<p>The auxiliary power supply system of electric locomotives is connected to various types of loads. In particular, nonlinear loads such as inverter-driven air conditioners cause significant output voltage distortion, resulting in degraded power quality. To mitigate the voltage distortion induced by nonlinear loads, feedforward compensation based on accurate load current estimation is required. In this study, AI-based time-series prediction models are employed to check feasibility of estimating the load current, and the prediction performance of state-of-the-art machine learning techniques is comparatively evaluated. Furthermore, model lightweighting was performed and execution time was compared to enable real-time implementation on embedded systems. Through this study, the feasibility of implementing AI-based load prediction techniques on embedded platforms was evaluated.</p>