Systematic Literature Review on Electric Vehicles and Multicriteria Decision Making: Trends, Rankings, and Future Perspectives

There is a concern with themes that are of general interest, such as the issue of sustainability and environmental questions. Climate change and global warming are adversely affecting environmental quality, and transportation is a crucial factor for the environment [1]. Therefore, EVs have received attention around the world, mainly in relation to the reduction of gas emissions. In the last 200 years, mankind has begun to explore the world's resources, looking for non-renewable energy sources such as petroleum, coal, and natural gas, causing severe long-term environmental impacts [2], such as the high emission of carbon dioxide, which is linked to the rising global temperature [3]. With increasing demand for renewable energy sources and low environmental fuel consumption, vehicles powered by environmentally friendly sources are receiving government incentives, and the industry is growing significantly [4]. EVs are a possible alternative to help with sustainable development; they can decrease greenhouse gas emissions and the use of fossil fuels, and they are an optimal solution for transportation due to the zero emission of exhaust gas [5].

EVs play a very important role in decreasing pollution from fossil fuel consumption and are capable of making a paradigm shift in the entire transport sector, as transport contributes a lot to urban air pollution, and a reduction in pollution rates is the key need of the moment [6]. EVs have achieved great success, indicating that the automobile industry will soon be emission-free [7]. The electric car industry has developed in the last few years as they produce vehicles with high specifications that do not require the use of fossil fuels [6]. Therefore, the use of EVs is essential to controlling environmental challenges [8].

The growing search for the use of EVs also creates challenges for managers, considering that it will be necessary to increase production, search for charging spot locations, and replace batteries, in addition to battery reverse logistics, among other aspects. As a result, researchers in the area will need to carry out studies on different subjects linked to this topic, seeking solutions to the various problems that must be resolved with the growth in the search for EVs.

Due to the importance of the topic in recent years, several studies on EVs have been carried out in search of solutions, and consequently, literature reviews related to sustainability have been carried out, mainly due to the global demand for renewable energy sources [9], [10], [11]. Maybury et al. [12] carried out a SLR about research related to adhesion of EV use, based on the procedure of Kitchenham [11] and on the orientation presented by Weidt and Silva [13]. They analyzed 53 papers and concluded that most researchers apply the discrete choice model (DCM) and that the lack of successful businesses is a significant barrier to the adoption of EVs. Weidt and Silva [13] mapped a literature review on EV consumer behavior, analyzing 254 articles and verifying the existence of 4 clusters (environmental sustainability, transport infrastructure batteries, purchase ways, and battery use). Ding et al. [14] performed a literature review about EVs, addressing the operational processes of several EVs, battery technology, and super capacitors, aiming to increase the energy capacity of plug-in hybrid electric vehicles (PHEVs). Soares et al. [15] investigated EV supply chain management through a bibliometric and SLR using the PRISMA method.

The results of the systematic review showed that the most studied topics were related to risk management, while the bibliometric review showed the importance of associated costs as well as studies involving transparency and sustainability in the supply chain. Regarding the components, the battery was the most studied. Liao et al. [16] analyzed consumer preferences about EVs, comparing the economic and psychological approaches, followed by a conceptual framework of preferences. It was found that economic and technical aspects of EVs are generally significant, in addition to the density of charging stations, demonstrating the importance of charging infrastructure in the development of EVs. Other reviews about EV can be found in the literature, addressing aspects related to: commercial EV routing in urban logistics [17], locating charging infrastructure for shared autonomous EVs and for vehicle-to-grid strategies [18], characterizing the flexibility of EV charging strategies [19], consumer adoption of EVs [20], EV adoption [21], and sustainability assessment of EVs [22].

The MCDA methods can help researchers make decisions related to EV issues like the location of charging stations and the choice of EV and its components. Ecer [23] states that the techniques of MCDM are efficient tools for making the correct EV purchase decision. The selection of the most sustainable location for the EV charging station plays an important role in its life cycle. This process needs to consider some conflicting criteria and has a complex decision problem that can be modeled as a MCDM problem [24]. Pradhan et al. [25] cite that current research in the literature accepts that many MCDM techniques have been extensively used for various vehicle analyses.

