在汽车气动外形优化设计中,往往需要大量的高精度CFD数据作为支撑。然而,高精度CFD数据获取难度大、成本高。为了缓解汽车气动优化设计中气动特性评估精度和效率之间的矛盾,根据迁移学习与数据融合的思想,提出了一种基于多精度深度神经...在汽车气动外形优化设计中,往往需要大量的高精度CFD数据作为支撑。然而,高精度CFD数据获取难度大、成本高。为了缓解汽车气动优化设计中气动特性评估精度和效率之间的矛盾,根据迁移学习与数据融合的思想,提出了一种基于多精度深度神经网络(multi-fidelity deep neural network, MFDNN)的汽车外形优化设计方法,以减少优化设计中所需的高精度数据个数,从而有效提升优化速度、降低优化成本。将所发展的优化方法应用于快背式MIRA标准模型减阻优化设计中,优化结果表明,该方法能够充分融合不同精度数据所蕴含的知识,加速气动外形优化进程,提升优化效率。以收敛用时作为评价指标,在取得相近或更优优化结果的前提下,基于多精度神经网络的优化框架的收敛速度是基于单精度神经网络的离线优化框架的5.85倍,是基于单精度神经网络的在线优化框架的2.81倍。展开更多
Focal depth is one of the most difficult seismic parameters to determine accurately in seismology. The focal depths estimated by various methods are uncertain to a considerable degree, which affects the understanding ...Focal depth is one of the most difficult seismic parameters to determine accurately in seismology. The focal depths estimated by various methods are uncertain to a considerable degree, which affects the understanding of the source process. The influence of various factors on focal depth is non-linear. The influence of epicentral distance, arrival time residual and velocity model （crust model） on focal depth is analyzed based on travel time formula of near earthquakes in this paper. When wave propagation velocity is constant, the error of focal depth increases with the increase of epicentral distance or the distance to station and the travel time residual. When the travel time residual is constant, the error of focal depth increases with the increase of the epicentral distance and the velocity of seismic wave. The study also shows that the location error perhaps becomes bigger for shallower earthquakes when the velocity is known and the travel time residual is constant. The horizontal error caused by location accuracy increases with the increase of the epieentrai distance, the travel time residual and the velocity of seismic waves, thus the error of focal depth will increase with these factors. On the other hand, the errors of focal depth will lead to change of the origin time, therefore resultant outcomes will all change.展开更多
文摘在汽车气动外形优化设计中,往往需要大量的高精度CFD数据作为支撑。然而,高精度CFD数据获取难度大、成本高。为了缓解汽车气动优化设计中气动特性评估精度和效率之间的矛盾,根据迁移学习与数据融合的思想,提出了一种基于多精度深度神经网络(multi-fidelity deep neural network, MFDNN)的汽车外形优化设计方法,以减少优化设计中所需的高精度数据个数,从而有效提升优化速度、降低优化成本。将所发展的优化方法应用于快背式MIRA标准模型减阻优化设计中,优化结果表明,该方法能够充分融合不同精度数据所蕴含的知识,加速气动外形优化进程,提升优化效率。以收敛用时作为评价指标,在取得相近或更优优化结果的前提下,基于多精度神经网络的优化框架的收敛速度是基于单精度神经网络的离线优化框架的5.85倍,是基于单精度神经网络的在线优化框架的2.81倍。
基金sponsored by the National Basic Research Program (2008CB425705)the Science Foundation for Young Scientists of CENC (404-1312)the Research on Earthquake Monitoring Rapid Prediction Capability Index System of the 12th"Five-year Plan",China
文摘Focal depth is one of the most difficult seismic parameters to determine accurately in seismology. The focal depths estimated by various methods are uncertain to a considerable degree, which affects the understanding of the source process. The influence of various factors on focal depth is non-linear. The influence of epicentral distance, arrival time residual and velocity model （crust model） on focal depth is analyzed based on travel time formula of near earthquakes in this paper. When wave propagation velocity is constant, the error of focal depth increases with the increase of epicentral distance or the distance to station and the travel time residual. When the travel time residual is constant, the error of focal depth increases with the increase of the epicentral distance and the velocity of seismic wave. The study also shows that the location error perhaps becomes bigger for shallower earthquakes when the velocity is known and the travel time residual is constant. The horizontal error caused by location accuracy increases with the increase of the epieentrai distance, the travel time residual and the velocity of seismic waves, thus the error of focal depth will increase with these factors. On the other hand, the errors of focal depth will lead to change of the origin time, therefore resultant outcomes will all change.
