基于多保真深度神经网络的船型优化

[目的]为了提高优化效率并获得更好的优化结果,将不同精度数据进行有机融合,利用多保真深度神经网络开展船型优化设计。[方法]基于多源数据融合和迁移学习思想,构建了一种多保真深度神经网络。通过将大量低保真数据与少量高保真数据融合学习,构建与高保真数据之间的线性项和非线性项,得到高保真近似模型。基于此方法开展针对DTMB 5415船静水阻力的优化设计。分别采用势流和黏流样本点阻力进行评估,势流计算结果作为低保真数据,黏流计算结果作为高保真数据,构建多保真深度神经网络近似模型。借助遗传算法获得优化解并与只使用单一高保真数据构建的Kriging近似模型的优化结果进行对比。[结果]基于多保真神经网络方法,DTMB 5415阻力减少了6.73%。基于Kriging模型,DTMB 5415阻力减少了5.59%。[结论]多保真深度神经网络近似模型可以兼顾效率和精度,可以用于优化求解,且由其得到的优化船型阻力优化效果更为显著。

自适应多保真数据融合的神经网络模型

数据驱动的深度学习建模在力学、材料等不同学科中得到了较多应用。深度学习建模的精度依赖大量高保真数据。在实际应用中,高保真数据往往是少量且昂贵的,而低保真数据却是成本低廉且数量较多的。当高保真数据量过少时,深度学习建模精度较低。近期发展的多保真深度神经网络,通过融合不同保真度的数据,在高保真数据较少时,依然保持了较高的建模精度。然而,已有的多保真深度神经网络模型的精度较为依赖针对模型参数的正则化调节。当添加的正则化过强时,网络对非线性关联式的拟合能力不足;当添加的正则化强度不够时,在学习多保真数据间的线性关联关系时又会出现过拟合现象。两者都会严重影响模型的预测精度。在缺乏高保真验证数据集时,较难得到最优的正则化系数。为此,通过改进已有多保真网络模型的损失函数,引入一个与线性关联式相关的参数,提出了自适应多保真数据融合的神经网络模型。该模型能根据给定数据自适应地拟合不同保真度数据间的线性或非线性关系,对正则化依赖较小,从而提高了建模的鲁棒性。在多个标准测试案例及实际应用的翼型气动参数的预测中,该模型均能表现出较高的精度和稳定性。

库车坳陷阿瓦特地区高精度地震成像技术及应用

阿瓦特区块位于塔里木盆地库车坳陷克拉苏冲断带西段,地表高程变化大、起伏剧烈,地下构造极其复杂。受地表及地下地质条件的影响,地震资料叠前深度偏移处理准确成像困难。笔者针对该区特殊的地质特征,通过近真地表叠前深度偏移技术、高精度浅表层速度建模技术、特殊岩性体速度精细刻画技术、多方位网格层析成像技术等系列高精度地震成像技术的应用,大幅度提升了该区地震资料的叠前深度偏移成像品质,为后续地质解释及油气勘探开发奠定了坚实的数据基础。

散点透视“读不懂”——“后新诗潮”侧论

引言:是诗让我们“不懂”,还是我们“不懂”诗?当朦胧诗人把持手中的意象以“挑战者”的姿态打出他们“怪异”的文本品牌,当代诗坛在这些芜杂的意象群落间经历了一种“震惊体验”,“读不懂”的呼声随之雀起。而今,那些曾令人劳心费神的意象已经为诗界耳熟能详,北岛和舒婷们连同他们的意象沉潜为一种经典的所指。然而,“后新诗潮”①又重复着朦胧诗初出茅庐时的遭际。不同的是,这一次的“读不懂”

Ocean Mixing with Lead-Dependent Subgrid Scale Brine Rejection Parameterization in a Climate Model

