2024 International Conference on Expert Clouds and Applications (ICOECA)

Optimized Scheduling of Electric Vehicles

Charging in Smart Grid using Deep Learning
K. Paul Joshua 1* Jarabala Ranga 2 P. Venkata Prasad 3
Department of Computer Science and Research Scholar-SSU, Department of Electrical and
Engineering, Department of Electrical and Electronics Engineering, Chaitanya
SCAD College of Engineering and Electronics Engineering, Ramachandra Bharathi Institute of Technology,
Technology, College of Engineering, Hyderabad, Telangana, India.
Tirunelveli, Tamil Nadu, India Eluru, Andhra Pradesh, India [email protected]
[email protected] [email protected]
2024 International Conference on Expert Clouds and Applications (ICOECA) | 979-8-3503-8579-3/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICOECA62351.2024.00079

Balasubbareddy Mallala 4 Rajendiran M 5 Ramya Maranan 6

Department of Electrical and Department of Computer Science and Department of Research and
Electronics Engineering, Chaitanya Engineering, Panimalar Engineering Innovation, Saveetha School of
Bharathi Institute of Technology, College, Poonamallee, Engineering, SIMATS,
Hyderabad, Telangana, India Chennai, Tamil Nadu, India. Chennai, Tamil Nadu, India,
[email protected] [email protected] [email protected]

Abstract— This research presents an integrated research Network (RNN), to allot sensibly for electric charge sites
framework for electric vehicle charging infrastructures under [3], [4]. As the world's mode of transportation is changing
smart grids. It employs Genetic Algorithms and machine from about the internal-combustion engine to full
learning models such as Artificial Neural Networks (ANNs), electrification, more and more attention is directed at smart
Decision Trees, and Recurrent Neural Networks. The GA is
able to minimally stack waiting times strategically and, by the
grids and what they can do to make the different operations
effective allocation of charging points to drivers at various on charging more smoothly and intelligently [5],[6].
stages in their route, maximizes usage. In contrast, requiring The goals of the research is to reduce the waiting
intensive accurate training on massive data sets are the ML time for charging the vechicle and effectively utilizing the
methods, which can also predict future demand for efficient resources. Electric car charging and optimization is a far-
use of resources. Results of the method demonstrate the reaching topic. As a result, many scholars have tried to get
strength of these techniques, with ANNs reaching nearly 99% their result into this problem. Time of use charging, load
accuracy. By applying this comprehensive methodology, balancing, and demand response techniques are three
operational metrics such as waiting time for EV owners are
common ways to handle problems involved in effective
significantly improved and sustainable power management
becomes more feasible. These findings underscore the potential
response to electric vehicle charging requests. However,
for artificial intelligence to revolutionize energy distribution-- these solutions often hit the wall when matched up against
so that it balances consumer convenience with tighter dynamic charging patterns [7] [8] [9].
environmental targets and ways of working even now taking
shape concerning electric vehicle charging. Smart grids have turned into a crucial facilitator for
integrating EVs intensively into our traditional power
Keywords— smart grids, electric vehicles, optimization, systems [9]. By making real-time checks and
genetic algorithm, machine learning communications, smart distribution grids can change energy
I. INTRODUCTION distribution in response to the special needs of EVs.
Numerous studies show that incorporating EVs into smart
For the current transportation network with grids will help to improve conventional network balance and
more and more electric vehicles (EVs), the electrical system reduce peak loads, and optimize energy usage [10]. Genetic
should be regulated by advanced technology. Therefore Algorithms (GAs) evolving out of biological mechanisms of
smart grids play a vital role in promoting a more efficient choice can be extremely effective tools for solving complex
use of resources and in coping with the fierce problems of optimization challenges in smart grids. In order to fit the
the rapidly integrating EVs. This research represents an in- dynamic nature of EV charging requirements, they can
depth examination of how to improve EV charging by using explore a vast solution space and respond rapidly to varying
features that combine Genetic Algorithm (GA), and machine situations. Previous research has applied GA to issues such
learning in smart grid situations [1], [2]. as grid optimization and load balancing in a bid to improve
overall grid performance [11], [12].
Our main goal is to design a smart charging
schedule that reduces the risk of waiting, raises the charging Machine learning models, as well as ANN, DT and
station's effectiveness, and can be harmonized with the RNN. are increasingly being used to forecast and optimize
broader prospects of sustainable management of energy. To techniques for electric vehicle charging. ANN's ability to
the best of our knowledge, this fusion of GA and ML memorize nonlinear correlations, DT's decision-making
models is a pioneering way, such as Artificial Neural transparency, and RNN's sequential recollection ability
Network (ANN), Decision Tree (DT), and Recurrent Neural make them highly efficient tools for dealing with the