SLR methods have advanced over time, with the creation of new methods or the adaptation or improvement of existing ones. Liberati et al. [26] proposed the PRISMA method for systematic reviews and meta-analyses. The PRISMA method was proposed in a meeting with several participants to analyze the QUOROM checklist and its flowchart [27]. Ensslin et al. [28] developed the ProKnow-C (Knowledge Development Process-Constructivist) method, which was organized into 4 blocks by Ensslin et al. [29]: bibliographic portfolio selection, bibliometric analysis, systematic analysis, and research question. These methods help researchers by guiding them in building a portfolio of articles. ProKnow-C and other SLR methods have spread throughout academia. The ProKnow-C method was adapted by Pagani et al. [30] for the creation of the Methodi Ordinatio.

Steffen et al. [31] proposed a SLR method that considers a raking index for ordering the publications in order of importance. There are 14 steps in the method that use normalized scientific indicators in the index calculation.

In this work, a SLR was carried out to create the ranking of articles based on the method proposed by Steffen et al. [31], with a modification in the ranking index calculation to make the method less subject and more robust. The index is obtained using the same normalized input data but in a weighted sum, in which the weights used for deciding the relevance of the papers are calculated using the AHP method. Furthermore, for generating the papers, ranking is made by using the arithmetic mean of the method TOPSIS.

Studies focused on sustainable issues are proving to be increasingly important. In this context, the main motivations of this study are:

Realize a SLR involving EVs and MCDM methods;

Propose a new way for ranking papers in a research field using MCDM methods in SLR techniques;

- Visualize methods more applied, main authors, among other aspects.

- Provide potential possibilities for future studies, presenting gaps in the literature.

To carry out this study, searches were carried out in two databases (Web of Science and Scopus), considering articles that fit the keywords chosen from exploratory searches.

This study is organized into four sections. Section 2 presents the methodologies adopted, pointing to the SLR method adopted and the MCDM and Cluster Analysis (CA) methods used. Section 3 discusses the results of the SLR. Finally, Section 4 presents a discussion around the results, directions for future studies, and limitations of the study.

SLR is a very useful tool that can make the review process more robust and efficient, resulting in a number of articles about a specific subject. From the final set of articles (normally named portfolios), various characteristics about the subject being reviewed can be easily determined. There are several SLR methods in the literature that are composed of many rigorous steps because they are not only a way to group existing papers but also a way to support the development of knowledge in many research fields [32].

Among the several SLRs available in the literature, there are some that provide an equation used to calculate an index that is the basis for ranking the papers in the final portfolio (FP). This ranking index is one of the SLR features that has improved in the last few years, as can be seen in the works of Pagani et al. [30], Pagani et al. [33], and Steffen et al. [31]. In this study, an adaptation of the SLR methodology proposed by Steffen et al. [31] is presented, and 14 steps are presented in Figure 1. The steps are divided into search, filtering, and ranking the papers; our aim was to improve the ranking procedure (more details about all steps can be found in the work of Steffen et al. [31]). This method incorporated MCDA methods for the generation of the criteria weights, aiming to generate the way of calculating the rank index used for the ordering of the FP. This new methodology was applied in the search for articles with the theme “MCDM methods involving EVs.”

The method proposed by Steffen et al. [31] consists of 14 steps, as does the modification proposed in this study. In Steffen et al. [31], the ranking obtained in Step 11 is calculated using Eq. (5), where researchers attribute the weights according to the importance they consider for each criterion. In this case, it is suggested that the weights must be calculated using a method widespread in the literature (AHP) and that the ranking is also an average between the calculation with the original method and using the TOPSIS method, providing greater robustness to the results.

Steps 1-5: In these initial steps, the authors must define research intentions, the databases that will be considered, and carry out exploratory searches to define keywords.

Steps 6.1-6.2: Usually some papers appear more than once in the search, so it is necessary to eliminate the duplicates (Step 6.1). This elimination can be carried out aided by software (like Endnote, Mendeley, JabRef, etc.) or manually. The next filtering procedure (Step 6.2) is the elimination by reading the title, keywords, and abstract, which help us analyze if the paper is really about the subject we want to review.

