The transition from decentralized Electronic Control Units (ECUs) to centralized Domain Control Units (DCUs) in automotive electric/electronics architecture intensifies the need for efficient thermal management soluti...
详细信息
The transition from decentralized Electronic Control Units (ECUs) to centralized Domain Control Units (DCUs) in automotive electric/electronics architecture intensifies the need for efficient thermal management solutions. High power density chips within minimized DCUs generate significant heat flux over limited spaces, challenging conventional cooling methods amid constraints like limited installation space, weight reduction, and cost efficiency. Natural convection air cooling remains a cost-effective and reliable solution, but optimizing fin arrangements and shapes is critical to enhance heat dissipation. This study presents a multi-objective optimization framework that integrates a deep learning-based geneticalgorithm to determine the optimal fin layout for a given heat source and Thermal Interface Materials (TIMs) configuration. A generative temperature field prediction model is developed using a U-Net backbone combined with a mask attention module, effectively predicting temperature fields under varying parameters, including heat source and TIM properties, and fin dimensions. The model demonstrates reliable performance with mean maximum absolute errors of 3.14 degrees C and 1.30 degrees C for the maximum and minimum temperatures of all heat sources, respectively. Such model effectively overcomes the limitations of traditional, computationally intensive optimization techniques that rely on iterative numerical simulations. By integrating a deep learning surrogate model with a geneticalgorithm, the framework offers a practical and efficient solution for thermal optimization in complex systems like DCUs, facilitating rapid design iterations and contributing to improved thermal management.
This paper proposes a real-time Energy Management System (EMS) for a low voltage (LV) Microgrid (MG). The system operation consists in solving the Unit Commitment (UC) and Economic Load Dispatch (ELD) simultaneously f...
详细信息
ISBN:
(纸本)9781467380416
This paper proposes a real-time Energy Management System (EMS) for a low voltage (LV) Microgrid (MG). The system operation consists in solving the Unit Commitment (UC) and Economic Load Dispatch (ELD) simultaneously for 24 hours ahead at every 15-minute period. This operation is formulated as a multi-objective optimization problem where the minimization of operational cost, total emissions and power losses is simultaneously pursued using the non-dominated sortinggeneticalgorithm II (NSGA-II). In this algorithm, crossover and mutation operators were improved with respect to existing approaches to achieve an adequate characterization of the energy management problem and a good algorithm performance. Simulation studies have outlined that, in fact, the NSGA-II can be used as a real-time optimization tool providing a good-quality Pareto front to operate optimally the MG in a limited time of 15 minutes.
暂无评论