979-8-3503-8579-3/24/$31.00 ©2024 IEEE 408

DOI 10.1109/ICOECA62351.2024.00079
Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on November 02,2024 at 17:10:20 UTC from IEEE Xplore. Restrictions apply.
 problems of EV charging. Research has shown that
forecasting charging demands, effectively allocating
resources, and enhancing average charging performance
[13] and [14] have been of practical use.
The comibination of genetic algorithms with
machine learning models is a novel frontier in optimizing
EV charging. Although GA optimizes charging
requirements and reduces peak loads, machine learning
models utilize ancient knowledge to predict future rates.
Their combination provides a complete solution for the
distribution of rate points. It can also minimize public
appearances and make smart grids' EV charging smarter
overall..
Fig. 1. Proposed Methodology
II. METHODOLOGY A. Genetic Algorithm
In navigating the future trajectory of power systems, the Genetic Algorithm (GA) proudly has a central place in
integration of smart grid generation appears as a vital route, our research methodology, playing a crucial role in both
in which electric motors, especially new energy vehicles, minimising peak grid demand and optimising electric
play a pivotal role as important components inside smart vehicle (EV) charging demands. GA is a heuristic search
grids. Despite their importance, a demanding goal facing and optimization approach encouraged by way of the
smart grids relates to the proficient scheduling of electrical mechanics of herbal selection and genetics. It relies on the
vehicle charging and the efficient regulation of strength in premise of generating a population of ability answers across
the grid. In answer to this problem, our study provides a consecutive generations to converge closer to an ultimate
contemporary approach for smart grid electric powered response. In addition to individuals or chromosomes that
vechicle charging scheduling and strength management. represent capacity solutions, the GA structure consists of
This approach combines 3 strong technologies: Genetic genetic operators (crossover and mutation) that generate
Algorithm (GA), Artificial Neural Network (ANN), new answers, a health function that evaluates each
Recurrent Neural Network (RNN), and Decision Tree (DT)- individual's performance, and a choice technique that
based Machine Learning methods. Through this synergistic determines which individuals move on to the following the
integration, our solution offers series prediction, and condition.
intelligent decision-making, ultimately enhancing electric
powered vechicle charging scheduling and power In our context, the GA tackles the endeavour of
management. discovering and developing charging schedules for electric
Initially, the Genetic Algorithm takes center stage in powered autos in a manner that minimizes the stress at the
optimizing electric powered car charging needs, deliberately grid during peak intervals. By iteratively improving and
acting to decrease height grid loads. This primary stage sets choosing charging profiles, the GA helps to a more opimal
the idea for an additional simplified and effective grid distribution of electric vehicle charging demands across the
functioning. smart grid. Implementation-sensible, MATLAB functions
since the instrumental platform for implementing the
Subsequently, our Machine Learning (ML) model, including Genetic Algorithm in our investigations. MATLAB's
ANN, RNN, and DT algorithms, takes responsibility of the comprehensive optimization toolbox, paired with its
predicting the remaining time time required for charging the suppleness in dealing with sophisticated algorithms, making
vechcile. Based on the result the ML model alloates the it a smart option for the difficult optimization duties
charging point to the vechicle. This predictive feature acts as inherent in GA.
a vital component in proactively modifying the charging
schedules to the constantly shifting conditions inside the B. Machine learning model
grid. Artificial Neural Networks (ANNs) hold a vital place in our
The suggested integrated system gives an exhaustive research, providing to the predictive abilties necessary for
remedy to the issues experienced by means of smart grids in effective EV charging scheduling and power management
dealing with electric powered vechicle charging. By inside smart grids. ANNs, triggered by the human brain's
carefully integrating the powers of Genetic Algorithm and architecture, comprise of linked nodes grouped into layers.
sophisticated ML methods, our methodology not only In our methods, ANNs work as dynamic models able of
optimizes charging needs but also develops flexibility to the understanding intricate styles and interactions within the EV
constantly changing grid scenarios. This guarantees smart charging device. The entry layer receives data about
and potent charging of electric vehicles, enhancing the characteristics such previous charging patterns, grid load
overall sustainability and efficiency of smart grid circumstances, and outside influences impacting power
operations. The methodology of the prposed sytem is shown consumption. The hidden layers assist the extraction of
as flow chart in figure 1. troublesome patterns, and the output layer offers projections
for future charging demands and grid masses. Through
repeated education on past data, ANNs change their core
parameters to improve forecast accuracy [15], [16]. The
architecture of ANN is shown in figure 2.

