See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/370978369

Data-Driven Analysis of EV Energy Prediction and Planning of EV Charging

Infrastructure

Conference Paper · May 2023

DOI: 10.1109/BigDataService58306.2023.00009

CITATIONS READS
3 1,443

5 authors, including:

Jerry Gao
San Jose State University
352 PUBLICATIONS 5,880 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jerry Gao on 23 May 2023.

The user has requested enhancement of the downloaded file.

 Data-Driven Analysis of EV Energy Prediction and
Planning of EV Charging Infrastructure

Arul Palaniappan, Purnima Bhukya, Sai Kiran Chitti,Vinupriya Sanjay Kumar Jerry Gao
Department of Applied Data Science Department of Applied Data Science,
San Jose State University Department of Computer Engineering
San Jose, United States San Jose State University
(arul.palaniappan, purnima.bhukya, saikiran.chitti,vinupriya.sanjaykumar)@sjsu.edu San Jose, United States
Corresponding Author:
[email protected]

Abstract—To reduce the current crisis on energy and environ- emissions come from road vehicles. The biggest way to solve
mental problems, the adoption of electric vehicles is an effective this specific problem is for consumers to start buying electric
way. But limited access and availability of charging infrastructure vehicles (EVs) because EVs are battery powered and do not
are restricting the EV market development. This paper presents
data-driven analysis of current EV charging infrastructure to emit any GHGs when generating energy to move.
plan future EV infrastructure in the city. This paper tackles However, there are a few problems that act as barriers for
the problem using three approaches. The first approach that is the growth of electric vehicles. One of the main problems
covered in this paper is to predict the number of EV charging that must be addressed as people start switching to EVs is
stations that need to be implemented in a geographic location.
This is done using the SARIMAX model on the California EV to implement better EV infrastructure in the cities and near
charging station dataset that is preprocessed to be split into zip highways, because people need to charge their vehicle in the
codes. The second approach that is tackled by this paper is the middle of a long trip just as they would be refueling a gasoline
energy consumption prediction by an EV charging station. This powered vehicle. A positive to EVs is that most of them are
information is useful to allow planners to understand how much sold with a Level 1 charger that can be plugged into an outlet
energy is consumed by an EV station across multiple different
time periods. An optimized Meta-Fusion regression model based at home, so some people can use this to charge at home. But
on a stacked model is proposed for this approach and is applied this is the slowest method of charging and is not useful for
to the Palo Alto charging station dataset. The third approach long road trips. Another problem with charging EVs is that
utilized the K-means clustering algorithm on the Palo Alto it takes much more time to charge an electric vehicle to full
charging station dataset to predict the optimal location for EV capacity than it takes a gasoline-powered vehicle to refuel at
charging stations in that city. The SARIMAX mode had an R2
score of 0.97, a MAE of 6.50, and RMSE of 17.43 when predicting a gas station. There are currently three types of EV chargers,
EV station counts in the zip code 92101. The stacked model had and the following table shows the features of each type. The
an R2 score of 0.98, a MAE of 0.91 and an RMSE of 2.26 when variations in the times listed in the table below are caused by
predicting the daily energy consumption by an EV station. the different battery types used in each vehicle. In our project,
Keywords—Big Data, EV Charging Infrastructure, EV Charg- we are primarily focusing on Level 2 and DC fast charging
ing, Data-Driven Analysis, EV Electricity Prediction, Machine stations because those are the fastest type of chargers available.
Learning.
TABLE I
I. I NTRODUCTION OVERVIEW OF E LECTRIC C HARGES (USDOT)