Steps 7-9: The definition of the journal indicators is realized in step 7. After this, searches are carried out for the indicators of each article (Step 8), and articles without indicators are eliminated (Step 9). In the journal indicators definition, all papers without citations published more than five years ago or in journals without an impact factor were excluded.

Steps 10-11: In step 10, the normalization (Eqs. (1)-(2)) of the decision matrix (composed by articles and your respective indicators, which in this case are considered criteria) is realized. The adaptation in the methods is related to step 11, as presented in Figure 1, which was split into many sub-steps. In addition to calculating the original index of the study proposed by Steffen et al. [31], the weights are calculated using the TOPSIS method, developed by Hwang and Yoon [34], and the final ranking will be defined by the average between the ranking proposed by Steffen et al. [31] ($A{{I}_{i}}$) and the ranking of the TOPSIS method (${{D}_{i}}$). The index proposed by Steffen et al. [31] uses some indicators, namely the journal science citation index ($JC$, like JCR, SNIP, and many others that can be chosen by the researcher), the approximate average citation per year of the article ($AC$), and the journal selfcitation fraction ($JS$, the fraction of the citations of the journal that are from the same journal). In general, these data are normalized by the maximums and minimums method, as proposed by Pal et al. [35].

Using this normalization process, all data will receive a value in a range from zero to one, meaning that for data that penalizes the index, the higher the value, the closer it becomes to zero after the normalization, and the smaller the value, the closer it becomes to one after the normalization. For data increasing the final value of the index, the opposite occurs: higher data become close to one and smaller data close to zero.

In the proposal of Steffen et al. [31], the index of article $i$ is calculated by Eq. (3):

where, $J{{C}_{jk}}$ is the $k$-th science citation index of the journal $j$, $\alpha $, $\beta$ and $\gamma$ are non-negative parameters whose values are chosen, in a subjective way, by the researcher, ${{w}_{k}}$ is the fraction weight attributed to the $k$-th science citation index, $J{{S}_{j}}$ is the journal $j$ fraction of self-citation, and ${{\overline{AC}}_{i}}$ is approximately the number of citations per year of the article $i$ calculated by Eq. (4):

where, $A{{C}_{i}}$ is the total number of citations of the paper $i$ up to the search date, information that can be found in Google Scholar [31], [36], [37], $A{{A}_{i}}$ is the “Article Age,” i.e., the difference between the year in which the research was carried out and the year the article $i$ was published.

The $JC$ and $\overline{AC}$ indicators are normalized using Eq. (1) and $JS$ is normalized by Eq. (2), because the self-citation practice makes more probable the citation of irrelevant paper.

As $JC$ we used JCR and SNIP. The indicators, SNIP and self-citations are provided by CWTS journal indicators and the numbers of citations are from Google Scholar [30], [37], [38], [39], [40], [41]. The indicator JCR (Journal Citation Reports) can be found on the Clarivate website (https://jcr.clarivate.com/jcr).

The method proposed by Steffen et al. [31] is very interesting, but, there is a disadvantage related to the choice of $\alpha$, $\beta$ and $\gamma$ values, that is totally subjective. Thus, our aim is to propose a more robust method related to the index used in the rank using the same indicator as input data, but with a little modification in the equation. As can be seen in Eq. (5), the index is obtained by the weighted sum of indicators.

The determination of criteria weights (${{w}_{1}}$, ${{w}_{2}}$ and ${{w}_{3}}$) for both methods (Steffen et al. [31] and TOPSIS method) is carried out using the AHP method, proposed by Saaty [42]. TOPSIS and AHP methods are described in the next subsections.

Steps 12-14: Steps 12-13 are based, respectively, on downloading all articles that make up the FP and excluding articles for which downloading the complete article is not possible. Step 14 consists of carrying out a bibliometric and systematic analysis of the articles in the FP, with the aim of determining important points in these publications, as well as possible gaps in the area.

