) base stations,
Fig. 3. Overall framework for applying edge computing to smart grid operation providing low latency communication to smart meters. In [3], a
and control. machine-to-machine communication network is proposed for
edge computing in smart grid communication, which claims to
Fig. 3 depicts the overall framework for applying edge be effective and reliable. More discussion and review on the
computing to smart grid operation and control. As shown, nested secure, reliable, and flexible communication techniques for
or cascaded edge computing structure may be feasible. A smart grids are presented in [4]. A comprehensive consideration
subsystem that implements an edge computing paradigm serves may need to be conducted to select the appropriate
an edge node within a larger system that also follows the edge communication network technology based on the requirements
computing paradigm. of the specific range, bandwidth, security, reliability, and
For example, a manufacturing factory can implement an asset implementation cost.
management system based on edge computing, where each asset
like an electric machine or transformer is constantly monitored, III. EDGE COMPUTING APPLICATIONS
diagnosed and maintained by distributed edge nodes/services
A. Power system asset management
within a department, and a centralized optimizer processes and
optimizes information for all the assets in the entire factory. At This section presents potential application of edge computing
the same time, the central optimizer can provide data and services for online fault detection and maintenance of edge assets such as
to another optimizer at a similar level or a higher hierarchical electric machines and power transformers.
level. Electric machines such as wind turbine generators or ocean
Another example is illustrated in the top portion of Fig. 2, wave energy generators have been broadly used in smart grids in
where cameras capture sky cloud image and weather stations the past decades. Specifically, the most common generators used
capture weather conditions like temperature and wind speed. in smart grids include permanent magnet synchronous machines
Edge nodes/ distributed generation (DG) controllers mitigate fast (PMSM) and doubly fed induction generators (DFIG) due to their
voltage variations due to transient DG output change. In the high efficiency and flexible control advantages [5]. However,
central/cloud level, the global optimizer obtains the controller these generators are easily subject to electrical faults such as
settings for distributed controllers across the system including stator winding faults, rotor faults, or bearing faults, caused by
load tap-changer transformer, voltage regulators, capacitor banks several different failure mechanisms, as reported in [6]. For
and DGs (wind, solar, etc.) [1], and only limited data need to be instance, an onshore wind generator installed in the turbine
communicated from edge nodes to the central controller at a nacelle is typically tied to a back-to-back power converter
specific time interval. located on the bottom of the tower through long power cables
(e.g., above 100 meters), and the surge impedance mismatching
C. Communication and network between the generator and converter combined with the fast-
As shown in Fig. 3, there are mainly three layers in the smart switching of the power converter may induce large reflected
grid architecture from the perspective of edge computing, surge voltage spikes superposed on the generator stator
including the edge devices (wind turbines, solar farms, EVs, windings, damaging the insulation of the generator stator
etc.), edge nodes (smart inverters, controllers, etc.), and the cloud windings and incurring short-circuit or open-phase faults [7]. In

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on May 17,2022 at 12:12:59 UTC from IEEE Xplore. Restrictions apply.
 2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)