Climate change is one of the most pressing issues in the Type Level 1 Level 2 Level 3(DC Fast)
world today. This issue is mainly due to the large amounts Power 1 kW 7 - 19 kW 50 - 350 kW
of carbon dioxide and other greenhouse gasses (GHG) being
Time Taken to
let into the air. According to the NIEHS [1], air pollution 20 - 60 min
Charge a Battery 40 - 50 hours 4 -10 hrs
causes over 6.5 million deaths each year globally. According (0 to 80% cap)
from 0% Capacity
to the WHO, this is because 99% of the population in the
Range per Hour
world is breathing in air that exceeds their limits on breathable 2 - 5 miles 10 - 20 miles 180 - 240 miles
of Charge
air. One of the major contributors of GHG is emissions from
Home ,
road transportation. Road transportation uses a lot of fossil Typical Locations
Home Office , Public
fuels and emits GHG when the fuels are converted into energy. Placed
According to Gurwa et. al (2019)[2], 27% of the world’s GHG Public
The above problems motivated us to come up with a solution our paper, a stacked model which combines multiple different
for the planning of EV infrastructure with an AI powered ensemble models, including the Random Forest and XGBoost
model using Machine Learning and Deep Learning Tech- is introduced and compared among the standard models.
niques. The main objective of this research is to create a Erotokritos Xydas et.al (2015)[7] used EV charging event
model that can optimally predict where EV infrastructure can data, EV charging station data and weather data. First, they
be placed in a city. The model will take into consideration did some manual data preprocessing to split the data into a
various factors related to EV charging such as the location of monthly and an hourly time series. They also combined the
current stations, EV energy consumption. weather data into some of the EV data into a daily time
This research also considered the weather data to understand series. Each of the series was sent to a different module
how EV charging demand is affected by weather. The optimal for data mining. The correlation model finds the correlation
EV infrastructure network can cater to the charging needs and between weather data and peak daily usage, the regression
reduce the life cycle costs of the transportation system. Recent model finds the patterns for monthly charging demands, and
advancement in machine learning helps analyze present data the clustering model finds the charging demands based on
and come up with efficient solutions that solve the planning geographic location. All three modules are fed into a fuzzy
of charging infrastructure in a city. It can help in planning and characterization model to combine each of the series into a
management such as prediction of location of new charging charging demand risk level. The problem with this paper is that
infrastructure, demand for charging and scheduling. Hence, they didn’t use multiple models to evaluate the performance.
knowledge of machine learning models and AI is a key In this paper, a stacked model, which combines multiple
solution for enhancing the sustainability of EV transportation. models is introduced and is compared to the individual models
Our project can be used for a few applications. We expect separately to evaluate which model was better and confirmed
other cities around the world to expand on our model and that stacked model performs better.
use it to build new EV infrastructure in their cities. Another Table II here gives a list of papers we researched, and it
application that can be created with the help of our model is includes their objectives, Region, different types of datasets
a mobile application that can be used to help owners to get they used, models, and evaluation metrics.
the various charging options when they don’t have access to
charging stations.
B. Technology Survey
This paper is divided into the following sections. Section II 1) Current Live Technologies: There are few companies
describes a literature survey of previous related work. Section that have created a software to plan EV charging infrastructure
III provides an overview of the technical survey and prior art. around various places in the United States. Almost all these
Section IV gives an overview of the Methodology, which is software are not available to the public and are proprietary. We
subdivided into Dataset, Data Preprocessing which is again are not able to get information on how they are implemented.
subdivided as Data Cleaning and Data Merging. Section V The National Renewable Energy Laboratory (NREL) created
discusses the machine learning model used in the project for one of these tools called Electric Vehicle Infrastructure –
load forecasting, forecasting of energy consumption and loca- Projection Tool (EVI-Pro)[28].
tion recommendation. Section VI details the results obtained
by the machine learning model. Section VII describes future The website states that users can use the software to analyze
improvements to the project and concludes with a summary travel patterns, charging demand, and design the infrastructure
of the project. capable of meeting demand. This tool was used to help analyze
the requirements of EV infrastructure planning in the entire
II. R ELATED W ORK state of California. The US Department of Energy has a tool
A. Literature Survey called EVI-Pro Lite which is based on EVI-Pro and is available
In this section, we have reviewed many papers related to for public use[29]. This is a web tool that allows users to plan
Electric vehicles, The EV charging infrastructure, types of the number of charging stations required in a city or state.
EV chargers to understand how previous research has been This tool has few pre-selected urban areas. Once you select
conducted in this area and how to proceed with our research. the location, then you enter in the number of vehicles that you
Mariz B et.al(2016)[6] analyzed the characteristics of EV want to support there, and then click on the Calculate button.
charging demand and identified the factors that affect it, such It will do some calculations and show you the number of Level
as weather conditions, time of day, and location. They use data 2 and Level 3 charging stations that need to be implemented
preprocessing to group their traffic data into clusters and then in the location. There are some parameters related to the
plot the traffic volumes across the time of day for each cluster. amount and types of EVs to support in the location that can be
Then, they used relational analysis to find the relationship modified on the results page before recalculating the results.
between the features in the weather dataset. Finally, they used We also did some research on a couple of methods that are
a decision tree to combine all these things to plot the graphs of used to plan EV infrastructure in a region which are listed
charging demand in a day. The problem with their data is that in Table III. The first paper talks about methods used to
they didn’t use different models to compare their results. In plan for locating where future EV infrastructure would go.
 TABLE II
C OMPARISION OF E XISTING R ESEARCH PAPERS

