粘弹性方程的有限元超收敛结果

研究粘弹性方程有限元近似解和真解Ritz-Sobolev投影之间的超收敛结果,当有限元空间指数k≥2时,得到了二者之间的L(p2≤p≤∞)模超收敛一阶,W1(,p2≤p<∞)模超收敛二阶,W1,∞模超收敛几乎二阶结果。     

基于可解释机器学习的粮食主产区农村居民点演化特征及其影响因素分析——以华中地区为例

分析粮食主产区的农村居民点演化特征及影响因素,对粮食主产区农村居民点布局优化和国家粮食安全具有参考价值。综合运用景观格局指数、空间关联指数等方法,分析了1980—2015年华中粮食主产区农村居民点用地时空变化特征,并采用极限梯度提升决策树(XGBoost),结合SHAP模型解析其主导影响因素及影响机制。结果表明：华中地区农村居民点用地整体呈扩张趋势,新增居民点用地主要表现为对耕地的占用,农村居民点聚集分布态势明显,主要以局部集聚的形式分布,整体分布密度相对较低;斑块扩张态势明显,破碎程度增加,且存在显著的高值聚集现象;整体分布呈现随距离道路、水系距离的增加而明显衰减的状态;XGBoost结果显示较传统线性回归模型有更好的拟合性,社会经济因素中人口因素的影响是农村居民点用地规模扩张的主要影响因素。     

Enhanced fatigue property by fabricating a gradient nanostructured surface layer in a reduced-activation steel

Reduced activation ferritic/martensitic (RAFM) steels have been selected as candidate structural materials for future advanced nuclear power systems. In the present work, the influence of a gradient nanograined surface layer on the fatigue properties of RAFM steels was studied. A gradient nanostructured (GNS) surface layer with a thickness of ～85 ?μm was prepared on RAFM steel utilizing surface mechanical rolling treatment (SMRT). The mean grain size was approximate 43 ?nm at the topmost surface and increased gradually with depth. The results of the stress-controlled tension-compression fatigue experiments showed that the fatigue life enhanced approximately 6 times in the SMRT samples compared to the corresponding base metal counterparts. The relationship between the applied stress amplitude and the fatigue lifetime, and the fracture morphology showed that the surface strengthening and strain delocalization were caused by GNS, which suppressed surface crack initiation process, and hence the fatigue properties of RAFM steels improved. In addition, the deformation compatibility in GNS and coarse-grained boundaries leading to more dislocation interactions and accumulation during the cyclic process, also plays a crucial role in enhancing the fatigue properties of RAFM steel.     

基于XGBoost模型的营养成分分析高血压预测方案

高血压是一种常见的慢性病，若能早发现、早采取措施可降低其引发并发症的风险。尽管高血压的产生与发展与诸多因素有关，但饮食被公认为影响高血压的主要因素之一。机器学习模型可以对疾病进行有效预测，并提供辅助治疗。笔者提出一种基于XGBoost的通过分析营养成分预测高血压的方案，该方案由数据转换、特征选择、数据清理与标准化、模型搭建、分类与评估5部分组成。实验结果表明，XGBoost在高血压预测中获得了0.859的F1分数且准确率超过85%,高于随机森林、支持向量机与人工神经网络。此外通过分析不同营养成分对高血压预测的影响因素，获得了影响高血压的前10个营养特征，大部分与医学结论相同，验证了模型的有效性。