The criteria weights were calculated using the AHP method, proposed by Saaty [43], by creating a judgment matrix of the A criteria (with ${{a}_{ij}}$ elements). Each matrix element represents the degree of importance of a criterion in relation to another criterion.

where:

i) ${{a}_{ij}}>0$

ii) ${{a}_{ij}}=\frac{1}{{{a}_{ij}}}\to reciprocal$

iii) ${{a}_{ii}}=1$

iv) ${{a}_{ik}}={{a}_{ij}}.{{a}_{jk}}$

This matrix was filled according to the scale of Saaty [43], as presented in Table 1:

From the matrix of judgments, a new normalized matrix B (composed of ${{b}_{ij}}$ elements) is calculated [41], where:

With the priority eigenvector (${{W}_{i}}$) defined as:

and

The eigenvector represents the value of the matrix elements in relation to the total. To verify the consistency of the calculations, Saaty [43] created a consistency index (CI), and this index cannot be greater than 10%. For this calculation, first multiply the elements of matrix A by their respective weights (${{W}_{i}}$), creating a new matrix C (${{c}_{ij}}$ elements):

Add each line of C and divide the value by its respective weight ${{W}_{i}}$, generating an ${{H}_{i}},$ value, and calculate the eigenvalue associated with this matrix (${{\lambda }_{max}}$), which is the arithmetic mean of ${{H}_{i}}$:

CI is calculated by:

where, RI (random index) proposed by Saaty [43] is described in Table 2:

The TOPSIS method was proposed by Hwang and Yoon [34]. The application of the TOPSIS method considering the normalization of maximums and minimums [44]. Therefore, the calculations started with the matrix X.

The next step of the method is the multiplication of the weights of each criterion (which in this case was considered by the average between the weights of the AHP and entropy methods) by the elements of matrix X, generating a new matrix V (with elements ${{v}_{ij}}$).

After that, the ideal and the negative-ideal solution are considered, generating the alternatives ${{A}^{*}}$ e ${{A}^{-}}$:

where:

$J=\text{ }\!\!\{\!\!\text{ }j=1,..,n\text{ }\!\!|\!\!\text{ }j\}~ {\text{is associated a benefit criteria}}$

$J\acute{\ }=\text{ }\!\!\{\!\!\text{ }j=1,..,n\text{ }\!\!|\!\!\text{ }j\}~{\text{is associated a cost criteria}}$

Hwang and Yoon [34] mention that the two alternatives (${{A}^{*}}$ e ${{A}^{-}}$) will indicate, respectively, the most preferable alternative (ideal solution) and the least preferable alternative (negative-ideal solution). The calculation of the distances of each alternative m, ideal and negative-ideal, can be measured by the n-dimensional Euclidean distance, presented, respectively in Eqs. (16)-(17):

The calculation of the ideal solution is performed by Eq. (23):

From value ${{D}_{i}}$, it is possible to create the ranking of the alternatives in descending order, that is, the highest value indicates a better performance of the alternatives in relation to the criteria.

After defining the research objectives, preliminary research was carried out to create the portfolio, where the keywords were defined and the search in the considered databases was carried out, with the results presented in Table 3:

The searches were carried out in other databases, which are not included in the analysis, as they did not show results that would alter the FP. Table 4 shows some numbers for the filtering procedures.

After the filtering procedure, it was decided not to limit the number of papers in the FP, i.e., to consider all 75 articles. Looking for the data on the considered indicators (SNIP, JCR, self-citations, and number of citations), two articles do not have citation and FI indicators, despite being publications over 5 years old, so they were excluded from the portfolio. Therefore, the FP was composed of 73 publications, with all complete files found for download. From the definition of the FP, the criteria weights were constructed using the AHP method. Three criteria were considered: IF, self-citations, and average citations (Citations/(Age+1)). The average of citations was calculated by dividing the number of citations by the age of the article + 1, in order to avoid divisions by 0 for articles published in the year 2023. Table 5 shows the matrix of judgments for the criteria.

From the matrix of judgments, the weights for the criterion were determined, and the CI was verified. The CI was 0.088, which agrees with what was proposed by Saaty [43]. The criteria weights obtained are presented in Table 6.

