Learning MPC based Predictive Energy Management for Hybrid eVTOL with Transformer-based Multimodal Power Prediction

This article proposes a Predictive Energy Management Strategy (PEMS) framework that employs a multimodal power forecasting model, specifically tailored for Hybrid Electric Propulsion System eVTOL (HEPS eVTOL) operating in urban low-altitude environments. Firstly, a Pilot-in-the-Loop (PiL) simulation data acquisition framework is devised for eVTOL in urban scenarios to gather the necessary motion, lidar and power data. Subsequently, to optimize power prediction performance of eVTOL, a Motion-Environment Fusion Transformer Network (METN) is proposed, enables better handling of power demands in complex urban environments. Then, a Learning MPC (LMPC) strategy to solve the nonlinear predictive optimization problem online is proposed, which considers fuel economy, battery State of Charge (SOC) and State of Health (SOH). Compared with LSTM baseline, METN has achieved 16.5% average improvement in the precision of power prediction by capturing the influence of motion and environmental features. Meanwhile, the proposed LMPC has demonstrated better performance, achieving a fuel consumption reduction of 4.163% and a SOH optimization of 13.4% compared with LSTM-MPC baseline. In addition, the METN-LMPC method has a single-step runtime of 33.8 ms. The simulation results verify the superiority of the proposed PEMS in power prediction and real-time online optimization. Furthermore, experiment in a robustness test bench also validate its effectiveness.