409

Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on November 02,2024 at 17:10:20 UTC from IEEE Xplore. Restrictions apply.
 vehicle is charged. These feedback measures have
contributed greatly to developing a more sophisticated
charge optimization strategy, allowing us to allocate
resources effectively during periods of greatest demand.
Moreover, noticing the transforming potential of
machine learning in electric vehicle charging, we used the
collected data to train a machine learning model. The goal
of this model is to predict, based on historic usage and
feedback patterns, the best way to allot charging stations for
vehicles coming up. Our research seeks to use machine
learning algorithms to improve the flexibility of the
charging infrastructure, making resources more readily
available.
Fig. 2. Architecture of ANN
TABLE I. CHARGING OF VECHICLES
Followed by ANNs, Recurrent Neural Networks (RNNs)
Time Interval Number of Vehicles Charging
offer a temporal component to our investigations, enabling
7:00 AM - 8:00 AM 10
the model to forecast the sequential structure of EV
8:00 AM - 9:00 AM 15
charging data. Unlike standard feedforward networks, RNNs
possess recurrent connections that carry data from prior time 9:00 AM - 10:00 AM 8
steps, making them effective at dealing with time series 12:00 PM - 1:00 PM 5
statistics. In our study, RNNs handle the dynamic and 4:00 PM - 5:00 PM 20
developing nature of EV charging requests and grid load 6:00 PM - 7:00 PM 18
circumstances. The temporal dependencies in charging 8:00 PM - 9:00 PM 12
styles, driven by means of daily and seasonal versions, are
efficaciously recorded by means of the recurring IV. EXPERIMENTAL ANALYSIS
connections within the community. This sequential memory In this research, the Genetic Algorithm is viewed as a
allows the RNN to produce forecasts mainly based on prior strategic method of reducing vehicle waiting times at
charge data [17]–[19]. charging stations. This will enable greater use of resources.
The incorporation of Decision Tree (DT)-primarily based The aim of the research is to assign vehicles to charging
Machine Learning (ML) algorithms enhance interpretability stations in an intelligent manner, so as to reduce idle time
and performance in our study technique. Decision trees and make better use of resources. By using genetic
enhances the human decision-making procedures with the algorithms the system becomes an intelligent decision
assistance of recursively splitting data into subsets based at maker, making the smartest possible assignments. By taking
the most comprehensive functions. This hierarchical form into account data like total charge completed and desired
provides a clear display of selection routes, important for finish time for any specific car, it predicts time remaining
studying the elements impacting EV charging and grid load until the battery is fully charged. This input makes the
scenarios. In our investigations, DT-based ML algorithms model more insightful and over time it has been able to
help to improving EV charging schedules using the selection learn from and analyze patterns in the behavior of cars at
criteria from past data. The collection of rules carefully charging bays. Exposing the algorithm to a varied set which
branches and splits the information, delivering practical includes many different charging scenarios is also a kind of
insights for the greatest quality scheduling of EV charging training. It then adjusts its criteria by identifying patterns in
within the smart grid environment [20], [21]. the closing time or completions of cars. This preparation is
necessary; it wants to better anticipate charging stations
while a long sentence follows regarding the complex
III. REAL TIME DATA COLLECTION FOR MODEL EVALUATION training process in which the Genetic Algorithm was
We installed an algorithm that comes with six exposed to variations of sets denoting different charging
charging docks. It will be able to fuel electric vehicles scenarios in order to perfect its decision-making criteria.
continuously through the whole day. In order to gauge the This meant learning from inherent trends in completion
actual constraints and changes in demand, our analysis status as well as time closed. In the long term, only by doing
centers heavily on peak charging demands, in the mornings this effectively can it improve the ability to foresee and
and evenings. From the wee hours of the morning until distribute charging stations.
dusk, Table 1 provides data pertaining to when there are
active vehicles on the charging system. Our empirical data Once trained, genetic algorithms are very good at
provides a great groundwork for the study, divining into performing real-time evaluations of charging capacity in
charging dynamics of this station in depth. cars coming in. Based on factors like charging and time
taken, the genetic algorithm optimizes the distribution of
Taking things further, in the years ahead we plan to points in order to minimize waiting time, making charging
utilize these measures to improve the effective use of as smooth and productive as possible from start to finish. As
charging resources. We have designed assessment seen in the data in Table II, this scheme significantly
procedures that capture key indices. Generally, you input decreased the waiting time of cars being charged both
data on the charging interval and duration each time a before and after the genetic algorithm implementation. The

