=Paper=
{{Paper
|id=Vol-3804/paper4dc
|storemode=property
|title=AI applications in SG for reliability, security, and stability
|pdfUrl=https://ceur-ws.org/Vol-3804/paper4dc.pdf
|volume=Vol-3804
|authors=Theodore Kindong
|dblpUrl=https://dblp.org/rec/conf/bir/Kindong24
}}
==AI applications in SG for reliability, security, and stability==
https://ceur-ws.org/Vol-3804/paper4dc.pdf
AI applications in SG for reliability, security, and stability
Theodore Kindong1
1 Linköping University, SE-581 83 Linköping, Sweden
Abstract
The new paradigm in clean, sustainable, dependable, and efficient energy generation, and delivery, has
led to the transformation and innovation of traditional grid to smart grid. The transformation and
innovation use advanced technologies such as AI and IoT to monitor and control power generation,
transmission, and distribution processes in a smart grid (SG). AI applications in SG have emerged as an
innovation that guarantees effective, flexible, reliable, sustainable, decentralized, secure, and cost-
effective distribution and management of energy in SG. This study is a research proposal for AI
Applications in SG for Reliability, Security, and Stability. It begins by introducing the SG and its related
challenges, followed by AI applications in SG and its implementation challenges. The study identifies
research problems in AI applications in SG to be AI interpretability and formulates three research
questions that can help address the problem identified. Also, this paper presents the results of the
literature review conducted to provide a sufficient grounding for this study and discusses the following
concepts of AI application in SG. Predictive Analytics in SG, AI-enabled Demand Response in SG, AI-
enabled Control and Coordination in SG, AI-enabled security, stability, and reliability analysis in SG, and
Implementation challenges of AI applications in SG. The study proceeds to discuss the proposed
theoretical approach and the chosen research methodology and then concludes with the expected
study contribution to research and practice.
Keywords
Smart grid, artificial intelligence, predictive analytics, SG stability, and AI interpretability1
1. Introduction
The term "smart grid" refers to the transformation of a traditional electric power grid, which was
originally regulated using electromechanical methods, into a network that is controlled using
information and communication technologies (ICTs). The Smart Grid (SG), as outlined in the US
Department of Energy's Smart Grid System Report, includes information management, control
technologies, digitally based sensors, information and communication technologies (ICTs), and
field devices[1]. Its deployment has arisen as a viable way to improve energy efficiency and
tackle the problems presented by increasing energy consumption and environmental issues. Its
rise has been due to the integration of Internet of Things (IoT) technology and other emerging
technologies into the traditional grid. Which has facilitated the implementation of sophisticated
monitoring and control systems to enhance energy management efficiency. SG system employs
IoT-enabled sensors and devices to gather real-time data on energy usage, generation, and
environmental conditions [2]. These sensors offer a plethora of data that may be utilized to study
energy patterns, detect inefficiencies, and make well-informed decisions to enhance energy
efficiency.
Due to the expansion of the SG, which includes more interconnection, greater integration of
renewable energy, widespread use of direct current power transmission systems, and the
liberalization of electricity markets[1-3]. The smart grid facilitates the gathering of vast
quantities of complex and diverse data on the operations of the electric power grid [1]. This is
BIR-WS 2024: BIR 2024 Workshops and Doctoral Consortium, 23rd International Conference on Perspectives in
Business Informatics Research (BIR 2024), September 11-13, 2024, Prague, Czech Rep.
theodore.kindong@liu.se (T. Kindong)
0009-0001-6502-8325 (T. Kindong)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
achieved by combining modern metering infrastructure, control technologies, and
communication technologies[1, 2]. This rapid expansion requires effective stability analysis and
control to ensure a stable, reliable, and secure operation of the smart grid. Thus, the stability
characteristics of the smart grid have become significantly more difficult compared to the past.
Necessitating a transition from conventional stability analysis and control methodologies to
more sophisticated stability analysis and control due to the limitations of traditional stability
analysis and controls in terms of speed, efficacy, and cost [3].
More so, the vast amount and multidimensional data collected pose analysis challenges for
conventional modeling, optimization, and control technologies that are restricted in their ability
to handle large amounts of data, leading to a growing recognition of the need for artificial
intelligence (AI) techniques in the smart grid [1]. Hence, various artificial intelligence algorithms
are being applied in SG for stability analysis and control, such as load forecasting, power grid
stability evaluation, fault detection, and security issues. Furthermore, research asserts that the
utilization of AI approaches can augment and enhance the dependability and robustness of smart
grid systems[1]. This has continued to be the case with the rise in machine learning (ML)
applications and the growing use of data-driven approaches to improve resilience, stability,
reliability, and security in SG. This shows significant prospects for utilizing ML techniques in SG
to anticipate power outages (POP), forecast energy demand, perform security, analyze stability,
and control. Hence, artificial intelligence (AI) applications in SG offer robust and encouraging
tools for analyzing stability and controlling smart grids, which continue to gain increasing
attention in academics and practice.