such a scenario, fast online health monitoring of the generator optimizer makes decision on overall maintenance and operation
will be very critical to avoid or minimize the downtime cost. schedule optimization for all the equipment across the
Likewise, large power transformers (LPT), as an essential manufacturing power system.
element in the power transmission-distribution system, are also
vulnerable to insulation failures, regardless of dry transformers
or oil-cooled transformers. The 2015 CIGRE survey of 964
prominent transformer failures found that the major reason for
transformer collapse was dielectric breakdown (i.e., insulation
failure) [8]. Such fault can be definitely predicted and detected
based on the instantaneous winding hottest-spot temperature and
the load mission profile data, as specified in the IEEE Std.
C57.91-2011 [9]. Therefore, online prognosis and diagnosis of
these critical power equipment has been of paramount
importance for the reliable operation of the smart grid.
In [10]-[12], health condition monitoring of the edge assets in
the smart grid has been proposed to be implemented based on Fig. 4. Power system asset management based on edge computing.
cloud computing. However, the fault detection speed for
generators is typically evaluated by milliseconds or even
microseconds, and the signal processing of the massive local data
transmitting to the SCADA will cause severe time delay and even
incur cascaded system failures, due to the delayed fault
protection and fault-tolerant operation command. On the
contrary, with the emerging edge computing approach, the fault
signatures from the generator voltages, currents, and power can
be extracted and computed locally, eliminating the necessity of
transmitting massive of data to the cloud. Only the data
associated to systematic actions will be delivered to the cloud,
which significantly reduces the overall bandwidth requirement
and amount of data sent to the cloud. A power system asset
management architecture based on edge computing is shown in
Fig. 4, where the health condition of the edge devices such as
sensors, generators, and transformers will be monitored locally,
and the identified fault information will be transmitted to the
central/cloud datacenter through the edge nodes to schedule a
maintenance, lifecycle management or fault-tolerant operation. Fig. 5. Manufacturing power system fault monitoring and diagnosis.
The scheme is generally applicable to a manufacturing power
system, a distribution system, and a large power system.
In addition to health monitoring, the presented concept is Edge nodes may provide a great deal of computation and
applicable to broader monitoring and protection against natural analysis such as wavelet transform, Fourier transform, and
and intentional physical and cyber attacks for the individual recurrent neural network training for local equipment diagnosis.
equipment and the entire power system of interest. Fig. 5 depicts Only processed data such as voltage and current magnitude and
the scheme for fault monitoring and diagnosis for a angle, frequency, and harmonic and unbalance information need
manufacturing power system. Across the power system, sensors to be communicated to the central level for processing. This
are strategically placed at the equipment (generator, motor, paradigm dramatically saves data bandwidth and makes the most
transformer) terminals and selected feeder locations. Quantities of the computing power of edge nodes.
like voltage and current waveforms for each equipment are
B. Optimal distributed charging station management
monitored and analyzed by edge services that provides data
analytic functions based on signal processing and machine This application is to develop a novel smart charge
learning techniques. Any faults including equipment internal management (SCM) system that consists of a central SCM
faults and terminal faults and possible equipment controller system managed by the utility and networked, distributed
faults or cyber attacks will be identified. Waveforms across the autonomous SCM agents deployed across the distribution grid,
system are analyzed by the central analyzer to locate the source as depicted in Fig. 6. The central SCM coordinates and controls
of the fault that may occur anywhere in the system [13]. The distributed SCMs to achieve optimal system performance, while
faults could be short-circuit faults or intentional cyber attacks SCM agents at the edge nodes are capable of operating
such as injection of harmonics and unbalances sources, controller autonomously for managing local charging as well as for
attacks, and denial of service attacks that will interrupt the providing valuable system services including frequency support,
operation or degrade the performance of equipment on the voltage support, peak load shaving, maximizing renewable
factory floor. Edge services also serve to fine tune equipment for integration, and the like.
optimal performance and maintenance scheduling, while central

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on May 17,2022 at 12:12:59 UTC from IEEE Xplore. Restrictions apply.
 2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)

