AI-BASED PORTABLE SUBSTATIONS AND TRANSFORMER DESIGN FOR FAST EV CHARGING: A CRITICAL REVIEW OF INNOVATIONS AND CHALLENGES

Abstract

This study addresses the pressing engineering problem of how to deliver high-power EV fast charging at the grid edge without compromising reliability, power quality, deployment speed, or cost. The purpose is to quantify how AI integration and transformer or portable-substation design choices relate to real-world field performance. We conduct a quantitative, cross-sectional, case-based investigation on 10 operating fast-charging sites that are cloud connected and enterprise managed, and pair it with a systematic literature review of 68 publications to ground constructs and measures. The sample comprises cloud or enterprise cases instrumented with SCADA and power-quality telemetry. Key independent variables include AI integration, modularity and standardization of portable skids, transformer technology and cooling class, with contextual moderators such as feeder short-circuit strength and ambient temperature. Outcomes include availability, energy throughput, total harmonic distortion, voltage deviation events, normalized hottest-spot temperature rise, installed cost per kW, and time-to-energize. The analysis plan follows three tiers: descriptive profiling, correlation screening, and multivariable modeling using OLS with heteroskedasticity-robust errors for symmetric outcomes and Gamma log-link GLMs for positive skewed cost and schedule metrics, with pre-registered interaction tests. Headline findings show that stronger AI integration is associated with higher availability and throughput and with lower distortion and thermal stress, modular standardized skids reduce installed cost and commissioning time, and solid-state transformer packages deliver clearer power-quality benefits on weak feeders. Implications for operators and OEMs include prioritizing standardized portable designs, instrumented data access, and an AI operations stack to optimize ramp management and predictive maintenance across heterogeneous sites.