<p>The rapid growth of electric vehicle (EV) adoption presents significant challenges for charging infrastructure planning and grid integration, particularly at the regional level. However, existing studies often apply machine learning techniques in isolation and lack integrated, region-specific behavioural modelling. This study introduces the Intelligent Sustainable EV Clustering and Analysis Platform (ISE-CAP), an integrated and interpretable analytical framework that advances beyond conventional ML-based EV studies by combining behavioural clustering, predictive modelling, explainable artificial intelligence (XAI), and adaptive optimisation within a regionally comparative decision-support architecture. A structured survey dataset comprising 256 EV users from the North East (n = 124) and West Midlands (n = 132) was analysed to examine charging behaviour, adoption motivations, and infrastructure preferences. K-Means clustering identified three distinct EV user groups in each region. Predictive models, including Random Forest, CatBoost, and a compact deep learning architecture, were trained using an 80:20 train-test split with cross-validation achieved high accuracy on the available regional dataset, with the North East model attaining <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\hbox {R}^{2}\)</EquationSource> </InlineEquation> = 0.9951 and MSE = 0.0694, indicating a very strong fit to the observed data. On the held-out regional test dataset, the West Midlands model achieved <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\hbox {R}^{2}\)</EquationSource> </InlineEquation>= 0.9548 and MSE = 0.6410, indicating strong predictive performance within the analysed sample. Charging behaviour analysis indicates that 85% of users prefer DC fast chargers, with most users (66% in the West Midlands and 53% in the North East) willing to travel up to 3 km for charging. Cost savings accounted for 65% of EV purchases in the North East, while environmental concerns accounted for 30% in the West Midlands, based on regional frequency distributions. SHapley Additive exPlanations (SHAP) analysis identified charging duration, real-time station availability, and cost as the most influential factors associated with charger selection preferences. The findings highlight regional heterogeneity in charging behaviour and infrastructure needs, emphasising the importance of adaptive and interpretable modelling approaches. While results are specific to the analysed regions, the ISE-CAP framework provides a scalable decision-support architecture for sustainable EV infrastructure optimisation in smart city contexts.</p>

Sustainable EV adoption with clustering and predictive modelling for optimal charging infrastructure in the West Midlands and North East UK

期刊论文
出版模式：

开放获取
发表日期：

2026年5月7日

查看全文

Muhammed Cavus,
Shouai Wang,
Sanchari Deb,
Anurag Sharma,
Margaret Bell,
Dilum Dissanayake

摘要

The rapid growth of electric vehicle (EV) adoption presents significant challenges for charging infrastructure planning and grid integration, particularly at the regional level. However, existing studies often apply machine learning techniques in isolation and lack integrated, region-specific behavioural modelling. This study introduces the Intelligent Sustainable EV Clustering and Analysis Platform (ISE-CAP), an integrated and interpretable analytical framework that advances beyond conventional ML-based EV studies by combining behavioural clustering, predictive modelling, explainable artificial intelligence (XAI), and adaptive optimisation within a regionally comparative decision-support architecture. A structured survey dataset comprising 256 EV users from the North East (n = 124) and West Midlands (n = 132) was analysed to examine charging behaviour, adoption motivations, and infrastructure preferences. K-Means clustering identified three distinct EV user groups in each region. Predictive models, including Random Forest, CatBoost, and a compact deep learning architecture, were trained using an 80:20 train-test split with cross-validation achieved high accuracy on the available regional dataset, with the North East model attaining \(\hbox {R}^{2}\) = 0.9951 and MSE = 0.0694, indicating a very strong fit to the observed data. On the held-out regional test dataset, the West Midlands model achieved \(\hbox {R}^{2}\) = 0.9548 and MSE = 0.6410, indicating strong predictive performance within the analysed sample. Charging behaviour analysis indicates that 85% of users prefer DC fast chargers, with most users (66% in the West Midlands and 53% in the North East) willing to travel up to 3 km for charging. Cost savings accounted for 65% of EV purchases in the North East, while environmental concerns accounted for 30% in the West Midlands, based on regional frequency distributions. SHapley Additive exPlanations (SHAP) analysis identified charging duration, real-time station availability, and cost as the most influential factors associated with charger selection preferences. The findings highlight regional heterogeneity in charging behaviour and infrastructure needs, emphasising the importance of adaptive and interpretable modelling approaches. While results are specific to the analysed regions, the ISE-CAP framework provides a scalable decision-support architecture for sustainable EV infrastructure optimisation in smart city contexts.