Despite the overwhelming interest in the application of AI in SG, they have yet to significantly
address how AI applications in SG can be transparent and human-understandable. Humans are
increasingly interested in understanding and interpreting AI decisions within SG. However,
existing literature on AI applications in SG has not given sufficient attention to the necessity to
address AI algorithms’ inherent opacity and their functioning as "black boxes"[4]. This raises the
challenge of transparency and AI interpretability, which is the capacity to provide explanations
to humans about the functions of AI and clarity about AI decisions. Hence, this study seeks to
explore AI interpretability within SG can be addressed. To meet users of AI applications
continuous desire to comprehend what the AI model has acquired from the data, understand and
interpret AI decisions, and determine bias when AI makes false predictions as well as return the
autonomy to make the ultimate decision. Furthermore, the absence of interpretability in current
literature poses a significant risk, particularly in application scenarios that demand high levels
of security, such as the smart grid. Therefore, the objective of this study is to investigate how
interpretable models can be applied in SG. Interpretable models offer the ability to identify and
pinpoint the underlying source of abnormalities, providing valuable insights into the reasons for
their failure[3]. The remainder of this study covers the background and research problem,
previous research, knowledge gaps, the study’s significance, and research questions. The second
section of this work presents a literature review and chosen theoretical lens. The chosen
research methodology follows this and finally concludes with the expected study’s contribution
to research and practice.
1.1. Background and research problem
Electricity has been a paramount and extensively utilized kind of energy since the 19th
century[3]. The electric power grid, responsible for transmitting electricity from power plants
to end-users, has been hailed as the most significant engineering accomplishment of the 20th
century[5] and has become an essential component of contemporary civilization. Nevertheless,
the electricity grid continues to evolve. The growing integration of renewable energy sources,
such as wind and solar energy, into power networks is being driven by worries over fossil fuel
depletion, climate change, global warming, and the decreasing prices associated with these
sources[3]. The evolution of the electricity grid has continued with the integration of emerging
disruptive technologies such as IoT and AI tackled primary obstacles in the energy system such
as environmental awareness, inadequate clean and sustainable energy management, insufficient
optimization of energy distribution and transmission, high costs associated with power transfer,
and increased customer awareness of energy charges [6].
Similarly, SG relies on communication and IT for its operation [7], which involves a network
of sensors in transmission-distribution infrastructure, remote monitoring, SCADA, lumbar
systems, and household appliances. Meters, sensors, and synchrophasors provide significant
data [5, 8]. This vast amount of data requires the best mechanisms for handling, investigating,
and evaluating it to ensure SG stability, security, reliability, and resilience. Hence, AI techniques
are being adopted to produce intelligent electricity that is stable and secure. AI techniques
enable control hubs and energy supply structures to remotely monitor power-driven campaigns
to advance and analyze the power system in real time, reducing principal times [9].
Also, integrating renewable energy sources and technologies such as IoT raises the problem
of SG's stability, reliability, security, control, and resilience. Hence, to ensure widespread access
to the grid while prioritizing high levels of stability, control, security, and resilience, AI
applications in SG have emerged as a potential solution. AI applications in SG have been used to
enable real-time grid analysis and predictive analysis using historical and current data to
provide an accurate representation of the current grid state to producers and consumers.
Additionally, it can predict future grid status. Therefore, it is feasible to detect power grid
transmission loss, locate the overheated line, identify missing power connections, make
decisions, and implement self-healing measures[6]. AI techniques promise to help prevent
power system failures, including minor outages and major blackouts, through power outage
prediction and weather forecasting. These types of events provide relief to both customers and
the nation, as they prevent significant economic and business losses.
Despite the promising future of AI applications in SG for communicating, retaining
information, and making decisions through analysis. There has been growing interest in
addressing AI interpretability in SG. As SG combines human and technological interactions, there
is a need for humans within SG to be able to interpret AI decisions and their operations for a
more accurate analysis, intuitive, and collaborative system. AI interpretability in SG provides an
opportunity to effectively integrate the behaviors of producers, consumers, and users to provide
reliable, cheap, and reliable power sources.
1.2. Previous research
In recent years, there has been increasing interest in both academia and industry on AI
applications in SG. This is due to a global increase in the popularity of artificial intelligence (AI)
and its transformative nature in other domains. Their inventive and disruptive nature has led to
their widespread recognition and practical applications in several industries [1, 10]. Previous
research has examined the implementation of artificial intelligence in the SG and substantial
changes resulting from integrating and utilizing artificial intelligence (AI) in SG [1, 5, 6, 8, 11, 12].
More so, past studies have explored how AI techniques such as machine learning can be used
in power outage prediction and energy demand forecast [9, 11, 13-16]. Kehkashan et al. [16],
proposed a systematic approach for selecting optimal machine learning models to improve the
performance of the SG. Their study examines the efficacy of machine learning algorithms by
utilizing performance evaluation measures as the fundamental criteria for selecting the most
optimal model [16]. Similarly, other studies have sought to address security concerns in SG using
AI techniques, such as the work of Abdullah et al. [17]. Their work examined how the integration
of AI with Blockchain distributed ledger technology (BDLT) can be applied in SG in the realm of
renewable energy and associated power automation [17]. Also, Liu et al. [18] address security in
SG by exploring a new method for detecting a specific type of attack, known as Bad Data Injection
(BDI), in the Smart Grid. Their method combines information from the network traffic flow and
the power system's physical laws to create a unified model called Abnormal Traffic-indexed State
Estimation [18].
Furthermore, prior research has mostly addressed distinct issues, such as the analysis and
control of stability in an efficient manner [3], as well as the utilization of Deep Learning (DL) for
predicting power consumption during peak hours[19, 20]. Similarly, Barth et al. [14] investigate
how distributed reinforcement learning can enhance decision-making autonomy and
collaboration in SG.
