A Learning Model Predictive Control Framework With Long Short-Term Memory for Energy Management in Fuel Cell Hybrid Electric Buses

To leverage the complementary physical characteristics (e.g., dynamic response) of fuel cells (FCs) and supercapacitors (SCs), effective energy management strategies (EMSs) need to be developed for FC/SC hybrid energy storage systems (HESSs). This paper proposes an EMS for fuel cell hybrid electric buses (FCHEBs), which consists of a learning model predictive control (LMPC) block and a speed predictor based on a long short-term memory (LSTM) network. Therefore, the proposed framework is referred to as the LMPC-LSTM framework. Specifically, the LMPC controller integrates a linear model predictive control (MPC) block, an iterative learning control (ILC) block, and a safety set for the SC state of charge (SOC). The objectives of the LMPC-LSTM framework are to properly control the SC SOC while reducing the hydrogen cost and the FC degradation cost. Extensive simulations and experiments are performed to demonstrate the applicability and effectiveness of the LMPC-LSTM framework. In terms of the hydrogen cost, the LMPC-LSTM framework performs better than the MPC and LMPC frameworks. In terms of the FC degradation cost, the LMPC-LSTM framework still outperforms the MPC framework although it performs marginally worse than the LMPC framework. Since the hydrogen cost is significantly higher than the FC degradation cost, the LMPC-LSTM framework outperforms the MPC and LMPC frameworks in terms of the system operating cost, which means that the LMPC-LSTM framework improves the system economy. On the other hand, the LMPC-LSTM framework enhances the system stability by reducing the SC SOC change over an entire drive cycle. Therefore, in terms of both the system economy and the system stability, the LMPC-LSTM framework outperforms the MPC and LMPC frameworks.