Efficient Design and Optimization of Large DC Distribution Architectures Using Descent Based Methods

Complex dc distribution architectures are often required in electronic systems to generate multiple point-of-load regulated dc outputs from a single dc input. Typically these systems are realized using a set of single-input, single-output dc-dc modules. The system optimization involves choosing an arrangement of dc-dc modules to minimize the system losses, size or a user-defined figure of merit. Exhaustive-search based system optimization methods become intractable for large systems, i.e., when the number of required outputs, n
o
, exceeds eight. Presented in this work is an efficient descent-based design automation technique that optimizes the system without the use of ad-hoc heuristics or complexities associated with an alternative efficient method based on simulated annealing. Test cases show that the proposed algorithm can, for small and medium problems, decrease solve time by up to 60 % with no more than a 1 % decrease in figure of merit compared to the exhaustive search and, for large systems, finds a solution with a figure of merit 1 % better compared to simulated annealing with a 38 % increase in solve time.