一种新的选择性神经网络集成方法及其在PTA中的应用

神经网络集成可以显著提高神经网络的泛化性能。传统的集成方法中大都采用将训练的所有网络直接进行组合的方式形成集成网络,而实际上这些网络可能具有一定的相关性。为此,选择性神经网络集成成为目前研究的热点,它能够进一步提高集成网络的泛化性能。本文提出了一种利用网络权值计算网络模型之间差异度的新的选择性神经网络集成方法DWSEN。UCI数据测试表明,与流行的集成方法Bagging和Boosting比较,本方法有着更好的泛化能力和稳定性。将DWSEN应用于精对苯二甲酸(PTA)溶剂系统脱水塔装置的建模过程,结果显示,利用该方法训练得到的集成模型具有更好的泛化性能,能够较好地模拟生产运行过程。

基于齐夫法则的江苏省旅游收入规模结构变化研究

借用齐夫法则研究了江苏省近10年的旅游收入规模结构变化,并与差异度变化加以对比。结果显示无论是国内旅游收入还是国际旅游收入,规模分布均基本遵循齐夫定律,与差异度变化趋势也相一致;无标度区范围均呈扩大趋势,规模结构逐步得到优化。但两者也有不同点:(1)前者从局部单分形演变为几乎覆盖全域的单分形,而后者则从局部单分形→几乎覆盖全域的单分形→局部单分形→双分形;(2)前者是先呈帕雷托模式后呈不规则的帕雷托模式,而后者则是前几年呈不规则的帕雷托模式,后几年标度区一呈对数正态分布模式,标度区二呈不规则的帕雷托模式;(3)前者的规模变差是先扩大后缩小,而后者则是前几年规模变差略有波动,后几年两个标度区之间的差距稍有扩大,但各标度区内部的规模变差却呈缩小趋势。齐夫法则可为分析旅游收入规模结构变化提供新的定量方法。

Locating method of fire source for spontaneous combustion of sulfide ores

in order to achieve a more efficient way to accurately detect the position of the fire source of spontaneous combustion underground mine, a simple fire source locating method, based on infrared scanning system which can determine the point where the highest temperature on the surface of igniting ores occurs, was proposed. First, the differential equations that describe heat flow in ore body were presented and the relationship between the surface temperature distribution and the depth and intensity of inner fire source was established with a relatively simple heat transfer model. With the solution of equation, the expression of the relationship between the surface temperature distribution and the inner fire source was deduced and the mathematical-physical model of heat transfer process was set up. Then, with the model, visualization of fire source on the basis of MATLAB simulation platform was realized. The results show that： 1） within 10 m, when the detecting depth is less than 2 m, the temperature perturbation on ores surface can change rapidly, and then slowly; after 4 m, in contrast, it changes very little, and is even close to zero at 10 m; 2） When it is close to self-ignition duration and the detective depths are 2, 5 and 10 m, respectively, the maximum temperature differences are correspondingly 0.5, 0.04 and 0.005 ℃ in the scope of 1 m×1 m; under the same condition, the maximum temperature differences are 1.391, 0.136 and 0.018 ℃, respectively, in the scope of 2 m×2 m. Therefore, this system can be used to measure the temperature differences on the surface of ore body and determine the highest temperature point directly. Also, it is possible to determine the depth of fire source and its intensity by locating method of fire source indirectly.

Estimating extreme temperature differences in steel box girder using long-term measurement data

The extreme temperature differences in fiat steel box girder of a cable-stayed bridge were studied.Firstly,by using the long-term measurement data collected by the structural health monitoring system installed on the Runyang Cable-stayed Bridge,the daily variations as well as seasonal ones of measured temperature differences in the box girder cross-section area were summarized.The probability distribution models of temperature differences were further established and the extreme temperature differences were estimated with a return period of 100 years.Finally,the temperature difference models in cross-section area were proposed for bridge thermal design.The results show that horizontal temperature differences in top plate and vertical temperature differences between top plate and bottom plate are considerable.All the positive and negative temperature differences can be described by the weighted sum of two Weibull distributions.The maximum positive and negative horizontal temperature differences in top plate are 10.30 ℃ and -13.80 ℃,respectively.And the maximum positive and negative vertical temperature differences between top plate and bottom plate are 17.30 ℃ and-3.70 ℃,respectively.For bridge thermal design,there are two vertical temperature difference models between top plate and bottom plate,and six horizontal temperature difference models in top plate.

Reduced Fidelity Susceptibility in One-Dimensional Transverse Field Ising Model

We study critical behaviors of the reduced fidelity susceptibility for two neighboring sites in the onedimensional transverse field Ising model. It is found that the divergent behaviors of the susceptibility take the form of square of logarithm, in contrast with the global ground-state fidelity susceptibility which is power divergence. In order to perform a scaling analysis, we take the square root of the susceptibility and determine the scaling exponent analytically and the result is further confirmed by numerical calculations.

烃源岩TOC测井评价方法及应用——以珠江口盆地文昌组为例

利用纵向高分辨率的测井信息评价烃源岩的总有机碳含量(TOC),弥补了利用有限地球化学分析数据表征烃源岩有机质丰度分布特征的不足,为珠江口盆地资源量计算和油气勘探决策提供依据。通过分析岩心总有机碳含量与对应的测井响应参数之间的关系,建立总有机碳含量预测的多参数回归模型、核磁共振与密度孔隙度差异模型和多矿物反演模型。运用以上3种模型对研究区内6口井文昌组烃源岩的总有机碳含量进行预测,预测结果与地球化学分析数据具有良好的一致性,相对误差平均值小于20%。对比分析表明核磁共振与密度孔隙度差异模型的精度最高,具有较好的应用前景。