However, despite the overwhelming interest in AI applications in SG, there needs to be more
literature on how AI interpretability is addressed to improve trust and transparency. The
absence of technical expertise and the inability to comprehend AI judgments [21] have not been
resolved, resulting in trust issues around the use of AI in SG. Current research has been scarce
on how artificial intelligence's predictive models in SG can be made transparent and easily
comprehensible to people. Hence, this study aims to build on the fact that human intervention
can improve AI applications when AI systems fail or exhibit bias [22, 23]. To explore how AI
interpretability can be integrated into SG when solving issues related to stability control,
dependability, security, and transmission cost systems in SG. It will also investigate how AI
interpretability in SG can increase the accuracy of predictive models and support human
operators in obtaining and exploiting data throughout the grid. The study will build on the
growing interest in AI applications in SG to explore the integration of AI interpretability for all
participants' more effective, intuitive, and collaborative use. This approach aims to achieve
global optimization and distribute the resulting expenses across all participants. AI applications
in the smart grid offer several advantages, including global optimization, cost-sharing among
players, cost-effectiveness, environmental friendliness, stability analysis, and developing a
dependable SG [1, 3].
1.3. Knowledge gaps
Current studies on AI applications in SG have often neglected the aspect of interpretability, and
the research on this topic is still in its early stages, with only a limited number of references
available. The black-box aspect of AI poses a challenge to its implementation in smart grids, as
most AI agents cannot achieve perfect accuracy and lack an understanding of the specific reasons
behind AI failures. Given the importance of SG stability, security, and management for safety,
operators must comprehend the decisions made by AI agents and identify the source of any
abnormalities, which has not been the case in the current literature. AI must earn the trust of
operators before it can be extensively utilized for smart grid stability analysis and control.
Also, studies around the AI applications in SG have predominantly addressed the technical
aspect of AI, principally focusing on improving AI performances and ignoring AI algorithms'
potential biases and failures. The research in this area is still in its early stages, with only a few
references available. The black-box aspect of AI poses a challenge to its implementation in smart
grids as most AI agents are unable to achieve perfect accuracy, and we need an understanding of
the specific reasons behind AI failures. Given the importance of power grid stability and
management for safety, operators must comprehend the decisions made by AI agents and
identify the source of any abnormalities. Only until AI gains the trust of operators can it be
extensively utilized for smart grid stability analysis and control.
1.4. Importance and significance of the research
With the rising adoption of AI techniques in SG to guarantee stability, reliability, resilience, and
security and the growing reliance on these techniques, it is crucial to understand the trade-off
between Accuracy and Interpretability for successful implementation in SG. Hence,
interpretability has been a significant focus of AI research. However, this feature has received
limited attention in the field of SG research. Typically, the effectiveness and comprehensibility of
AI methods are two variables that must be balanced against each other [24] , making this study
very important. The current approaches of AI applications in SG have achieved exceptional
performance at the expense of extensive abstraction, hence the necessity to achieve a somewhat
effective equilibrium between the accuracy and interpretability of AI.
Furthermore, SG is a critical infrastructure requiring effective security, stability, reliability,
and resilience. Thus, the application of AI in SG must adopt an effective method for creating AI
algorithms that are easier to understand by incorporating a user interface that presents the AI
algorithm's results and reveals part of the underlying reasoning behind the AI decision process
[21]. The need to make AI techniques in SG offers explanations for important AI decisions that,
when partially accurate and, in some cases, can give misleading reassurance, [21] reinforces the
importance of this study. Also, it is important to develop new training methods for creating high-
quality AI algorithms in SG that are easily understandable for security assessment. A study that
integrates AI interpretability in SG by providing a coherent explanation for the AI algorithms'
output and identifying the factors contributing to system instability and security issues is crucial.
1.5. Research questions
The objective of this study is to examine the current body of literature on the use of artificial
intelligence (AI) in smart grids (SG), assess specific instances of AI implementation, and establish
the boundaries for ensuring the interpretability of AI in SG. Existing research on AI in SG
primarily emphasizes AI's technical capabilities while neglecting the crucial aspect of human
comprehension and interpretation of AI decisions. In contrast, this study takes a distinct
approach by addressing AI's technical abilities and the human perspective.
Therefore, our research aims to utilize the latest advancements in artificial intelligence (AI)
and its applications, as well as big data and data generated by smart grids (SG), to investigate
how AI can be applied transparently to smart grids. The objective is to augment human
comprehension of AI determinations and promote their involvement and self-governance to
optimize energy distribution and administration in smart grids. This will be achieved by
answering the following research questions:
RQ1: What is the present state of AI research in SG?
RQ2: How can we balance AI's effectiveness in SG and its ease of understanding?
RQ3: How does AI interpretability in SG affect its reliability, security, and stability?
To provide sufficient grounding for the study and answer the above research questions, the
next sections of the paper present a preliminary literature review.
2. Literature review
A comprehensive analysis of the existing literature was undertaken to provide additional clarity
regarding the framework and themes of this study. A comprehensive literature evaluation is
essential for any study. So, it is for this PhD thesis as it gives ideas and concepts to support the
chosen approach to the issue, helps determine the appropriate methodology, reveals areas
where information is lacking, and demonstrates the distinctive contribution of the thesis[25].
The literature evaluation aimed to facilitate the formulation of theories, address research gaps,
and identify areas that require more investigation [26].
