Machine Learning-Assisted DOA Beam Steering for Misalignment Mitigation in Vehicular Wireless Power Transfer

SOUKAINA JAAFARI, Ahmed Khallaayoun, Esmail Ahouzi

Abstract Wireless power transfer (WPT) for electric vehicles (EVs) suffers from significant power degradation under transmitter-receiver misalignment, particularly in dynamic scenarios. This paper presents an integrated machine-learning (ML) and beam-steering framework for mitigating misalignment through direction-of-arrival (DOA) estimation and adaptive transmit beam control. An 8×8 patch antenna array operating at 5.82 GHz is modeled to characterize the impact of distance, lateral displacement, and angular misalignment on WPT performance. Three ML models, Decision Tree, Support Vector Machine, and CNN+BiLSTM, are trained to estimate angular DOA and lateral offset under noise, CFO, gain/phase imbalance, and calibration errors. Results show high angular DOA estimation accuracy and acceptable lateral prediction accuracy within steering tolerances. The predicted DOAs are then used to dynamically steer the transmit array. Validation is conducted using both array-level beamforming analysis and full-wave electromagnetic simulations. Without correction, received power degrades by more than 20-30 dB beyond ±15° misalignment. With DOA-driven beam steering, the system achieves gain improvements exceeding 13 dB, 16 dB, and 18 dB at misalignment angles of 20°, 40°, and 60°, respectively. These results demonstrate the feasibility of ML-assisted DOA beam-steering as a practical approach for improving misalignment resilience in dynamic EV WPT.