Identification of Geologic Fault Network Geometry by Using a Grid-Based Ensemble Kalman Filter

Discrete geologic features such as faults and highly permeable embedded channels can significantly affect subsurface flow and transport characteristics. Therefore, they must be properly identified, parameterized, and represented in subsurface simulation models. In this work, we use an improved ensemble Kalman filter (EnKF) for history-matching fault network geometry from production data. EnKF is a sequential Monte Carlo data assimilation method that simultaneously propagates and updates an ensemble of model states, resulting in a set of calibrated model realizations that can be readily used for model prediction and uncertainty analysis. A pattern-based stochastic simulation algorithm was used to generate fault network realizations based on a priori fault trace data. The classic EnKF algorithm was enhanced with a grid-based covariance localization scheme to better handle non-Gaussian permeability distributions resulting from the presence of faults. Numerical experiments indicate that the modified EnKF can be a promising method for uncovering unmapped faults by using production data.     

Stochastic Interpolation of Spatial Random Fields by BF/MCF-ISM

In the past, interpolation of random fields was successfully treated by Kriging methods for Gaussian fields, and by conditional simulation techniques for a class of non-Gaussian translation fields. Recently, bootstrap filter/Monte Carlo filter (BF/MCF) is extensively used for interpolation of general non-Gaussian fields. However, while BF/MCF is a versatile tool to interpolate non-Gaussian fields, that is an algorithm of generating a set of sample realizations of both a predicted state vector and a filtered state vector, the computational cost is expensive due to the required sample size. In order to reduce the required sample size, an importance sampling function derived from the updating theory of Gaussian fields is applied to the ordinary BF/MCF. Interpolation of spatial fields is first demonstrated by using numerically simulated data, and the BF/MCF incorporated with importance sampling technique (BF/MCF-ISM) for the state estimation of conditional non-Gaussian fields is performed with respect to its efficiency in variance reduction.     

基于图像混合核的列生成PM2.5预测

传统PM_2.5预测方法获取污染物浓度数据需要大型精密仪器，成本较高。本文尝试利用图像数据进行PM_2.5浓度预测。大气PM_2.5浓度的变化与图像的暗通道强度、对比度和HSI（Hue-saturation-intensity）颜色差异有密切联系。大气中PM_2.5浓度的升高会导致非天空区域的暗通道强度值下降，图像对比度下降和HSI空间颜色差异变小。通过分析PM_2.5浓度与图像特征的关系，提出了一种基于图像混合核的列生成空气质量PM_2.5预测模型。首先，以1 h为采样周期，每日8:00～17:00为采样范围，采集多种天气条件下的景物图像，提取图像的对比度、暗通道强度和HSI颜色差异共5个图像特征。其次，数据存在样本规模大、样本不平坦分布等特点，单个核函数构成的预测模型难以满足预测精度需求，因此本文按照核结构从简单到复杂的原则，选择线性核函数、多项式核函数和高斯核函数三种核函数建立组合模型。然后计算每个核基于训练样本的Gram矩阵，并将所有Gram矩阵并列成一个混合核矩阵。利用列生成算法和混合核矩阵建立预测模型，求解模型参数。最后，进行仿真实验，实验结果表明本文提出的可满足预测精度要求，与单核预测模型相比，该预测模型预测精度更高，模型稳定性更好。计算复杂度分析结果显示基于图像混合核的列生成模型与单核预测模型相比计算量无明显增加。      

Observation and Numerical Simulations with Radar and Surface Data Assimilation for Heavy Rainfall over Central Korea

