Deep learning compound architecture-based coordinated control strategy for hybrid electric power system with turboshaft engine

Hybrid electric power system based on turboshaft engines (HEPS-TE) emerges as a practical solution for powering the land and air vehicles (LAVs), which are capable of operating both on land and in the air. The coordinated control strategy is crucial for achieving coordinated operation of the components of HEPS-TE, based on the optimal reference commands from the upper-level energy management strategy, to enhance the system's efficiency. In this paper, a nonlinear Model Predictive Control (MPC) strategy based on a deep learning compound architecture is proposed for the coordinated control of HEPS-TE. First, the architectures of deep learning neural network, radial basis function neural network and extreme learning machine neural network are compared to determine the optimal basic NN architecture for building the thermodynamic performance prediction model of turboshaft engine. Subsequently, employing a multi-mode NN integration method coupled with buffer zone design, the final engine prediction model is established. By integrating the mechanism model of the mechanical transmission system, a generation unit prediction model based on a deep learning compound architecture (DL-CA) is presented, achieving a 95.09 % reduction in computational burden compared to component-level models. Furthermore, a DL-CA based nonlinear MPC coordinated control strategy is formulated for HEPS-TE to attain economic reference target tracking. Numerical simulation results demonstrate that, compared to the PID control strategy, the proposed control strategy can fully harness the potential of HEPS-TE while satisfying constraints. Moreover, even when compared to linear parameter-varying MPC, the fuel consumption can be reduced by 4.03 % on the air-land driving cycle. The DL-CA based MPC is conducive to promoting the application of hybrid electric LAV.