Optimizing Charging Infrastructure for Electric Autonomous Vehicles in Smart Manufacturing using Hyperparameter-Tuned Artificial Neural Networks

Abstract

An optimised charging infrastructure is required for the integration of electric autonomous vehicles (A-EVs) into smart industrial systems in order to improve operational efficiency, minimise downtime, and guarantee sustainable energy management. An AI-driven framework for the planned and strategic placement of A-EV charging stations in industrial settings is presented in this paper. The platform uses predictive analytics, real-time IoT data, and machine learning models to dynamically modify charging station allocation according to industrial workflow requirements, grid stability, and car demand. To guarantee the best possible positioning and use of charging stations, the suggested methodology combines optimisation algorithms, reinforcement learning-based scheduling, and spatiotemporal data analysis. This study maximizes charging infrastructure to electric autonomous vehicles with the A-EV Grid Management Dataset through the use of KNN-based imputation, Z-score outliers, Min-Max scaling, SMOTE balancing, and feature selection with RFE and Lasso. An ANN model is hyperparameter-tuned and stratified 5-fold cross-validation is used to evaluate the model. The experimental results indicate that the proposed ANN is more accurate and has 94.5% accuracy, AUC-ROC as 95.2%, and 5.3% as MAE, which is higher than RF, SVM, KNN, and Decision Tree baselines. The results indicate that the model is efficient in smart manufacturing systems to support the real time charging decisions. The results of comparative performance evaluations utilising Random Forest, SVM, KNN, Decision Tree, and a hyperparameter-tuned Artificial Neural Network (ANN) show that the ANN model predicts optimal charging decisions with greater accuracy (94.5%) and precision (93.7%). Compared to traditional static deployment options, the results demonstrate a significant improvement in energy efficiency, less congestion at charging stations, and cheaper operating costs. This study emphasises how crucial AI-enhanced infrastructure planning is for smart manufacturing's autonomous EV fleets. To better optimise A-EV charging networks, future studies may investigate blockchain for decentralised energy management, multi-agent reinforcement learning for adaptive station allocation, and 5G-enabled V2X communication. The results further the development of highly automated, intelligent, and energy-efficient industrial transportation systems.