Abstract
In satellite layout design, heat source layout optimization (HSLO) is aneffective technique to decrease the maximum temperature and improve the heatmanagement of the whole system. Recently, deep learning surrogate assisted HSLOhas been proposed, which learns the mapping from layout to its correspondingtemperature field, so as to substitute the simulation during optimization todecrease the computational cost largely. However, it faces two main challenges:1) the neural network surrogate for the certain task is often manually designedto be complex and requires rich debugging experience, which is challenging forthe designers in the engineering field; 2) existing algorithms for HSLO couldonly obtain a near optimal solution in single optimization and are easilytrapped in local optimum. To address the first challenge, considering reducingthe total parameter numbers and ensuring the similar accuracy as well as, aneural architecture search (NAS) method combined with Feature Pyramid Network(FPN) framework is developed to realize the purpose of automatically searchingfor a small deep learning surrogate model for HSLO. To address the secondchallenge, a multimodal neighborhood search based layout optimization algorithm(MNSLO) is proposed, which could obtain more and better approximate optimaldesign schemes simultaneously in single optimization. Finally, two typicaltwo-dimensional heat conduction optimization problems are utilized todemonstrate the effectiveness of the proposed method. With the similaraccuracy, NAS finds models with 80% fewer parameters, 64% fewer FLOPs and 36%faster inference time than the original FPN. Besides, with the assistance ofdeep learning surrogate by automatic search, MNSLO could achieve multiple nearoptimal design schemes simultaneously to provide more design diversities fordesigners.