The average number of citations was the criterion with the highest weight, followed by IF and, with the lowest weight, self-citations. The first two criteria are considered “benefit,” while self-citations are considered “cost,” that is, the lower the better. With the weight of the criteria in hand, the next step was to normalize the data using the maximums and minimums method by Chakraborty and Yeh [44]. With normalized data, the indices proposed by Steffen et al. [31] and the TOPSIS method. For the calculation of the mean, the index values were normalized (${{D}_{i}}$ and $A{{I}_{i}}$) and the average of these was calculated, generating the $I$ index. The results of the index $I$, as well as the ranking of the 10 first articles that make up the FP, are presented in Table 7, while the other articles that make up the final ranking are presented in the appendix.

The article that ranked first is “Optimal site selection of electric vehicle charging stations by using fuzzy TOPSIS based on sustainability perspective,” authored by Gu and Zhao [45], and published in 2015 in the journal Applied Energy. This article had the highest number of citations (375) and the approximate average citation per year, 41.67, with a high IF (the third highest JCR of the FP) and a number of self-citations of 8.1% (among the 15 lowest). Second in the ranking is the article entitled “A consolidated MCDM framework for performance assessment of battery electric vehicles based on ranking strategies,” authored by Ecer [23], published in the journal “Renewable and Sustainable Energy Reviews” in 2021. This article did not have the second highest number of citations (101), but it had the second highest approximate average number of citations per year, 33.67, showing the importance of calculating the average number of citations in the analysis, as proposed by Steffen et al. [31], where it is considered that an older article had more time to be cited in relation to a more recent one. The IF (JCR and SNIP) considered in this analysis were the highest for the second article, and the self-citation is among the lowest for the FP (6%). On the other hand, the article “A Fuzzy Analytic Hierarchy Process and VIKOR Framework for Evaluation and Selection of Electric Vehicle Charging Technology for India,” authored by Mall and Anbanandam [113] and published in 2022 in the journal Transportation in Developing Economies, ranked last. Despite being recent, the article has 2 citations, but IF equals 0 for SNIP and JCR, which explains its final position. Figure 2 presents the temporal evolution in the number of publications from the FP, where one can see the rise in the number of publications mainly in the last few years.

The theme “EV” is recent, even more so considering the solution of problems involving MCDM methods, which is noticeable by temporal analysis. Considering the year 2023, whose review was carried out on March 30 (1st quarter), approximately 85% of papers were published in the last 5 years. The first publication found was in 2009, with a 6-year gap for the next two publications in 2015. 2021 had the highest number of publications (21), followed by 2022 with 14. Figure 3 shows the countries with the most contributions to the composition of the FP.

China, with 22 publications, is the country that most contributed to the analysis, followed by India, with 18 publications, and Turkey and Poland, with 7 and 6 publications, respectively. These 4 countries represent approximately 72.6% of the FP, while the other 16 countries that are part of the FP had 3 publications or less. To illustrate possible groups that are strongly active in themes related to the subject under study, an analysis was carried out of the universities/companies that participated in the analysis, considering the main author, as presented in Figure 4.

The analysis of possible groups allows researchers interested in the subject to have a direction to seek information and/or partnerships for future studies. There are 11 universities with 2 publications or more and another 43 institutions with 1 publication each, among those that are in the FP. The highlight is North China Electric Power University, which has six publications, followed by the University of Szczecin, with three publications. Figure 5 presents the main journals in the FP.

Energies is the journal that contributed more to the FP, with 8 publications. Four journals are presented in sequence, with three occurrences: Energy Sources, Part A: Recovery, Utilization, and Environmental Effects, Journal of Cleaner Production, Journal of Energy Storage, and Sustainable Cities and Society. There are nine journals presenting two articles each, and there are another 35 journals that have one publication.

Figure 6 shows the average citation per year. For this construction, the average citations per year have been divided by the number of citations in the year by the number of publications and by the age of the articles. 2015 is the year with the highest average citations, and 2023, on the other hand, is the year with the lowest number of average citations. This can be explained by the fact that the publication that was in first place in the ranking and had the highest number of publications was from 2015, and there was only one more publication in that year, which on average made the number high. Articles of 2022 and 2023 are very recent and, considered the temporal window, had less time to be cited compared to the 2009 article.

Figure 7 presents the density map of the main authors, built using the VOSviewer software. Shabbiruddin is the author with the most participation in publications, with six occurrences, one as the main author and the other five as co-authors. Deveci, M., Liu, H.-C., Pamucar, D., Pradhan, S., and Ziemba, P. contributed 4 publications, and Ghose, D., has 3 publications from the FP. There are 20 authors with two publications, and 188 authors participated in 1 publication each from the portfolio. The author with more publications as the main author was Ziemba, P., namely, 4 publications.