Sea ice thickness is highly spatially variable and can cause uneven ocean heat and salt flux on subgrid scales in climate models.Previous studies have demonstrated improvements in ocean mixing simulation using parameterization schemes that distribute brine rejection directly in the upper ocean mixed layer.In this study,idealized ocean model experiments were conducted to examine modeled ocean mixing errors as a function of the lead fraction in a climate model grid.When the lead is resolved by the grid,the added salt at the sea surface will sink to the base of the mixed layer and then spread horizontally.When averaged at a climate-model grid size,this vertical distribution of added salt is lead-fraction dependent.When the lead is unresolved,the model errors were systematic leading to greater surface salinity and deeper mixed-layer depth(MLD).An empirical function was developed to revise the added-salt-related parameter n from being fixed to lead-fraction dependent.Application of this new scheme in a climate model showed significant improvement in modeled wintertime salinity and MLD as compared to series of CTD data sets in 1997/1998 and 2006/2007.The results showed the most evident improvement in modeled MLD in the Arctic Basin,similar to that using a fixed n=5,as recommended by the previous Arctic regional model study,in which the parameter n obtained is close to 5 due to the small lead fraction in the Arctic Basin in winter.

Effects of Soil Depth and Season Variation on Community Structure of Arbuscular Mycorrhizal Fungi in Greenhouse Soils Planted with Watermelon

The characteristics of arbuscular mycorrhizal fungal (AMF) community structure in various soil depths and growing seasons of watermelon (Citrullus vulgaris) grown in commercial greenhouses in Daxing of Beijing and Weifang and Laiyang of Shandong, China were investigated using both morphological identification and denaturing gradient gel electrophoresis. The sampled soils had been used for continuous greenhouse production of watermelon for 0, 5, 10, 15, or 20 years. Glomus claroideum was the dominant species in the greenhouse soils planted for 5, 10, and 15 years in Laiyang, while Glomus mosseae and Glomus etunicatum were dominant in the nearby open farmland soil. Sorenson's similarity index of AMF community composition ranged from 0.67 to 0.84 in the soils planted for 5 years, and from 0.29 to 0.33 for 20 years among the three locations. Spore abundance, species richness, and the Shannon index were highest near the soil surface (0-10 cm) and decreased with soil depth, and higher in June and October than in August and December. Canonical correspondence analysis showed that available P and the number of years that soil had been used for greenhouse production were the main factors contributing to the variance of AMF community composition. It was concluded that the community structure of AMF was mainly influenced by soil available P and planting time of watermelon as well as by soil depth and seasonal variation in the commercial greenhouse.

Loss of total phenols from leaf litter of two shrub species: dual responses to alpine forest gap disturbance during winter and the growing season

Aims Alpine forest gaps can control understory ecosystem processes by manipulating hydrothermal dynamics.Here,we aimed to test the role of alpine forest gap disturbance on total phenol loss(TPL)from the decomposing litter of two typical shrub species(willow,Salix paraplesia Schneid.,and bamboo,Fargesia nitida(Mitford)Keng f.).Methods We conducted a field litterbag experiment within a representative fir(Abies faxoniana Rehd.)forest based on‘gap openness treatments’(plot positions in the gap included the gap center south,gap center north,canopy edge,expanded edge and closed canopy).The TPL rate and litter surface microbial abundance(fungi and bacteria)of the two shrub species were measured during the following periods over 2 years:snow formation(SF),snow cover(SC),snow melting(ST),the early growing season(EG)and the late growing season(LG).Important Findings At the end of the study,we found that snow cover depth,freeze–thaw cycle frequency and the fungal copies g−1 to bacterial copies g−1 ratio had significant effects on litter TPL.The abundances of fungi and bacteria decreased from the gap center to the closed canopy during the SF,SC,ST and LG periods and showed the opposite trend during the EG periods.The rate of TPL among plot positions closely followed the same trend as microbial abundance during the first year of incubation.In addition,both species had higher rates of TPL in the gap center than at other positions during the first winter,first year and entire 2-year period.These findings suggest that alpine forest gap formation accelerates litter TPL,although litter TPL exhibits dual responses to gap disturbance during specific critical periods.In conclusion,reduced snow cover depth and duration during winter warming under projected climate change scenarios or as gaps vanish may slow litter TPL in alpine biomes.