Ref Objective Region Dataset ML Models Deep Learning Other Models Eval Metrics
Calculating the daily charging Forecasting model with
Traffic
[6] demand of EV stations South Korea N N big data technologies Compared to standard
Weather models
based on weather and traffic data P-median
Calculate the energy demand
risk level of an EV charging station Charging Event
[7] United Clustering N N Corealation
wrt to Charging event and Kingdom Charging Station Coefficient
station data.
Include supply and demand EV ranges
Set and Max Covering
[8] constraints when planning new Beijing Population N N Best solution satisfy-
P-median ing restrictions
EV charging stations Govt budget constraints
Using allocation objectives Location- Compared to standard
[9] Hong Kong EV usage rate survey N N
to plan EV infrastructure Allocation model
RMSE
Spatial and temporal information Spatial
[10] Dundee, Ireland EV infrastructure N N MAE
to predict EV station availability Temporal CGN
R2̂
ELaadNL
Predictive model for estimating LinReg RMSE
[11] Netherlands Google Maps N N
the EV infrastructure usage XGBoost MAE
Nokia HERE
Accuracy
EVnetNL LogReg
Evaluate factors affecting EV Precision
[12] Netherlands OpenGIS Rand Forest N N
infrastructure popularity Sensitivity
EV Charging Records GBRT
MCC
Predict energy demand for next KNN Pattern Sequence
[13] UCLA EV Charging Records N Accuracy
24 hours at each charging outlet ARIMA Forecasting
EV Infrastructure planning based Shanghai International Mixed- Best model
[14] Shanghai N N
on user travel data Motor City Integer Nonlinear given constraints
CEC Rand Forest Precision
Bridge the gap for EV access
[15] Orange County Energy Dept Multi-Log N N Recall
between the wealthy and the poor
U.S. Census Regression F1-score
EMD- MAE
Predict charging demand and EV
[16] Georgia Tech Charging station usage N AOA- N RMSE
infrastructure placement
LSTM Accuracy
Data-driven long-term EV Traffic Flow
[17] Sydney N CNN Cournot Competition Model
infrastructure planning Distribution Network
Models classifying EV charging Charging station BERT Accuracy
[18] United States N N
user reviews user reviews XLNet F1 score
RNN MSE
Predicting charging duration at Rabat,
[19] Real-time EV station N LSTM N RMSE
public chargers Morocco
GRU MAE
LinReg RMSE
[20] Predicting PEV charging demand Nebraska Level 2 EV stations GBRT N N MAE
SVM R2̂
Infrastructure planning for
[21] Vienna Taxi dispatch center N N Decision Support System
electric taxi vehicles
Accuracy
LogReg
Describe the urban context and Precision
[22] Netherlands EVnetNL Rand Forest N N
activities near EV infrastructure Sensitivity
GBRT
MCC
Minimize the total charging cost LTA CSPP Most EVs
[23] Singapore N N
of EVs Stats Dept Nash Transfer Fit Budget
CG - SMV RMSE
Comparing 3 ML models to Agartala,
[24] Traffic Flow Bagged and N N MAE
understand EV charging demand India
Boosted Tree R2̂
Precision
Predict charging station Logistic
[25] Italy EV Charge Transactions N N Recall
availability using ML Big Data Regression
F1 score
Ensemble RMSE
Predicting charging times for
[26] Japan Auto Research Institute Rand Forest N N MAE
different EV charging modes
Grad Boost R2̂
Mixed-
EV planning from the perspective Transport & Highest investor
[27] China Integer NLP N N
of private investor Sioux-Fall Networks profit
Huff model
 TABLE III
A PPROACHES TO PLAN FUTURE EV INFRASTRUCTURE