2.1. Search strategy
The comprehensive literature requires a search strategy to identify relevant literature. Thus, a
keyword search was conducted in the relevant databases LiU UniSearch, Scopus, IEEE Xplore,
Google Scholar, ACM, and Web of Science to collect pertinent publications related to the subject
of interest. This search followed the systematic procedure of doing a literature review outlined
by [26] . A comprehensive literature search was conducted, which included a systematic search
of relevant titles in reverse order to ensure thorough coverage. This involved searching for
related literature among the references cited by studies identified in the initial keyword search.
The author initially identified the key themes of "Smart Grids," artificial intelligence,"
“reliability,” “predictive analytics,” "smart grid security," and "stability of smart grid" in Scopus,
a comprehensive database of peer-reviewed literature. Four concepts were discovered while
investigating AI-enabled Reliability, Security, and Stability in Smart Grid Systems: Demand
response, artificial intelligence techniques, intrusion detection, and security. Subsequently, we
identified specific terms or acronyms associated with each concept. For instance, "power outage
prediction" and "deep learning" serve as terms or acronyms for the concept of "machine
learning". Ultimately, we employed the OR operator to combine terms or concepts associated
with the same concept, whereas the AND operator was utilized to merge unrelated concepts. Due
to the query phrase employed in our search on Scopus, we obtained many search results. The
same was done for other databases, and inclusion and exclusion criteria were used to select
quality papers for analysis. The selected articles were synthesized into the themes below.
2.2. Predictive analytics in SG
Predictive analytics are crucial in the administration of SG systems. Forecasting models for
unregulated factors (such as the generation of renewable energy sources and building energy
usage) are necessary for the optimal management of SG, as these models enable informed
decision-making and facilitate fault identification and diagnostics[13]. SG consists of smart
meters and sensors that capture real-time data on energy use, voltage, current flows, and other
factors across the grid. The collected data is processed, aggregated, and stored in a central
repository called the data management system[2]. This system integrates data from several
sources to provide a complete perspective of grid performance. It uses information, two-way
communication technologies, and computational intelligence to ensure stability, security,
reliability, resilience, sustainability, and efficiency[7, 12, 27]. Thus, predictive analytics have
emerged in the literature, especially for real-time data collection and analysis in SG, which is
used for efficient control and coordination through data-decision making. Also, with the growing
demand for environmentally friendly local electricity production and delivery in an urban area,
predictive analytics is used as an optimization technique to assess historical data and forecast
future energy demand patterns[13, 16, 20, 28].
More so, predictive analytics makes modern AI models essential for risk prediction and
decision-making [23]and has gained interest in the area of AI applications in SG. It uses data
mining, predictive modelling, and machine learning to analyze historical and real-time data. The
technique is the “modern oracle of our networked digital age” [23]. SG uses AI to improve
predictive analytics for fault diagnosis, prediction, decision-making, and optimization, such as in
the work of [11, 13, 15, 20, 28].
Also, Ahmad et al. [13] comprehensive study compares tree-based ensemble machine
learning models (random forest – RF and extra trees – ET), decision trees (DT), and support
vector regression (SVR) to forecast solar thermal collector system useable hourly energy[13].
Their approach involved training and testing machine-learning models with experimental data.
Similarly, AI-enabled predictive analysis helps fight climate change, lower energy transmission
costs, and predict energy demand and grid stability[16]. Thus, energy producers and consumers
accept predictive analytics and seek ways to improve it [28]. SG power systems benefit greatly
from AI techniques, according to Bose [11]. He gives a brief but comprehensive explanation of
expert systems (ES), fuzzy logic, and artificial neural networks.
Furthermore, Zhongtuo et al. [3] present a clear summary of SG's use of predictive analytics
for grid stability analysis and control. AI in SG analyzes smart grid security, stability, fault
diagnostics, and stability control. Abdullah et al.'s study focuses on One of the real-time analyses
of the physical layer of the smart grid, leveraging predictive analytics for intelligent information
processing (IIP) to SG and its management and bidirectional data channel applications, ensuring
secure communication through an effective control mechanism. Predictive analytics in SG
improves supervisory control and data acquisition (SCADA) systems' monitoring, control, and
coordination[29]. It also allows for reconfiguring the power system, advanced metering
infrastructure (AMI), protection, distribution automation (DA), and embedded intelligence that
prioritizes self-healing, optimization, and recovery from anomalies, which are also improved
[30].
2.3. AI-enabled demand response in SG
Demand response (DR) is a key concept in SG, encouraging users to discharge non-essential
electricity during peak hours to balance peak-hour electricity supply. AI applications in SG have
also been used to optimize DR by integrating different AI techniques in SG to estimate consumer
electricity consumption and automate DR. DR in SG has been addressed in the literature, such as
the work of Qunzhi et al. [31]. They used semantic Web approaches to create an integrated Smart
Grid information model and present semantic information case studies for dynamic DR. Their
work demonstrates that the semantic model simplifies information integration and knowledge
representation for subsequent Smart Grid applications[31].
Similarly, José R. and Zoltán [32], in their work, reviewed the use of reinforcement learning
(RL) for demand response applications in the SG to control diverse energy systems such as
electric vehicles, heating, ventilation, and air conditioning (HVAC) systems, smart appliances, or
batteries[32]. More so, Ma et al.[20] , in their work, explored the method of distribution
optimization on the multi-agent system to determine the ideal network weights for different
stakeholders, resulting in an optimal dynamic pricing strategy that provides smart grid economic
efficiency and security[20]. Their proposed method improves the optimization of the RL
algorithm, stakeholder privacy, and decision-making autonomy in SG. Furthermore, Boopathy et
al. [19] , in their study, cover deep learning (DL) applications for intelligent smart grid demand
response. Presented DL fundamentals in SG demand response and examined cutting-edge DL
applications in SG, including electric load forecasting, state estimation, energy theft detection,
energy sharing, and trading[19]. Finally, we discuss existing research problems, critical issues
and potential paths in DL for smart grids and demand response. Deep learning (DL) models can
discover patterns from the massive SG network data and estimate electricity demand and peak
hours. Several studies have examined DL principles for DR in SG.
