Modular Artificial Neural Networks for Wireless Power Transfer Optimization in Sensor-Driven Industrial IoT

This work presents a novel approach, utilizing modular Artificial Neural Networks (ANNs), to model complex and confined electromagnetic (EM) environments, when the far-field approximation is inadequate. The main objective is to optimize energy harvesting and sensor placement within Wireless Power Transfer (WPT) systems, which are crucial for the autonomous functioning of Wireless Sensor Networks (WSNs) in harsh EM environments. To enhance computational efficiency, the Integral Solver method is adopted to create parameterized EM simulation scenarios, for the generation of the training data. Additionally, an active learning algorithm is employed to identify an optimal, minimal dataset for training and testing the modular ANN architecture. This architecture comprises distinct sub-networks aimed at predicting both optimal sensors spatial coordinates and maximum power density levels. The evaluation of these sub-networks demonstrates the effectiveness of ANN-based methods in tackling the challenges associated with WPT optimization for WSN applications in demanding EM environments.