Ref Objective Methods Description Negatives

Locates the charging points/ stations Does not always leads to the
[30] Methods used for Node-based approach to meet the area’s overall demand at best coverage
locating charging
infrastructure the given instance. with the minimal cost
Based on vehicle paths given origin
Approach is worse when predicting slow
Path-based approach and destination of trips. Have to make a
charging stations
stop in between trips for charging.
More likely to reflect the real charging
Requires detailed individual travel data
Tour-based approach needs of EV users. Take into account the
and high computing capacity
whole journey of EV.
Easy to track bus location and less chance
Routes are long and buses
[31] Planning EV Infras- Buses only of excess charge being consumed because
tructure for different must charge during down time
types of vehicles of their fixed routes
Converting taxis to be fully EVs makes
Taxis only the most sense because they use a lot of Must charge between trips Losing money if
charging time is too long
fuel to mostly travel short distances
Most realistic method because it includes Requires lot of data - battery capacities,
All Private Vehicles
data from all vehicles that take part in traffic charging behaviors, etc. of each EV
Initially assume that location can
Size of station influences station
accommodate infinite stations.
Maximize Charging Infinite size shrinks to demand-based location - incorrect assumption at start
Infrastructure Once location has been found, shrink
Utilization will lead to improper planning
station size according to demand
Some models add a constraint of station
Station size constraint provided at start size to better serve areas of high charging Charging station size constraint at the start
doesn’t allow for flexibility of station location
demand concentration with multiple stations.
Important to choose station power levels
before modeling. Must take into account grid
Power sizing None provided
capacity at location of charging points to avoid
overloading

It describes the three different approaches used for locating Electric vehicles, and the amount of Energy that is consumed
charging infrastructure. The second paper talks about the EV by each charging transaction. More information about these
infrastructure planning with respect to the type of vehicles and datasets can be found in the table below.
how to utilize the charging to its maximum level.
TABLE IV
III. DATA E NGINEERING DATASET C OLLECTION
A. Data Collection San Jose Vehicle Palo Alto
The primary focus of this research is on the state of Dataset EV Charging population EV Charging
California, specifically in the cities and zip codes of San Jose stations data - zip codes Station Usage data
and Palo Alto. We have decided to use three different datasets Data type csv xlsx csv
to solve our problems. The first dataset is from the Alternative Size 15.7 MB 4.4 MB 13.3MB
Fuels Data Center website from the US Department of Energy. No of rows 14444 172582 259,443

This dataset contains information about charging stations and Yearly Y Y Y

their locations throughout the state of California, such as Weekly N N Y
latitude, longitude, address, and charger type. The second Monthly N N Y
dataset is the vehicle population dataset by zip code that we Duration 2010 -2022 2010-2021 Real-Time
obtained from the California Energy commission. We used this
dataset to enhance the first dataset in order to understand how
the number of electric vehicles on the road affects the number The dataset is of real-time data and is updated with the latest
of EV charging stations available in the zip code. The third information. The data has been extracted through api format
dataset that we obtained is from the Palo Alto government and and the dataset has been downloaded in CSV format.This
it contains detailed data for a few EV charging stations in Palo dataset has been geocoded by the latitude and longitude details
Alto, including charging transactions, the plug-in time of the of the charging station details.
B. Dataset Preprocessing
Data preprocessing is an important step which involves
cleaning and organizing the raw data collected so that it is Fig. 2. Sarimax Model Equation
more suitable for feeding into the model[34].
a) Data Cleaning: This step takes care of dropping null
and unnecessary data, handling missing values, renaming the
columns, filtering the data with respect to California and
finally processing the datasets.
b) Data Merging: In this stage, all the datasets are
merged to the EV station dataset based on various parameters.
The vehicle registration data is merged based on the zip code.
EV population dataset is merged based on the zip code and Fig. 3. Sarimax Model
opening year. After merging all the datasets, the final dataset
is processed, and it consists of 32 columns and 10k rows. B. Meta Fusion Regression Model
The below pie chart here represents the percentage of It is an integrated model [36] which is based on the stacked
number of features in every dataset which can be used for model and consists of the traditional machine learning models
model building. like Random Forest Regression, XGBoost and Decision Tree
Regression models. The input data is given to the regressor
model, which predicts the energy consumption as the output.

