IEEE Consumer Electronics Magazine - January/February 2024 - 92

Special Section on Security and Privacy-Aware Emerging Computing
top priority for mitigating the effects of climate
change.
With the advancement of information and
communication technology, artificial intelligence
(AI) technologies are increasingly being used in
the management of electricity grids, giving rise
to a new field known as smart energy grids.1
Smart grids control the flow of electricity and
information through the use of the Internet of
Things and smart meters.2,3 They aim to improve
the efficiency, sustainability, reliability, flexibility,
and security of electricity grids is through
the use of AI techniques.
When it comes to sustainable electricity grids,
renewable energy generation is an indispensable
issue in smart grids;4,5 however, renewable energy
is not as competitive as conventional energy sources
in terms of price in the electricity market. The
economic mechanism of subsidy is widely used in
real grid management. Traditionally, subsidies for
each type of electricity generation are determined
by government,6 fair competition, and classical
economic analysis.7,8 For state-of-the-art research,
Q-learning9 and game theory10 are used to determine
subsidies. However, existing models make
their decisions based on limited and low information
that is not robust enough to determine appropriate
subsidies. In this article, we develop a deep
reinforcement learning approach for determining
subsidies using high-dimensional information.
In terms of flexibility and reliability, unexpected
electricity shortages may occur. Therefore,
a portion of the electricity capacity should be
reserved as a backup, which is called an operating
reserve. Operating reserves are essential in fragile
electricity grids, such as high-green-energy or lowincome
countries. Since the reserved capacity
must meet the demand in real time,11
dynamic operating reserve systems are proposed8
with Q-learning,9 multiobjective optimization,12
and probability forecasting.13 To avoid waste,
reserved electricity can be stored and scheduled
for later demand. Traditionally, excess electricity
is stored in large-scale pumped storage systems.14
With the rapid development of battery technology,
electricity can be stored in household electricity
storage, and even within electric vehicle batteries,15
which are decentralized and flexible for electricity
storage and scheduling. However, there are
nonnegligible losses in storing and transmitting
92
the electricity and a tradeoff between grid reliability
and energy efficiency. Therefore, accurate forecasting
of demand, dispatching of supply, and
determination of operating reserves are essential
to the flexibility, reliability, and efficiency of a
smart-grid system;5,13 those factors are involved in
the designed gridmanagement system.
The more information are deployed in smart
grids, the more they are exposed to cyber-physical
attacks, so security and privacy issues are
receiving much attention. In the current research,
authentication systems,16 cloud edge-based fault
detection,17 and neural network-based anomaly
detection18 have been proposed to prevent malicious
users and man-in-the-middle attacks. Distributed
learning is also a potential technology to
improve privacy and security, which distributes
learning tasks across multiple nodes to achieve
higher scalability and efficiency.19 Popular algorithms
for distributed learning include federated
learning and split learning. Federated learning
algorithms allow multiple nodes to learn a model
together without sharing their data,20 while, split
learning algorithms share knowledge without
sharing data and models, and divide the model
into segments that are trained jointly by servers
and edge devices.21 This research will leverage
distributed learning, federated learning, and split
learning technologies to improve the privacy and
security of the electricity grid management
system.
BLUEPRINT OF GRID MANAGEMENT
In this research, we aim to propose a design
several
for an AI-based electricity grid management system
that focuses on sustainability, reliability,
and security, as shown in Figure 1. The system
consists of a prediction module, an anomaly
detection module, a grid management module,
and a market equilibrium module.
In Figure 1, there are multiple suppliers and
demanders in the electricity market, and the
developed forecasting module predicts their electricity
supply and demand. In addition, the anomaly
detection module detects potential attacks or
faults for security reasons. The prediction and
anomaly detection results are used to set renewable
energy subsidies (sustainable electricity generation)
and operating reserve rates (reliable
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2024

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2024