2.4. AI-enabled control and coordination in smart grids
Control and coordination are becoming vital as smart grids transmit large amounts of
information. Hence, existing literature has examined how SG uses cutting-edge power
electronics, computer systems, information technology, emerging technologies, and cyber
technology. To coordinate, produce, distribute, and use electricity sustainably, environmentally
friendly, and reliable[3]. Qunzhi et al.[31] argue that Smart Grids' ability to collect huge
information enables new software applications and tools to revolutionize macro and micro
power consumption management to satisfy rising electricity demand[31]. Also, AI techniques
have emerged as more agile tools for controlling and coordinating generation transmission and
distribution of electricity in SG from and to the power grid, as well as real-time pricing by third-
party service providers to control home energy use[31].
Zhongtuo et al. [3] argue that transmissive information in SG requires sophisticated
information processing to add value and meaning in real time to ensure effective stability
analysis, control, and coordination. The expansion of the smart grid, which includes more
interconnection, greater integration of renewable energy, widespread use of direct current
power transmission systems, and the liberalization of electricity markets, has made grid stability
more complex and communication more transmissive[3]. AI-enabled Intelligent information
processing allows smart grids to integrate computer processing logic, vast database repositories,
and communications network connectivity, expanding the concept of control and coordination
before "facilitating or enabling certain tasks"[3, 11, 14]. Conversely, growing artificial
intelligence (AI) approaches offer robust and promising tools for smart grid stability analysis
and control and are gaining interest.
2.5. AI-enabled security, stability, and reliability analysis in SG
SG integrates renewable energy sources, decentralized power generation units, energy storage,
and plug-in hybrid electric vehicles (PHEV), raising security, stability and reliability concerns.
Hence, current literature has explored AI-enabled security approaches and stability and
reliability analysis[3, 9, 15, 17, 18, 33]. Rahman et al. [33] proposed a new approach for customer
reliability in SG by forecasting distribution power systems using a fault tree technique with
customer-weighted component failure frequencies and downtimes[33]. Their method goes a
notch higher than traditional electric grid customer reliability forecast that uses system average
(SA) component failure frequency and downtime weighted by component amount[33]. To
include weight component, failure frequency and downtime predictions with customer
disturbance data. Similarly, other studies have investigated how different AI systems can detect
abnormalities and predict assaults to protect stakeholders' privacy by prohibiting the flow of
personally identifiable information[9, 15, 17, 18].
More so, Zhongtuo et al.[3] provide a thorough and lucid overview of the current progress of
AI applications in SG with a comprehensive introduction to AI, encompassing its definitions,
historical background, and cutting-edge approaches. Their study thoroughly examines how it
might be applied to evaluate security, measure stability, diagnose faults, and control stability in
smart grids[3]. Also, Abdullah et al.[17] review the latest integrated artificial intelligence and
blockchain-enabled smart grid and power distribution automation scheduling, management,
optimization, privacy, and security[17]. Their research focuses on real-time smart grid physical
layer analysis and automation using AI and blockchain.
2.6. Implementation challenges of AI applications in SG
Despite the growing interest in AI applications in SG, AI capabilities and the ability to address
many more difficult problems than conventional mechanism-based approaches. Also, AI
applications in SG have been extensively researched and yielded outstanding results. The
practical implementation of AI applications in SG confronts obstacles such as high data
requirements, learning from imbalanced data, interpretation, transfer learning, robustness to
communication quality, and robustness to attack or adversarial examples[3].
More so, Smart grids face the ongoing issue of managing massive amounts of data with
significant fluctuation due to their reliance on power systems for communication networks[1].
This also raises another significant barrier to AI applications in SG, especially for power outage
prediction, which is data quality, since it depends on how the ML algorithms are trained and
tested[16].
Finally, the issues of AI interpretability in SG hugely affect the implementation of AI
applications in SG, as it limits the integration of domain expertise. AI algorithms can be complex
and challenging to comprehend[1, 3, 16], and the absence of interpretability in AI models
presents a difficulty to leveraged domain expertise gained over the years, of power system
operation. The expertise reduces AI's data dependence and boosts performance. The skill could
be used to integrate data-driven approaches with symbolic AI or knowledge engineering.
2.7. Proposed theoretical approach
SG has emerged as an innovative technology that addresses the growing energy demand, and it
has continued to evolve with the application of AI in SG. AI applications in SG are a recent
technology that requires methods that should be used immediately in their design and
implementation to ensure their adoption. Involving an inter-organizational community with
many different types of people creates and uses a vision of the SG innovation that is key to its
early and later spread [34]. This vision is very important for understanding, validating, setting
up and accelerating economic tasks and changes in SG. Several institutional forces affect how an
organizing idea grows and what effect it has[34]. Also, considering the early stage of AI
applications in SG, there are many doubts about its benefits, usability in diverse contexts, usage
patterns, and future[35]. Hence, organizations need to understand a new idea before adopting it.