Fig. 1. Number of features in refined datasets

Fig. 4. Meta Fusion Regressor Model
IV. M ACHINE L EARNING M ODELS
A. SARIMAX C. K - Means Clustering Recommendation model
The SARIMAX model is an extension of the ARIMA model
The research paper presents a data-driven approach for the
that includes seasonality. When patterns are not consistent but
planning of electric vehicle (EV) charging infrastructure in a
occur periodically, there is a seasonality in the series. There
smart city. The approach uses K-Means clustering [37] and
are parallel time series that are not modeled directly but can be
geographical visualization tools to identify clusters of high
used as a weighted input known as an exogenous variable. The
EV traffic and low charging station density and calculates the
disadvantage of the ARIMA model is that it cannot include
number of charging stations needed in each cluster based on
seasonality and exogenous variables. By including an addi-
historical charging data. The results are used to optimize the
tional set of autoregressive and moving average components a
deployment of charging stations in the identified clusters.
SARIMA model is derived. There are additional lags from the
offset by the frequency of seasonality. The SARIMAX model
is an updated version of the ARIMA model [35].
The SARIMA model has the following four orders: a
seasonal autoregressive order (P), an order of seasonal integra-
tion(D), a seasonal moving average order (Q), and a seasonal
order, which is the length of the cycle of repeating frequency.
The Seasonal order is given by four numbers: the seasonal
AR specification, the seasonal Integration order, the seasonal
MA, and the seasonal periodicity. SARIMAX is a variation of
SARIMA where it considers the exogenous variable, which is
the external data of forecast. The following figure shows the Fig. 5. K Means Clustering
equation of the SARIMAX model.
 V. R ESULTS AND D ISCUSSION individual charging stations to predict its energy consumption
A. EV charging station count prediction because we do not want data from different stations to interfere
with our models. For this task we introduce our new model
Since the load dataset contains monthly data and they
called Meta Fusion Regressor which combines the Random
contain information based on zip code, we must separate the
Forest, XGBoost, Decision Tree and other traditional Machine
data by zip code and then run the models on each zip code.
Learning models.
We decided to run the models on the top 10 most populous zip
We ran the model on different time periods - monthly,
codes to understand how the models work. For all the models,
weekly and daily - to see the differences between the model
we split each zip code dataset into 80% training data and 20%
accuracies and the energy consumption in each period. For
test data. The following figure shows a visualization of how
this prediction, we also split the dataset by using the earliest
we split the original dataset into test and training sets. We used
80% of the data for one station as the training dataset and the
the SARIMAX model to predict the load for each zip code.
remaining 20% as the test dataset. The following figure shows
This model is one of the most complex models yet is the one
how we split the dataset after we prepared it for weekly data.
with the best prediction accuracy.
For this model, we used the Auto Arima function from the
“pmdarima” library to predict the values of the parameters
that we would use in the model. The result of the Auto Arima
function is shown in the following figure. We plugged in the
parameters into our SARIMAX model to run it and predict
the test data.

Fig. 7. Splitting of train and test dataset

The dataset has a feature, which is the amount of charging
time that each transaction takes. This feature is helpful in
predicting the amount of energy used because as the time
taken to charge the electric vehicle increases, the more energy
is consumed from the grid. The below figures represent the
Fig. 6. Test and Prediction value results for 92101 Zipcode
test and prediction results of Meta Fusion Regressor on daily,
weekly and monthly data.
The above graph shows the test and prediction values using
this model and the table below shows the results of evaluating
the model.