The central SCM system aims to optimize the scheduling of

charging and discharging of plug-in electric vehicles (PEVs)
across the network based on statistical parking patterns of PEVs
and distribution network data, which will be integrated into the
power distribution management system. Such proposed SCM
system will utilize stochastic optimization and co-optimize
multi-objectives for providing benefits to grid operator, PEV
owner and charging network operator. The multi-objective
optimization algorithm consists of day-ahead planning and real-
time operation. The SCM aims to reduce distribution transformer
loading variation during the study horizon, minimize system
losses, and maximize renewable generation while mitigating
duck curve phenomena during the study horizon (e.g., 24 hours),
while the system should abide by the power flow, voltage and Fig. 6. Proposed coordinated control of distributed SCM systems.
ampere limits, as illustrated in Fig. 7. The PEV owner will be
incentivized for participating in the demand response program by
supplying or consuming real and reactive power. Reactive power
support is enabled by smart inverters of the charging stations.
The proposed optimization framework will also realize an
integrated voltage and var optimization and control scheme by
fully utilizing the capabilities of smart inverters and PEV
resources as well as existing voltage and var regulating
equipment including capacitor banks, voltage regulators, and
load tap changer transformers. Smart inverters are capable of
mitigating fast voltage fluctuations due to renewable generation
intermittency and PEV charging and save mechanical controllers
from tear and wear. Fig. 7. Illustration of peak load shaving with the SCM system.
The proposed solution may possess the following prominent
features: (1) Coordinated control architecture enables scalability,
improve bandwidth, and increases response time. (2) Stochastic
where, is the set representing study hours, represents the
multi-objective optimization techniques are employed to provide
transformer set, denotes the transformer loading at hour for
incentives to all stakeholders while dealing with various
transformer , E represents expected value. P and Q denote real
uncertainties. (3) Achieving integrated voltage, reactive power,
and reactive power, V is the node voltage, and u symbolizes the
and frequency control and maximizing utilization of renewable
control variables that consist of the settings of voltage regulating
energy, (4) Enabling large-scale deployment of charging
controllers and charging and discharging actions for PEVs. g
stations, which provides optimal incentives for wider penetration
stands for the general power flow equations, and h denotes the
of PEVs, and motivating more PEV owners to get involved in the
general form of constraints such as voltage limits, cable ratings,
V2G/G2V programs. (5) Providing grid services to empower the
transformer ratings, power factor limits, harmonic limits, and
grid operators for better operating and controlling the power grid
with the presence of a large number of PEVs. controller setting limits. h designates the limits for distributed
There are various uncertainties in system planning and generations (DGs) and PEV smart inverters. Pr represents the
operation including load and the number and location of PEVs probability of DG power output, which models uncertainty of
parked in the charging stations. Chance-constrained stochastic renewable generation including solar generation, and the
optimization approach can be developed to consider the probability of available PEVs at parking locations, which are
uncertainties. Take reducing the expected value of transformer ready to be harnessed. The solution to the problem will
loading variation during the study horizon as an example. Such a determine the optimal charging, discharging schedule for PEVs,
problem is formulated as follows: settings of voltage regulating devices, reactive support schedules
for DGs and PEVs.
In real-time operation, an intelligent local reactive and voltage
Minimize ∑ ∑ E
control algorithm will be carried out at the edge nodes to mitigate
voltage fluctuations by utilizing smart inverter capability. At a
Subject to: g P , Q , P , Q , P , Q , u, V =0 certain operating point, the sensitivity of system voltages with
h P,Q ,P ,Q , P , Q , u, V 0 respect to real and reactive power of DG and PEV can be
h P ,Q ,P ,Q 0 calculated based on power flow calculations. The nodes of the
Pr P , Q , P ,Q grid will be partitioned into groups, each of which is associated
Pr P , Q , P ,Q with a DG/PEV, such that the DG/PEV power fluctuation
imposes the greatest impact on the voltage change of the nodes
in this group. The average value of the sensitivity of voltages in
a group to generation fluctuation can then be determined. The
amount of reactive power to counter the voltage change can also

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on May 17,2022 at 12:12:59 UTC from IEEE Xplore. Restrictions apply.
 2020 IEEE International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm)

