Accounting for Model Uncertainty in the Prediction of University Graduation Rates

Empirical analysis requires researchers to choose which variables to use as controls in their models. Theory should dictate this choice, yet often in social science there are several theories that may suggest the inclusion or exclusion of certain variables as controls. The result of this is that researchers may use different variables in their models and come to disparate conclusions with respect to predicted effects and their statistical significance. In such cases one is uncertain of which particular set of regressors forms the model that represents the data. The approach used below accounts for uncertainty in variable selection by using Bayesian model averaging (BMA). Accounting for uncertainty, we demonstrate that BMA provides better out-of-sample prediction for university graduation rates than results based on alternative variable selection methods.     

杂交稻分蘖和乳熟期~(32)P、~(14)C 分布与分蘖数和谷产量

8个杂交水稻组合的分蘖数,与分蘖期~(32)P及~(14)C在分蘖的分布(％)呈显著正相关,相关系数分别为r=0.569(n=21),r=0.532(n=22)。谷粒产量与分蘖期~(32)P在分蘖分布(％)的相关系数r=0.604(n=14)。谷粒产量与乳熟期~(14)C—葡萄糖同化物在稻穗分布(％)呈显著正相关,r=0.616(n=25)。高产杂交水稻威优64、汕优66、源优植和广优56的谷粒产量与~(14)C在乳熟期稻穗的分布和从剑叶的输出率相关系数大,低产杂交水稻汕A×芦36—1,D优66等相关系数小。讨论了~(32)P、~(14)C在分蘖和稻穗的分布动态,可作为预测杂交水稻产量的指标。     

小康示范村总体规划中人口规模预测研究——以南阳新野县李湖村为例

首先分析了影响村镇人口规模的主要因素有计划生育政策、环境因素、产业结构、行政因素等，然后对常用人口预测方法进行了评价，在此基础上以南阳新野县李湖小康示范村为例，根据得到的资料，结合规划区的实际情况和规划人员的能力选择适宜的方法进行预测，通过比较综合，得到规划期的人口规模。     

基于神经网络的数控机床可靠性预计研究

本文提出了利用人工神经网络来辅助实现数控机床的可靠性预计的方法，建立了用于数控机床可靠性预测的三层BP神经网络模型，给出了具体的算法，并通过实例证明该方法比传统的数学模型预计方法更准确和可靠。     

基于XGBoost算法的社交网络链路预测

提升链路预测精度是复杂网路研究的基础问题之一。传统基于局部信息相似性、基于全局信息相似性与基于随机游走相似性的链路预测都是基于单个相似性指标进行预测的,而没有充分利用这些相似性指标的综合信息。将链路预测问题看作机器学习中的二分类问题,将有连接的样本标签记为1,无连接的样本标签记为0,将基于局部信息、基于全局信息与基于随机游走相似性等15个指标作为样本特征。综合考虑以上信息,使用XGBoost算法,选取AUC作为模型评价准则,在facebook真实数据集上进行实验。结果表明,该算法在测试集上的AUC高于基于单个相似性指标链路预测的AUC。     

蠕变条件下完整岩石的起裂和扩展准则

The mechanism and criterion of crack initiation and propagation of rocks were investigated by many researchers,And the creep behaviour of rocks was also theoretically and experimentally studied by some scientists and engineers.The characteristics of crack initation and propagation of rocks under creep condition.however,are very improtant for rock engineering and still not paid enough attention by researchers,In this paper,the criterion and mechanism of crack initiation and propagation under creep condition were investigated using specimens collected from sandstone rock formations outcropping in the Emei Mountain,the Sichuan Province of China.Cuboid specimens under three point bending were used in this investigation.All specimens were classified into four sorts and used for Mode-I fracutre of creep frcture tests.The experimental result shows that due to creep deformation.rock crack will inevitably initialt and propagate under a load of KI,which is less than fracture toughness KIC but not less than a constant(marked as KIC2),KIC2 indicates the ability of rock to resist crack initiation and propagation under creep conditions and is less than fracture tough ness KIC.defined as creep fracture toughness in this paper,KIC2 should be considered as an importnat parameter on design and computation of rock engineering.The microstructureal mechanism for crack initiation and propagation of rock materials under creep condition was introduced based on competitive model between softening effect and hardening effect,and the validity of test result was explained.The test result was also verified in rheological theory.When KI is more than KIC2 but less than KIC,rock crack will initiate and propagate after a time interval of sustained loading under creep condition.In order to find the relation between duration of sustained lading.which can lead to crack initiation and propagation,and the initial stress intensity factor KI,an unequal0interval time sequence forecasting and predicting model was introduced,and the relation was obtained for homogeneous and isotropic fine-grained red sandstone.Finally a modified fracture toughness formula was given,in which the influence of fracture process zone(FPZ) was fully considered.     

改进深度置信网络在城市用水量预测中的应用

城市用水量的准确预测可以为供水管网智能调度、异常报警提供支持,便于及时发现漏损、排查及检修,具有极大的现实意义与经济利益。针对现有用水量预测方法忽视用水量数据自身特征及不能模拟更复杂的数学运算的问题,提出一种改进深度置信网络（DBN）的用水量预测方法。对有高斯分布的连续受限玻尔兹曼机（CRBM）引入稀疏正则项,解决特征同质化现象的同时也适用于用水量数据输入。实验结果表明,在实际用水量预测中,改进DBN模型相比传统神经网络和传统DBN预测模型,预测准确率得到了较大的提高。     

改进的灰色模型在基础沉降预测中的应用

一般的灰色残差修正GM(1,1)模型只是单纯利用差分来代替微分,而且用原始沉降数据第一点的值作为时间响应函数的初始值,导致了一定的误差。提出一种多项式逼近法改进后的残差修正GM(1,1)模型,该模型利用多项式在离散观测点上的导数代替常用的差分,同时添加一个初值参数C_0做出调整。实验验证,改进模型具有较好的准确性和工程应用价值。     

云南省产业结构现状分析及发展趋势预测

本文通过调查我国云南省及云南省各地州建国以来产业结构的变动情况，利用三次产业分类法对云南省产业结构的发展和现状进行了分析，并对云南省未来25年产业结构的发展作出预测。