TABLE V
E VALUATION OF EV CHARGING STATION COUNT PREDICTION

Model RMSE MAE Rˆ2

SARIMAX 17.43 6.50 0.967

B. Predicting EV Energy Usage

This task mainly focuses on how much energy an EV
Fig. 8. Test and Prediction data results on Daily Basis
charging station consumes. This is important because we need Table VI has the evaluation metrics for each of the models
to advise our users to understand the consequences of building implemented across both time periods.
a new charging station. One of the major considerations when
building new EV stations is to figure out how much energy C. Optimal EV Charging station Location Prediction
is being consumed from the power grid. If an area of town is The final problem that we performed is to give a recommen-
already consuming too much electricity, we would recommend dation of the new optimal EV charging station’s locations. This
that they should build their station at a different location. We is important because as there is an increase in the adoption for
used the Palo Alto EV station dataset for this problem. The electric vehicles, there should be an increase in the number
dataset also has a feature, which is the amount of charging of EV charging stations to meet the EV charging demand.
time that each transaction takes. We must run our models on We have taken the historical EV charging station data of Palo
 TABLE VI
M ODEL COMPARISION FOR P REDICTING THE EV E NERGY U SAGE

Timeframe Model R2̂ Score MAE RMSE

Meta Fusion
Daily 0.98 0.91 2.26
Regressor
Random
0.98 0.99 2.47
Forest
XG Boost 0.98 0.81 2.65
Decision Tree 0.97 0.90 3.34
Meta Fusion
Weekly 0.95 14.30 20.59
Regressor
Random
0.89 18.93 30.58
Forest
Fig. 9. Test and Prediction data results on Weekly Basis XG Boost 0.93 16.06 24.14
Decision Tree 0.87 20.58 34.29
Meta Fusion
Monthly 0.97 46.39 62.10
Regressor
Random
0.68 145.54 225.60
Forest
XG Boost 0.87 88.10 140.86
Decision Tree 0.14 245.24 373.08