410

Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on November 02,2024 at 17:10:20 UTC from IEEE Xplore. Restrictions apply.
 gains in efficiency were substantial: rather than having to 100 Precision (%)
Recall (%)
wait for anywhere from twelve-to-twenty-five minutes with F1 Score (%)
ordinary means. Before we used the genetic algorithm, Accuracy (%)
waiting times at the station for cars to be charged during

Performance score
peak hours varied from 12 minutes to 25 minutes.
Subsequent to the genetic algorithm, waiting times
registered a significant decrease from eight minutes to
fifteen minutes. This marks a substantial improvement in
functionality, as charging stall location is decided
automatically for electric vehicles. There were waiting time
drops of four to ten minutes in every interval, which shows 80
that the genetic algorithm for speeding up charging This not ANN DT RNN

only enhances customer satisfaction but also proves that Model

advanced algorithms hold great potential for optimizing

Fig. 3. Performance of Each Model
electric vehicle charging infrastructure.
Figure 3 represent the performance of machine learning
TABLE II REDUCTION IN WAITING TIME AFTER IMPLEMENTING models predicting future charge points. The ANN rates of
GA precision, recall, and F1 scores are 90.0, 98.8, and 98.7
respectively, with an accuracy of 98.7%. The accuracy of
Time Average Average Reduction in
Interval Waiting Waiting Waiting the DT model, for predicting charge points, is 93.2%, and
Time Time (After Time min has a precision of 93.5%, a recall of 92.9%, and an F1 of
(Before GA) GA) min 93.2%. The RNN had an accuracy of %91.2 and precision,
min recall, and F1 score stood at 91.0% 91.5%, and 91.2%. The
7:00 AM - 15 10 5 research reveals precision is critical for going after charge
8:00 AM points by these models.
8:00 AM - 20 12 8
9:00 AM V. CONCLUSION
9:00 AM - 18 11 7 In conclusion, this study proposed a novel structure
10:00 AM incorporating GA into charging point time scheduling to
12:00 PM 12 8 4 make it more efficient, and also introduced the ANN, DT,
- 1:00 PM and RNN methods for intelligent charging point allocation.
4:00 PM - 25 15 10 The modified GA is charged to the needs of the moment,
5:00 PM decreasing wait times for charging; the charging points will
6:00 PM - 22 13 9 be extended. Of these models, ANN achieved the best
7:00 PM
performance with an accuracy rate of 98.7%. These results
8:00 PM - 17 9 8
are exceptional in that the cooperative allocation of
9:00 PM
resources optimally enhance the operational effectiveness of
the charging point and ensured that EV drivers do not have
To enhance the prediction capability of the proposed to wait. The obtained output shows how the evolution of
technique, ML models were applied, as previously computation can be combined with system analysis in
described. These models were trained using the dataset, predictive power usage models. Flexibility and precision
covering critical factors such as the amount of charge may be important in administrating a smart grid. In a word,
finished and remaining time for each vehicle. Following the these models also have the merit of getting the goods
training phase, extensive testing was done to assess the through cars to take maximum advantage of energy. With
prediction accuracy of each model in predicting future the development of the times, this type of discovery offers
charge points for optimum vehicle allocation. The ANN additional evidence boding well for future developments of
model displayed extraordinary precision, obtaining an smart grid technology and brings us more directly to electric
accuracy rate of 98.7%. This underlines the model's vehicle charging with advanced algorithms. Essentially, the
capability in finding the charging time. The DT model combination of GA and machine learning methods enhance
shows the accuracy at 93.2%, The RNN model, with an the energy efficiency in charging performance and overall
accuracy of 91.2%, predicts the responses. operational efficiency.
REFERENCES