be computed. Then the reactive power supplied by the DG and across the system. Future research will provide more detailed
PEV can be calculated according to the changes of the real analysis of implementing the proposed methods.
power.
A DG/PEV controller itself can serve as the edge node which REFERENCES
processes local information and calculates local settings, or
several DG/PEV controllers can be served by a more dedicated [1] Invited book chapter, Yuan Liao, Power Management and Protection of
edge node, which can perform a localized optimization to Distribution Systems with Distributed Generations in book Handbook of
Distributed Generation, Springer International Publishing AG, Cham,
achieve a specified system performance such as minimizing the Switzerland, 2017.
voltage variation of selected buses. The global optimization will [2] A. F. R. Trajano, A. A. M. de Sousa, E. B. Rodrigues, J. N. de Souza, A. de
be carried out by the central optimizer at the cloud/central level. Castro Callado and E. F. Coutinho, "Leveraging Mobile Edge Computing
Local computation and processing at the edge nodes greatly on Smart Grids Using LTE Cellular Networks," 2019 IEEE Symposium on
Computers and Communications (ISCC), Barcelona, Spain, 2019, pp. 1-7.
reduces the need of transferring real time data to a control center,
[3] E. Oyekanlu, C. Nelatury, A. O. Fatade, O. Alaba and O. Abass, "Edge
timely processes the data locally, and provides real time response computing for industrial IoT and the smart grid: Channel capacity for M2M
to the DG/PEV controllers. Only a limited amount of processed communication over the power line," 2017 IEEE 3rd International
data is transmitted to the higher level for coordinated control and Conference on Electro-Technology for National Development
optimization. On one hand, this helps relieve the latency and (NIGERCON), Owerri, 2017, pp. 1-11.
computation burden on the cloud datacenter. On the other hand, [4] M. Erol-Kantarci and H. T. Mouftah, "Energy-Efficient Information and
Communication Infrastructures in the Smart Grid: A Survey on Interactions
this approach avoids transmitting massive end-user data to the and Open Issues," in IEEE Communications Surveys & Tutorials, vol. 17,
public cloud posing severe privacy and cyber security concerns. no. 1, pp. 179-197, 2015.
[5] H. Li and Z. Chen, "Overview of different wind generator systems and their
C. Microgrid protection comparisons," in IET Renewable Power Generation, vol. 2, no. 2, pp. 123-
In MG protection based on edge computing, strategically 138, June 2008.
placed sensors across the system measure the voltage and current [6] J. Li, Q. Li and J. Zhu, "Health condition assessment of wind turbine
generators based on supervisory control and data acquisition data," in IET
waveforms including traveling waves generated by the faults. Renewable Power Generation, vol. 13, no. 8, pp. 1343-1350, 10 6 2019.
Edge nodes process the traveling wave data to obtain wavefront [7] J. He, G. Y. Sizov, P. Zhang and N. A. O. Demerdash, "A review of
information including magnitude, arrival time, and traveling mitigation methods for overvoltage in long-cable-fed PWM AC drives,"
wave direction, in addition to the fundamental frequency voltage 2011 IEEE Energy Conversion Congress and Exposition, Phoenix, AZ,
and current phasors [14]. Edge nodes can make a decision on the 2011, pp. 2160-2166.
occurrence of the fault and where the fault is based on local [8] M. Rafiq, et al. “The impacts of nanotechnology on the improvement of
liquid insulation of transformers: Emerging trends and challenges” J. Mol.
information by utilizing both circuit theory centered analytical Liq. 302 2020 112482 p.1-2.
methods and data centered artificial intelligence techniques. [9] IEEE Guide for Loading Mineral-Oil-Immersed Transformers and Step-
More accurate fault detection and location can be determined by Voltage Regulators," in IEEE Std C57.91-2011 (Revision of IEEE Std
the central system that collects the processed information from C57.91-1995), vol., no., pp.1-123, 7 March 2012.
the edge nodes. [10] R. T. and A. S., "A cloud computing approach for power management of
microgrids," ISGT2011-India, Kollam, Kerala, 2011, pp. 49-52.
IV. CONCLUSIONS [11] P. Lipčák, M. Macak and B. Rossi, "Big Data Platform for Smart Grids
Power Consumption Anomaly Detection," 2019 Federated Conference on
Smart power grid consists of a large number of networked Computer Science and Information Systems (FedCSIS), Leipzig, Germany,
components, controllers, sensors, network and communication 2019, pp. 771-780.
systems, and advanced algorithms. Smart grid control system [12] H. Bai, Z. Ma, and Y. Zhu, “The Application of Cloud Computing in Smart
consists of central and distributed controllers. Emerging data Grid Status Monitoring,” In: Wang Y., Zhang X. (eds) Internet of Things.
Communications in Computer and Information Science, vol 312. Springer,
sources call for intelligent data-driven, machine learning based Berlin, Heidelberg, 2012.
analytics that demand computational power. Edge computing [13] Wen Fan, Yuan Liao, “Fault location for distribution systems with
paradigm provides a promising way for enhancing smart grid distributed generations without using source impedances,” 51st North
operation and control. Edge nodes provide computation and American Power Symposium, Wichita, Kansas, October 13-15, 2019.
processing for a great deal of information captured from local [14] Wen Fan, Yuan Liao, “Traveling wave based method for protecting power
edge devices and make real time decision with minimum distribution systems with distributed generations,” International Energy &
Sustainability Conference 2019, Farmingdale, NY, USA, October 17 - 18,
response delay, rather than send all data to the public cloud 2019.
datacenter causing latency and security issues. The central/cloud
level controller carries out the global, coordinated optimization
and control process to obtain settings for distributed controllers

Authorized licensed use limited to: INTERNATIONAL ISLAMIC UNIVERSITY. Downloaded on May 17,2022 at 12:12:59 UTC from IEEE Xplore. Restrictions apply.