From the bibliometric analysis of the articles, a more detailed analysis was carried out to verify the most used methods, and main applications, among other contributions. The topics addressed in the publications are described in Table 8.

There are 31 studies that carried out an analysis about the installation of a charging station, ranging from charging stations on streets to the installation of charging stations at workstations. EV Selection, Factors/Barriers of Adoption of EVs, and Battery Swapping Station appear next, with 11, 6, and 5 occurrences, respectively. Evaluating EVs and supplier selection, with two contributions each, are also themes found in the FP articles. Table 9 presents the methods most utilized in this analysis.

The methods were considered separately, and only those that had more than three occurrences were cited. Fuzzy was the most used method, with 32 occurrences. Other widely used methods are the AHP and TOPSIS methods, with 20 and 17 occurrences, respectively. The CRITIC and EWM methods, as well as the AHP method, are widely used in the literature for calculating scale constants, and both appear among those cited with more than 3 occurrences. On the other hand, only one study used methods from the ELECTRE family [64]. Compared to other methods, Fuzzy stood out. This fact may indicate that there is an environment of uncertainty in solving problems related to the topic of EV, as it is a characteristic of this method. In this context, a literature review has already been carried out around fuzzy MCDM methods in the formulation of energy policies [117], highlighting that these methods are widely used in the area. Another study that indicated fuzzy methods as a widely used technique in the area of sustainability [118]. Table 10 presents the most used methods related to the themes most considered in the FP.

Relating the most used methods with the most discussed themes in the articles, it is possible to verify that the most used theme, “Charging Station,” when combined with the 3 most used methods (Fuzzy, AHP, and TOPSIS), are the ones that occupy the first positions among the most used combinations, with 15, 12, and 8 occurrences.

Ziemba, P., the author with the most publications as the first author, used methods from the PROMETHEE family in the 4 articles, ranking fourth when considering the methods individually, combined with the topic “EV Selection,” the second most discussed. From the 4 articles, 3 were published in the journal Energies, the journal more utilized per article that composed the FP.

This study addressed a SLR verifying the main aspects between "EVs" and "MCDM." For this, the method proposed by Steffen et al. [31] and the MCDM methods for calculating a ranking of the articles that make up the FP. Citations remain the most important criterion; however, the importance of considering the average citations, depending on the age of the articles, was shown. In addition, the consideration of other important aspects of journals, such as IF and self-citations, makes the analysis more robust and the placement of articles more coherent.

In the articles that are part of the portfolio, the wide use of fuzzy methods stands out, considering them in general, which are widely used in environments of uncertainty when the decision-maker has doubts about the analysis. Alao et al. [119] mention that to capture possible inaccuracies and uncertainties in the judgment of experts, fuzzy is applied to transform linguistic variables into fuzzy numbers that can later be manipulated. A lot of the time, compensatory multi-criteria methods are used instead of non-compensatory methods. This may be because most of the criteria that are looked at in this type of subject are compensatory. Among the most used methods, non-PROMETHEE methods, widely used in the literature, appear only in fourth place, appearing in 8 publications, while ELECTRE is present in only 1.

The literature has addressed, in most cases involving “EVs” and "MCDM,” the issues of charging stations. Although there is a wide choice of analysis in combustion-powered cars, only 11 publications have analyzed this theme in "EVs," or approximately 15.07% of the FP. Generally, electric cars sold in different countries are not the same, as they depend on several factors, such as government and brand interest in entering the market in each country. With this, possible future studies can broadly explore the application of MCDM methods for choosing the best electric car in a given country. It is still possible to explore other techniques, in addition to MCDM, for creating hybrid methods.

As it is a recent topic, the first article found dates from 2009, with most articles published in the last 5 years. Themes involving sustainability are increasingly in focus, in view of the global problems with rising temperatures and carbon dioxide emissions, which directly impact nature. Problems that can solve or minimize these impacts must be increasingly addressed, as they can help decision-makers in their decision-making.