Approximate Scenario-Based Model Predictive Control for Battery Thermal Management in Electric Vehicles
提出一种近似场景模型预测控制框架，利用深度神经网络逼近场景控制律，在长期速度预测不确定下提升电池温度控制鲁棒性，降低能耗2.18%-2.76%并减少计算时间。
The performance and safety of the battery are affected by temperature. To maintain the battery temperature within a suitable range, the battery thermal management (BTM) system in electric vehicles (EVs) consumes considerable energy, which significantly reduces the driving range of EVs. Due to the slow dynamics of the thermal system, a long prediction horizon is required to prevent overheating of the battery and achieve low energy consumption. However, the large uncertainties associated with long-term velocity prediction significantly affect performance and robustness. This article proposes an approximate scenario-based model predictive control (ASCMPC) framework to improve the battery temperature control robustness under long-term preview uncertainty and reduce energy consumption while decreasing computation time. The scenario-based model predictive control (SCMPC) determines the control law by optimizing multiple previous velocity examples on the same route, which reduces energy consumption and satisfies battery temperature constraint in the presence of preview uncertainties. But SCMPC inevitably increases the number of optimization problem; thus, a trained deep neural network (DNN) is adopted to obtain an approximate SCMPC control law for easy online implementation. The effectiveness of the approximate controller is verified using Hoeffding’s inequality. Co-simulation results show that the proposed ASCMPC enhances the enforcement of battery temperature constraints and reduces energy consumption by 2.18%–2.76% compared to nondeterministic MPC under the uncertainty of preview information and different ambient temperatures, which also has less computation time.