VI. C ONCLUSION
This paper presents three different approaches to address the
challenges of planning new EV charging station infrastructure.
The first challenge that was tackled is predicting the number of
EV charging stations that should be placed in a particular zip
code in the next three years. This was done by calculating the
Fig. 10. Test and Prediction data results on Monthly Basis historical number of EV charging stations in the zip code and
then using the SARIMAX model to forecast the future counts.
The second challenge which was addressed is predicting how
Alto and applied the K-means clustering algorithm to identify much energy (kWh) is consumed by a charging station in
the clusters of high EV traffic and low EV charging station daily, weekly, and monthly time periods based on different
density. weather conditions. The proposed stacked model, called Meta
The centroids of each individual cluster are identified, and Fusion Regression, provided the best prediction results. The
the count of EV charging stations needed is calculated by third challenge utilized the K-means clustering algorithm to
dividing the total EV charging time by the number of unique predict the optimal locations for EV charging stations.
EV charging stations identified in that cluster. The obtained Overall, the above approaches helped to provide valuable
number is then rounded up to the nearest integer. The results information in planning new EV charging infrastructure in a
obtained are plotted over a Palo-Alto centered map which was zip code. In the future, live traffic data can be incorporated
created using the Folium library. into the approaches in order to provide better understanding
As shown in the below figure, the current EV charging of how drivers behave when they want to charge their electric
stations are displayed using green icons, and the predicted vehicles.
EV charging stations are displayed using red icons. R EFERENCES
[1] International Symposium “Transforming Mobility for Humanity” Com-
memorating the 50th Anniversary of The Club of Rome “The Limits to
Growth” 14 December 2022 Venue: Chulalongkorn University, Bangkok
/ Online
[2] N. Garwa and K. R. Niazi, “‘impact of EV on integration with grid
system – A review,’” in 2019 8th International Conference on Power
Systems(ICPS), 2019, pp. 1–6
[3] “Electric vehicle charging speeds,” Transportation.gov. [Online]. Avail-
able: https://www.transportation.gov/rural/ev/toolkit/ev-basics/charging-
speeds.[Accessed: 22-Oct-2022].
[4] “The current state of EV battery swapping,” Nio.com. [Online]. Avail-
able: https://www.nio.com/blog/current-state-ev-battery-swapping. [Ac-
cessed: 19-Oct-2022].
[5] N. I. Nimalsiri, E. L. Ratnam, C. P. Mediwaththe, D. B. Smith, and S. K.
Halgamuge, “Coordinated charging and discharging control of electric
Fig. 11. Optimal EV Charging station Location Prediction vehicles to manage supply voltages in distribution networks: Assessing
the customer benefit,” Appl. Energy, vol. 291, no. 116857, p. 116857,
2021.
 [6] Mariz B. Arias, Sungwoo Bae, Electric vehicle charging de- Electrical Engineering and 2021 IEEE Industrial and Commercial Pow-
mand forecasting model based on big data technologies, Ap- erSystems Europe (EEEIC / ICPS Europe), 2021.
plied Energy, Volume 183, 2016, Pages 327-339, ISSN 0306-2619, [26] I. Ullah, K. Liu, T. Yamamoto, M. Zahid, and A. Jamal, “Prediction of
https://doi.org/10.1016/j.apenergy.2016.08.080. electric vehicle charging duration time using ensemble machine learning
[7] Erotokritos Xydas, Charalampos Marmaras, Liana M. Cipcigan, Nick algorithm and Shapley additive explanations,” Int. J. Energy Res., vol.
Jenkins, Steve Carroll, Myles Barker, A data-driven approach for char- 46, no. 11, pp. 15211–15230, 2022.
acterising the charging demand of electric vehicles: A UK case study, [27] Y. Zhao, Y. Guo, Q. Guo, H. Zhang, and H. Sun, “Deployment of the
Applied Energy, Volume 162, 2016, Pages 763-771, ISSN 0306-2619, electric vehicle charging station considering existing competitors,” IEEE
https://doi.org/10.1016/j.apenergy.2015.10.151. Trans. Smart Grid, vol. 11, no.5, pp. 4236–4248, 2020.
[8] S. Y. He, Y.-H. Kuo, and D. Wu, “Incorporating institutional and spatial [28] “EVI-Pro: Electric Vehicle Infrastructure – Projection Tool,” Nrel.gov.
factors in the selection of the optimal locations of public electric vehicle [Online]. Available: https://www.nrel.gov/transportation/evi-pro.html.
charging facilities: A case study of Beijing, China,” Transp. Res.Part C [Accessed: 25-Oct-2022].
Emerg. Technol., vol. 67,pp. 131–148, 2016. [29] “Electric vehicle infrastructure projection tool (EVI-pro) lite,”
[9] S. Y. He, Y.-H. Kuo, and K. K. Sun, “The spatial planning of public Energy.gov. [Online]. Available: https://afdc.energy.gov/evi-pro-
electric vehicle charging infrastructure in a high-density city using a lite.[Accessed: 05-Oct-2022].
contextualised location-allocation model,” Transp. Res. Part A Policy [30] M.-O. Metais, O. Jouini, Y. Perez, J. Berrada, and E. Suomalainen,
Pract., vol. 160, pp. 21–44, 2022. “Planning EV charging infrastructures: A literature review,” in 23rd