[1] Z. S. Warraich and W. G. Morsi, “Early detection of cyber–

physical attacks on fast charging stations using machine learning
considering vehicle-to-grid operation in microgrids,” Sustainable
Energy, Grids and Networks, vol. 34, p. 101027, 2023, doi:
10.1016/j.segan.2023.101027.
[2] Y. Qiu et al., “Two-stage distributionally robust optimization-

411

Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on November 02,2024 at 17:10:20 UTC from IEEE Xplore. Restrictions apply.
 based coordinated scheduling of integrated energy system with [12] E. Ileberi, Y. Sun, and Z. Wang, “A machine learning based
electricity-hydrogen hybrid energy storage,” Protection and credit card fraud detection using the GA algorithm for feature
Control of Modern Power Systems, vol. 8, no. 1, 2023, doi: selection,” Journal of Big Data, vol. 9, no. 1, 2022, doi:
10.1186/s41601-023-00308-8. 10.1186/s40537-022-00573-8.
[3] E. Paffumi, M. De Gennaro, and G. Martini, “Innovative [13] L. Jing, Q. Niu, D. Zhan, S. Li, and W. Yue, “Predictive
Technologies for Smart Cities: Towards Customer Driven Modeling of Surface Roughness and Feed Force in Al-50wt% Si
Infrastructure Design for Large Scale Deployment of Electric Alloy Milling Based on Response Surface Method and Various
Vehicles and Vehicle-to-Grid Applications,” Transportation Optimal Algorithms,” Arabian Journal for Science and
Research Procedia, vol. 14, pp. 4505–4514, 2016, doi: Engineering, 2022, doi: 10.1007/s13369-022-07114-8.
10.1016/j.trpro.2016.05.373. [14] T. Banerjee, S. Dey, A. P. Sekhar, S. Datta, and D. Das, “Design
[4] A. Rejeb et al., “Unleashing the power of internet of things and of Alumina Reinforced Aluminium Alloy Composites with
blockchain: A comprehensive analysis and future directions,” Improved Tribo-Mechanical Properties: A Machine Learning
Internet of Things and Cyber-Physical Systems, vol. 4, no. May Approach,” Transactions of the Indian Institute of Metals, vol.
2023, pp. 1–18, 2023, doi: 10.1016/j.iotcps.2023.06.003. 73, no. 12, pp. 3059–3069, 2020, doi: 10.1007/s12666-020-
[5] K. Koteish, H. Harb, M. Dbouk, C. Zaki, and C. Abou Jaoude, 02108-2.
“AGRO: A smart sensing and decision-making mechanism for [15] M. P. Sarr, A. Thiam, and B. Dieng, “ANFIS and ANN models to
real-time agriculture monitoring,” Journal of King Saud predict heliostat tracking errors,” Heliyon, vol. 9, no. 1, p.
University - Computer and Information Sciences, vol. 34, no. 9, e12804, 2023, doi: 10.1016/j.heliyon.2023.e12804.
pp. 7059–7069, 2022, doi: 10.1016/j.jksuci.2022.06.017. [16] A. Zendehboudi, X. Li, and B. Wang, “Utilisation des modèles
[6] A. V. Jha et al., “Smart grid cyber-physical systems: ANN et ANFIS pour prédire un compresseur à spirale à vitesse