This study adopts organizational vision theory as its theoretical lens to investigate how
organizations understand AI applications in SG. According to what Creswell and Creswell [36]
call a "theoretical lens," this study uses organizing vision theory to shed light on AI applications
in SG. We use this idea on purpose because organizations and individuals try to understand
innovations before adopting them, and by using this knowledge, people become adopters or non-
adopters[37]. The comprehension process offers much room for interpretation regarding how
IT artefacts work collaboratively with humans in a team. The team members are analyzed
regarding their motives and behaviours [38].
To better understand how humans benefit from the computing capabilities of AI and how AI
uses information and large amounts of data generated from the SG. The organizational vision, a
“focal community idea for applying information technology in organizations”, [34] influence
potential adopters' decision-making and comprehension. The organizing vision is shaped by
public discourse between suppliers, consultants, journalists, university researchers, early
adopters, practitioners, and executive groups. The community calls each organizational vision
by its name and incorporates metaphors, scenarios, stories, difficulties, and dilemmas [34]. The
organizational vision will enhance AI interpretability, allowing users to interpret, legitimize, and
mobilize new technologies. Interpretation clarifies the AI's existence, decisions, and function to
reduce uncertainty. Legitimization relates AI applications in SG to benefits and commercial
operations[34] such as stability, reliability, and security.
3. Methodology
The methodology encompasses planning and procedures for data collecting, analysis, and
interpretation, from basic assumptions to specific methods. It also presents the philosophical
tradition, methodological approach, and method of data collection and analysis.
3.1. Philosophical tradition (paradigm)
The philosophical paradigm is the researcher's philosophical assumptions, that informed the
strategies, data collecting, analysis, and interpretation methodologies that should guide this
selection. To construct an effective plan, researchers must consider their philosophical
worldview assumptions, the inquiry strategy associated with this worldview, and the research
procedures or processes that translate the approach into practical application[39]. Information
systems research is predominantly categorized into positivism, interpretivism, and critical
realism. This research focuses on comprehending the epistemological assumptions, which are
the principles for creating and assessing reliable knowledge about a particular phenomenon
[40], hence adopting the interpretive paradigm.
The choice for interpretive paradigm is because it is primarily linked to qualitative research
as it asserts that reality is socially constructed (ontology) and can only be comprehended by
interpreting the underlying significance that people attribute to it (epistemology) [36]. This
study seeks to understand events by accessing the subjective interpretations that individuals
attribute to them. Unlike the previous "descriptive" studies on AI applications in SG, this study
does not accept the notion of an "objective" or "factual" explanation of events and situations.
Instead, it aims to achieve a relativistic yet commonly understood comprehension of
phenomena[40]. The study's objective is to heavily depend on the participants' viewpoints
regarding the topic being examined.
3.2. Methodological approach
This study adopts ethnomethodologically informed ethnography (EM) as a research method, as
it seeks the integration of ethnography into the systems development process. This is driven by
the belief that the social context in which systems are put greatly impacts their success[41]. AI
applications in SG are utilized in social situations, regardless of their technological features,
investigating their design and use is best suited in their natural setting. Hence, ethnography,
which focuses on observing interactions in natural situations, can provide a social perspective
on system design[42]. The choice of EM is because it might inform system design and the
challenges of integrating research findings with information system designers' needs. It is
suitable for addressing challenges in transdisciplinary collaboration between designers and EM
researchers. This study explores how developing information systems under a specific
framework can address the valid complaints of thoughtless technical implementation[42].
More so, examining circumstances in their authentic contexts and comprehending or
interpreting occurrences related to the significance individuals attribute to them[43], offers a
better approach to addressing the issues of AI interpretability in SG. Also, Christin's [4] research
expands upon the work of prior ethnographers such as Seaver, presenting an alternative
epistemological perspective that diverges from the conventional "black box" framework[4]. This
design involves a comprehensive description, analysis, and interpretation of social expressions
between people and groups, which typically refers to a program, event, activity, process, or one
or more individuals[36]. This research design is selected because the study focuses on a current
subject that necessitates a specific time frame and organized research activities for the
researcher to gather comprehensive information using different data-collecting methods during
the specified duration.
3.3. Methods of data collection
The method of data collection employs "scavenging" methods to collect relevant information
from various sources (such as informal conversations, official announcements, reviews from AI
apps in SG, and industry conference areas). To illuminate the intricate relationship between
computer systems' social, cultural, and technological components in our daily lives.
More so, the study will observe and interview specialists to acquire actual data on algorithm
creation and use in SG. It will include participatory and non-participatory observations
supported by semi-structured qualitative interviews with administrators, data scientists, and
community users.
3.4. Method of data analysis
To get clarity and valuable information from the collected data, the study will use thematic
analysis. A technique for finding, examining, and interpreting meaning patterns or "themes" in
qualitative data[44]. Hence, the study will record and evaluate qualitative inquiry transcripts
using coding methods comparable to Cech [45]. To identify similarities between workplace
cultures and algorithmic design, interviews and field notes will be transcribed and categorized
using grounded theory and thematic analysis[46]. To show how social practices influence
algorithm use and user impact.
4. Expected contribution
This study uses research techniques to solve a practical problem. Thus, it will benefit
practitioners working on AI applications in SG. Initially, AI app developers for SG will gain, by
learning how to incorporate domain expertise to design and deploy interpretable AI in SG. The
findings will also assist systems or software developers, IT security specialists in creating secure,
robust, and interpretable AI models, and company leaders and managers preparing to use AI.