[10] R. Luo et al.,“Deep learning approach for long- term prediction of EUROWorking Group on Transportation Meeting, EWGT 2020, 2020.
electric vehicle (EV) charging station availability,” in 2021IEEE Inter- [31] M. O. Metais, O. Jouini, Y. Perez, J. Berrada, and E. Suomalainen, “Too
national Intelligent Transportation Systems Conference (ITSC), 2021. much or not enough? Planning electric vehicle charging infrastructure:
[11] D. Pevec, J. Babic, M. A. Kayser, A. Carvalho, Y. Ghiassi-Farrokhfal, A review of modeling options,” Renew. Sustain. Energy Rev., vol. 153,
and V. Podobnik, “A data-driven statistical approach for extending elec- no. 111719, p. 111719, 2022.
tric vehicle charging infrastructure: A data-driven statistical approach for [32] California Energy Commission, “Zero emission vehicle and in-
extending electric vehicle charging infrastructure,” Int. J. Energy Res., frastructure statistics,” California Energy Commission. [Online].
vol. 42, no. 9, pp. 3102–3120, 2018. Available: https://www.energy.ca.gov/data-reports/energy-almanac/zero-
[12] M. Straka et al., “Predicting popularity of electric vehicle charging emission-vehicle-and-infrastructure-statistics. [Accessed: 13-Oct-2022].
infrastructure in urban context,” IEEE Access, vol. 8, pp. 11315–11327, [33] “Electric Vehicle Charging Station usage,” [Online]. Available:
2020. https://data.cityofpaloalto.org/dataviews/257812/electric-vehicle-
[13] M. Majidpour, C. Qiu, P. Chu, R. Gadh, and H. R. Pota, “Modified pat- charging-station-usage-july-2011-dec-2020/. [Accessed: 18-Oct-2022]
tern sequence-based forecasting for electric vehicle charging stations,” [34] G. Lawton, “Data preprocessing: Definition, key steps and
in 2014 IEEE International Conference on Smart Grid Communica- concepts,” SearchDataManagement, 31-Jan-2022. [Online]. Available:
tions(SmartGridComm), 2014. https://www.techtarget.com/searchdatamanagement/definition/data-
[14] Pan, A., Tian, M., Tang, B., amp; Yang, X. (2019). Urban Electric preprocessing. [Accessed: 06-Nov-2022].
Vehicle (EV) charging point design based on User Travel Big Data. IOP [35] B. Artley, “Time Series Forecasting with Arima , Sarima
Conference Series: Earth and Environmental Science, 300(4), 1–10. and SARIMAX,” Medium, 27-Jun-2022. [Online]. Available:
[15] A. Roy and M. Law, “Examining spatial disparities in electric vehicle https://towardsdatascience.com/time-series-forecasting-with-arima-
charging station placements using machine learning,” Sustain. Cities sarima-and-sarimax-ee61099e78f6. [Accessed: 24-Feb-2023].
Soc., vol. 83, no. 103978, p. 103978, 2022. [36] “Simple model stacking, explained and automated.” [Online]. Avail-
[16] J. Shanmuganathan, A. A. Victoire, G. Balraj, and A. Victoire, “Deep able: https://towardsdatascience.com/simple-model-stacking-explained-
learning LSTM recurrent neural network model for prediction of electric and-automated-1b54e4357916. [Accessed: 7-Nov-2022]
vehicle charging demand,” Sustainability, vol. 14, no. 16, p. 10207, 2022. [37] “Improving Collaborative Filtering With Clustering.” [Online].
[17] C. Li, Z. Dong, G. Chen, B. Zhou, J. Zhang, and X. Yu, “Data- Available: https://medium.com/@jwu2/improving-collaborative-
driven planning of electric vehicle charging infrastructure: A case study filtering-with-clustering-88c63bdae7cc
of Sydney, Australia,” IEEE Trans. Smart Grid, vol. 12, no. 4, pp.
3289–3304, 2021.
[18] O. I. Asensio, D. J. Marchetto, S. Ha, and S. Dharur, “Extracting user
behavior at electric vehicle charging stations with transformer deep
learning models,” in CARMA 2020 - 3rd International Conference on
Advanced Research Methods and Analytics, 2020.
[19] B. Mouaad, M. Farag, B. Kawtar, K. Tarik, and Z. Malika, “A deep
learning approach for electric vehicle charging duration prediction at
public charging stations: The case of Morocco,” ITM Web Conf., vol.
43, p. 01024, 2022.
[20] A. Almaghrebi, F. Aljuheshi, M. Rafaie, K. James, and M. Alahmad,
“Data-driven charging demand prediction at public charging stations
using supervised machine learning regression methods,” Energies, vol.
13, no. 16, p. 4231, 2020.
[21] J. Asamer, M. Reinthaler, M. Ruthmair, M. Straub, and J. Puchinger,
“Optimizing charging station locations for urban taxi providers,” Transp.
Res. Part A Policy Pract., vol.85, pp. 233–246, 2016.
[22] S. Deb, K. Tammi, K. Kalita, and P. Mahanta, “Charging station
placement for electric vehicles: A case study of Guwahati city, India,”
IEEE Access, vol. 7, pp.100270–100282, 2019.
[23] Y. Xiong, J. Gan, B. An, C. Miao, and A. L. C. Bazzan, “Optimal
electric vehicle fast charging station placement based on game theoret-
ical framework,” IEEE Trans. Intell. Transp. Syst., vol. 19, no. 8, pp.
2493–2504, 2018.
[24] D. Adhya, A. Pal, A. K. Chakraborty, and A. Bhattacharya, “Machine
learning application for prediction of EV charging demand for the
scenario of agartala, India,” in 2022 4th International Conference on
Energy, Power and Environment (ICEPE), 2022.
[25] F. Soldan, E. Bionda, G. Mauri, and S. Celaschi, “Short-term forecast
of electric vehicle charging stations occupancy using big data streaming
analysis,” in 2021 IEEE International Conference on Environmentand

View publication stats