communication technologies, standards and challenges,” Wireless variable avec injection de vapeur,” International Journal of
Networks, vol. 27, no. 4, pp. 2595–2613, 2021, doi: Refrigeration, vol. 74, pp. 473–485, 2017, doi:
10.1007/s11276-021-02579-1. 10.1016/j.ijrefrig.2016.11.011.
[7] L. Zjavka, “Power quality 24-hour prediction using differential, [17] D. Kotios, G. Makridis, G. Fatouros, and D. Kyriazis, “Deep
deep and statistics machine learning based on weather data in an learning enhancing banking services: a hybrid transaction
off-grid,” Journal of the Franklin Institute, no. xxxx, pp. 1–25, classification and cash flow prediction approach,” Journal of Big
2022, doi: 10.1016/j.jfranklin.2022.06.048. Data, vol. 9, no. 1, 2022, doi: 10.1186/s40537-022-00651-x.
[8] N. E. H. Ben Youssef, Y. Barouni, S. Khalfallah, J. B. H. Slama, [18] M. Nabipour, P. Nayyeri, H. Jabani, A. Mosavi, E. Salwana, and
and K. Ben Driss, “Evaluating Content-centric Communication S. Shahab, “Deep learning for stock market prediction,” Entropy,
over Power Line Communication Infrastructure for Smart Grids,” vol. 22, no. 8, 2020, doi: 10.3390/E22080840.
Procedia Computer Science, vol. 73, no. Awict, pp. 217–225, [19] A. Moghar and M. Hamiche, “Stock Market Prediction Using
2015, doi: 10.1016/j.procs.2015.12.015. LSTM Recurrent Neural Network,” Procedia Computer Science,
[9] R. Verma, “Smart City Healthcare Cyber Physical System: vol. 170, pp. 1168–1173, 2020, doi: 10.1016/j.procs.2020.03.049.
Characteristics, Technologies and Challenges,” Wireless [20] F. Tao, Q. Qi, L. Wang, and A. Y. C. Nee, “Digital Twins and
Personal Communications, vol. 122, no. 2, pp. 1413–1433, 2022, Cyber–Physical Systems toward Smart Manufacturing and
doi: 10.1007/s11277-021-08955-6. Industry 4.0: Correlation and Comparison,” Engineering, vol. 5,
[10] Sivaramkrishnan, M., Ramkumar, M. S., Subramanian S, S., & no. 4, pp. 653–661, 2019, doi: 10.1016/j.eng.2019.01.014.
Ladu, N. S. D. (2022). A Bridgeless LUO Converter with [21] Julia Sebastian and S. Jhancy Mary, “Structural, Thermal and
Glowworm Swarm Optimized Tuned PI Controller for Electrical Electrochemical Behavior of Poly(2-ethylaniline)-
Applications. Mathematical Problems in Engineering, 2022. nanocomposite-Fe2O3 and Poly(2-ethylaniline)-nanocomposite-
[11] K. E. Mary Reena, A. Theckethil Mathew, and L. Jacob, “An SiO2 for Antibacterial and Antioxidant Studies,” Polymer
Occupancy Based Cyber-Physical System Design for Smart Science - Series B, vol. 64, no. 3, pp. 340–353, 2022, doi:
Building Automation,” Mathematical Problems in Engineering, 10.1134/S1560090422200040.
vol. 2015, 2015, doi: 10.1155/2015/132182.

412

Authorized licensed use limited to: Motilal Nehru National Institute of Technology. Downloaded on November 02,2024 at 17:10:20 UTC from IEEE Xplore. Restrictions apply.