Answers are grounded in this paper and cite their sources.
This paper presents a multi-layer model predictive control (MPC) framework suitable for integrated power/propulsion systems, which can decouple system-level energy optimization from dynamic power management in a holistic manner. This control platform can be built on distributed heterogeneous computing hardware, including real-time processors and FPGAs (field programmable gate arrays). The MPC schemes, which may be readily implemented on NI’s CompactRIO modules or PXI FPGA modules, can be programmed to perform certain computational steps in parallel to find the optimized control actions. The integral part is a real-time quadratic programming (QP) based optimizer, which is accelerated by FPGA and used to generate optimal control actions based on online model prediction. Machine learning algorithms can be leveraged to learn model parameters from real-time operation data, in addition to known physical models. This QP optimizer aims to minimize a defined quadratic objective function, subject to sets of equality and inequality constraints, and solve the convex optimization problem using the first-order optimality conditions. The Jacobian matrix and gradient vector are updated accordingly, and then are sent to a linear solver which is used to solve a series of linearized equation sets within each of the Newton’s iterations in the QP solver. Following this, a linear solver solves a system of linear equations to find the search direction and incremental step, based on the Jacobian matrix and gradient vector. Finally, after the linear solver completes the entire solution process, the solution vector of the QP optimizer will be updated. In the multi-layer structure, the higher-level MPC aims to minimize the energy conversion losses and power generation costs, subject to dynamic load profiles, and maintain a dynamic level of reserved energy in storage to meet future changes in power demand. The lower-level MPC is to optimize the dynamic power management, i.e., controlling the currents to satisfy transient power needs while regulating the bus voltage. Simulation results from a pilot study show that the two-layer control system operates well in managing both energy flows and dynamic current/voltage regulations under dynamic conditions. Results also suggest that the QP optimizer in the MPC controller works properly in comparison with other methods.
No takes yet. Share an insight, caveat, or question.
Jiang et al. (2020) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: