Computing Complexity Reduction for Predictive Control of Engine Thermal Management System

摘要

This paper presents the design, implementation, and performance evaluation of a reduced complexity algorithm for a predictive control which is based on our previously published SAE paper (2021-01-0225) titled, "Model Predictive Control for Engine Thermal Management System." That paper presented a model predictive control (MPC) design concept and demonstrated energy efficiency improvements by enabling engine pre-cooling based on GPS/Navigation data to recognize future vehicle speed limit and road grade in anticipation of high engine load demand. When compared to conventional control, the predictive control demonstrated considerable energy and fuel savings due to delayed timing of both knock mitigation and activation of radiator cooling fan during high engine load demand. However, this predictive control strategy is much more complicated due to its highly coupled nonlinear behavior. Also, in reality, the previous developed MPC strategies are limited to the computational resources in engine control units (ECU). Therefore, to address these challenges, a reduced-complexity MPC controller for the powertrain thermal system is developed in this paper where, by "reduced-complexity," it is meant that the MPC controller achieves control objectives and to be executed on a modern ECU within a computation budget. To maximize fuel economy, one of the key control logics in the previously published SAE paper is to use estimated radiator outlet coolant temperature as an indicator to determine when to lower the target engine coolant temperature. Since the heat transfer coefficients (HTC) of the thermal system are time varying, the computation load is high if the inputs to the plant models are physics-based. Therefore, lookup table (LUT)-based and mean value model (MVM) modeling approaches are developed and combined in order to reduce computational burden. The complexity reduction is obtained via lookup tables and mean value models, which lightens the computational load of the algorithm with a minimal loss in precision. Experimental results showed that the proposed approach is able to deliver performance similar to originally proposed approaches. At the same time, the proposed algorithm is able to cut the computational complexity of the physics-based algorithm by an 18% factor.