This PhD thesis research could help make AI technology more inclusive by incorporating domain
expertise, semantic information, and interpretability into its material design.
Also, energy producers and consumers will utilize this knowledge to construct stable, reliable,
and secure SG using interpretable AI algorithms for real-time detection of anomalies and
intrusions, power outage prediction, and demand response. According to Gregor [47], the
finding's theoretical contribution to practitioners is design and action. Similarly, the study’s
contribution to knowledge according to Gregor's categories [47] will be explanatory because it
explains interpretable AI in SG. AI applications in SG are gaining tremendous interest, hence the
impact of AI's "black box" characteristics on smart grid operations and how smart grid operators
can tackle this issue. It will help academics create human-centered AI solutions in SG. This study
will demonstrate how interpretable AI solutions in SG might benefit a growing research area.
References
[1] O. A. Olufemi, H. N. Haoran, Artificial intelligence techniques in smart grid: A Survey Smart
Cities, 2021.
[2] S P. Sunkari, K. Shreeya, S. R. Muthyala, D. S. Datrika, S. S. Munigala, Analysis and functioning
of smart grid for enhancing energy efficiency using optimization techniques with IoT, in
International Conference on Cybernetics, Cognition and Machine Learning Applications
(ICCCMLA). 2023, IEEE.
[3] S. Zhongtuo, Y. Wei, L. Zhouping, Z. Lingkang, Z. Yifan, Z. Runfeng, T. Yong, W. Jinyu. (2020),
Artificial intelligence techniques for stability analysis and control in smart grids:
Methodologies, applications, challenges and future directions. Applied Energy, Elsevier
2020.
[4] A. Christin, The ethnographer and the algorithm: beyond the black box. Theory and Society:
Springer Nature 2020.
[5] G. W. Arnold. (2011), Challenges and Opportunities in Smart Grid: A Position Article.
Proceedings of the IEEE, 2011.
[6] H. S. Mhamud, M. C. Safwat, S. K. Dey, A. Hafiz, H. A. Sarafat, H. Mehedi, K. S. Subrata, T. Zinat.,
Towards electric digital twin grid: Technology and framework review. Elsevier, Energy and
AI, 2023. 11.
[7] M. Faheem, S. B. H. Shah, R. A. Butt, B. Raza, M. Anwar, M. W. Ashraf, M. A. B. Ngadi, V. C.
Gungor, Smart grid communication and information technologies in the perspective of
Industry 4.0: Opportunities and challenges. Computer Science Review, 2018.
[8] E. A. Bhuiyan, M. Z. Hossain, S. M. Muyeen, S. R. Fahim, S. K. Sarker, S. K. Das, Towards next
generation virtual power plant: Technology review and frameworks. Renewable and
Sustainable Energy Reviews, 2021.
[9] R. K. Beniwal, M. K. Saini, A. Nayyar, B. Qureshi, A. Aggarwal, A critical analysis of
methodologies for detection and classification of power quality events in smart grid. IEEE
Access, 2021.
[10] J. Wang, M. K. Lim, C. Wang, M.-L. Tseng, The evolution of the Internet of Things (IoT) over
the past 20 years. Computers & Industrial Engineering, 2021. 155.
[11] B. K. Bose, Artificial Intelligence Techniques in Smart Grid and Renewable Energy Systems—
Some Example Applications. IEEE XPLORE, Proceedings of the IEEE, 2017. 105(11): p. 2262
- 2273.
[12] S. Dhara, A. K. Shrivastav, P. K. Sadhu, Smart grid modernization: Opportunities and
challenges. Electric Grid Mod, 2022.
[13] M. W. Ahmad, J. Reynolds, Y. Rezgui, Predictive modelling for solar thermal energy systems:
A comparison of support vector regression, random forest, extra trees and
[14] D. Barth, B. Cohen-Boulakia, W. Ehounou, Distributed Reinforcement Learning for the
Management of a Smart Grid Interconnecting Independent Prosumers. Energies, 2022. 15.
[15] H. Cao, D. Zhang, S. Yi, Real-Time Machine Learning-based fault Detection, Classification,
and locating in large scale solar Energy-Based Systems: Digital twin simulation. Solar
Energy, 2023.
[16] F. Kehkashan, S. Hussain, F. B. Cautillo, A. A. Basir, & A. W. Ling, Machine learning for power
outage prediction during hurricanes: An extensive review. The International Journal of
Intelligent Real-Time Automation, 2024.
[17] A. K. Abdullah, A. L. Asif, R. Mamoon, A. R. Jahed, & T. R. Ghosh, Artificial intelligence and
blockchain technology for secure smart grid and power distribution Automation: A State-of-
the-Art Review. Sustainable Energy Technologies and Assessments, 2023.
[18] T. Liu, Y. Sun, Y. Liu, Y. Gui, Y. Zhao, D. Wang, & C. Shen, Abnormal traffic-indexed state
estimation: A cyber–physical fusion approach for Smart Grid attack detection. Future
Generation Computer Systems, 2015.
[19] P. Boopathy, M. Liyanage, N. Deepa, M. Velavali, S. Reddy, P. K. R. Maddikunta, N. Khare, T. R.
Gadekallu, W.-J. Hwang, Q.-V. Pham, Deep learning for intelligent demand response and
smart grids: A comprehensive survey. Elsevier, 2024. 15740137.
[20] H. Ma, H. Zhang, D. Tian, D. Yue, G. P. Hancke, Optimal demand response based dynamic
pricing strategy via Multi-Agent Federated Twin Delayed Deep Deterministic policy gradient
algorithm. Engineering Applications of Artificial Intelligence, Elsevier, 2024.
[21] J. Burrell, How the machine ‘thinks’: Understanding opacity in machine learning algorithms.
sage Journals: Big Data & Society, 2016. Vol. 3(No 1).
[22] P. R. Chandre, V. Vanarote, R. Patil, P. N. Mahalle, G. R. Shinde, M. Nimbalkar, J. Barot,
Explainable AI for Intrusion Prevention: A Review of Techniques and Applications, in ICT
with Intelligent Applications. 2023, Springer Link.
[23] C. Lazaro, M. Rizzi, Predictive analytics and governance: a new sociotechnical imaginary for
uncertain futures. International Journal of Law in Context, 2023.
[24] W. Shuang, Z. Le, H. Wei, Y. Rui, L. Baisi, Improved Deep Belief Network and Model
Interpretation Method for Power System Transient Stability Assessment. Journal of Modern
Power Systems and Clean Energy, 2020. Vol. 8(No. 1).
[25] C. Hart, Doing a Literature Review: Releasing the Social Science Research Imagination. Sage
London, 1998.
[26] Y. Levy, T. J. Ellis, A Systems Approach to Conduct an Effective Literature Review in Support
of Information Systems Research. Informing Science: The International Journal of an
Emerging Transdiscipline, 2006.
[27] M. C. Falvo., L. M. Luigi, S. Danilo, E. B. Enrico, Technologies for smart grids: A brief review,
in 12th International Conference on Environment and Electrical Engineering. 2013, IEEE:
Wroclaw, Poland
[28] L. Feng, Y. Zhou, Q. Luo, Y. Wei, Complex-valued artificial hummingbird algorithm for global
optimization and short-term wind speed prediction. Expert Systems with Applications,
2023.
[29] T. A. Rizzetti, P. Wessel, A. S. Rodrigues, B. M. Da Silva, R. Milbradt, & L. N. Canha, Cyber
security and communications network on SCADA systems in the context of Smart Grids, in
50th International Universities Power Engineering Conference (UPEC). 2015, IEEE: Stoke
on Trent, UK.
[30] M. El-Hawary, The Smart Grid—State-of-the-art and future trends, in Eighteenth
International Middle East Power Systems Conference (MEPCON). 2016, IEEE: Cairo, Egypt.
[31] Q. Zhang, N. Sreedhar, S. Yogesh, P. Viktor, Semantic Information Modeling for Emerging
Applications in Smart Grid, in 2012 Ninth International Conference on Information
Technology - New Generations. 2012, IEEE: Las Vegas, NV, USA.
[32] J. R. Vázquez-Canteli, N. Zoltán, Reinforcement learning for demand response: A review of
algorithms and modeling techniques. Applied Energy, Elsevier, 2019.
[33] F. A. Rahman, K. V. T. Chai, H. Basri, W. O. Sean., Application of fault tree analysis for
customer reliability assessment of a distribution power system. Reliability Engineering &
System Safety, 2013. 111: p. 76-85.
[34] E:B. Swanson, N.C. Ramiller, The organizing vision in information systems innovation.
organization science 1997. 8(5).
[35] O. Tona, S.A. Carlsson. The organizing vision of mobile business intelligence. in European
Conference on Information Systems. Association for Information Systems.
[36] J.W. Creswell, J.D. Creswell, Research design: qualitative, quantitative, and mixed methods
approaches fifth edition. . 2017: SAGE Publications, Inc.
[37] E.B. Swanson, N.C. Ramiller, Innovating mindfully with information technology. MIS
Quarterly, 2004. Vol. 28(No. 4): p. pp. 553-583.
[38] G. Walsham, Doing interpretive research. European Journal of Information Systems, 2006.
15.
[39] J.W. Creswell, J.D. Creswell, Research design: qualitative, quantitative and mixed methods
approaches 3rd Edition. SAGE. 2008, Sage London.
[40] W.J. Orlikowski, D. Robey, Information Technology and the Structuring of Organizations.
Information Systems Research, 1991. Vol. 2(No. 2).
[41] A. Crabtree, D.M. Nichols, J. O'Brien, M.Rouncefield, M.B.Twidale, Ethnomethodologically
informed ethnography and information system design. JOURNAL OF the american society
for information science., 2000. Vol. 51(No. 7): p. pp. 666 – 682.
[42] R.Srinivasan, Ethnomethodological Architectures: Information Systems Driven by Cultural
and Community Visions. Journal of the American society for information science and
technology, 2007. Vol. 58(No. 5): p. pp. 723–733.
[43] M. Crang, I. Cook, Doing ethnographies. 2007, London, Thousand Oaks, New Delhi: SAGE
Publications.
[44] V. Clarke, V. Braun, Thematic analysis. The Journal of positive psychology, 12(3), 297–298. .
2016.
[45] F. Cech, Tackling Algorithmic transparency in communal energy accounting through
participatory design. in proceedings of the 10th international conference on communities &
technologies - wicked problems in the age of tech. 2021. Vienna, Austria: ACM.
[46] S. Passi, S.J. Jackson. Trust in data science: Collaboration, translation, and accountability in
corporate data science projects. in Proceedings of the ACM on Human-computer interaction.
2018. New York, NY: ACM.
[47] S. Gregor, The nature of theory in information systems. MIS Quarterly, 2006. 30(3): p. 611-
642.