T This study investigated the impact of multiple-Doppler radar data and surface data assimilation on forecasts of heavy rainfall over the central Korean Peninsula; the Weather Research and Forecasting (WRF) model and its three-dimensional variational data assimilation system (3DVAR) were used for this purpose. During data assimilation, the WRF 3DVAR cycling mode with incremental analysis updates (IAU) was used. A maximum rainfall of 335.0 mm occurred during a 12-h period from 2100 UTC 11 July 2006 to 0900 UTC 12 July 2006. Doppler radar data showed that the heavy rainfall was due to the back-building formation of mesoscale convective systems (MCSs). New convective cells were continuously formed in the upstream region, which was characterized by a strong southwesterly low-level jet (LLJ). The LLJ also facilitated strong convergence due to horizontal wind shear, which resulted in maintenance of the storms. The assimilation of both multiple-Doppler radar and surface data improved the accuracy of precipitation forecasts and had a more positive impact on quantitative forecasting (QPF) than the assimilation of either radar data or surface data only. The back-building characteristic was successfully forecasted when the multiple-Doppler radar data and surface data were assimilated. In data assimilation experiments, the radar data helped forecast the development of convective storms responsible for heavy rainfall, and the surface data contributed to the occurrence of intensified low-level winds. The surface data played a significant role in enhancing the thermal gradient and modulating the planetary boundary layer of the model, which resulted in favorable conditions for convection.     

Improvement in Background Error Covariances Using Ensemble Forecasts for Assimilation of High-Resolution Satellite Data

Satellite data obtained over synoptic data-sparse regions such as an ocean contribute toward improving the quality of the initial state of limited-area models.Background error covariances are crucial to the proper distribution of satellite-observed information in variational data assimilation.In the NMC (National Meteorological Center) method,background error covariances are underestimated over data-sparse regions such as an ocean because of small differences between different forecast times.Thus,it is necessary to reconstruct and tune the background error covariances so as to maximize the usefulness of the satellite data for the initial state of limited-area models,especially over an ocean where there is a lack of conventional data.In this study,we attempted to estimate background error covariances so as to provide adequate error statistics for data-sparse regions by using ensemble forecasts of optimal perturbations using bred vectors.The background error covariances estimated by the ensemble method reduced the overestimation of error amplitude obtained by the NMC method.By employing an appropriate horizontal length scale to exclude spurious correlations,the ensemble method produced better results than the NMC method in the assimilation of retrieved satellite data.Because the ensemble method distributes observed information over a limited local area,it would be more useful in the analysis of high-resolution satellite data.Accordingly,the performance of forecast models can be improved over the area where the satellite data are assimilated.     

Ladle Furnace Steel Temperature Prediction Model Based on Partial Linear Regularization Networks with Sparse Representation

Partial linear regularization networks (PLRN) combined with sparse representation technique is developed to establish the steel temperature prediction model for LF. Parametric linear part is introduced into the classical regularization networks in order to fit the partial linear structured temperature model, which is obtained by analyzing the mechanism of LF thermal system in detail. Improvement in prediction accuracy is achieved due to the well learning performance of regularization networks and the modification according to the special structure. Furthermore, sparse representation technique is adopted on original PLRN for the sake of reducing computational cost and further improving the generalization performance. Learning scheme of recursive version is designed to train the sparsely represented PLRN, in which support vectors is selected one‐by‐one and recursive algorithm is adopted for computational efficiency. The proposed method is examined by practical data. The experiment results demonstrate that the proposed method can both improve the prediction accuracy and lead to sparse solution, so that it reduce the storage need and the prediction time for practical application.     

Kalman Filtering with Regional Noise to Improve Accuracy of Contaminant Transport Models

Spatially independent Gaussian noise has been widely assumed in examining the Kalman filter (KF) properties in different areas of engineering practice. However, for subsurface modeling, it is more reasonable to consider both data and noise as regional. In this study, regional noises are employed in KF and finite-difference schemes in solving the subsurface transport problem. A KF is constructed as a data assimilation scheme for a subsurface numeric model. Also, a regional random field simulation scheme is proposed and employed to examine the impact on effectiveness of KF correction processes. The results indicate that the prediction error of the KF data assimilation scheme is 30% smaller than the error from the deterministic model. Furthermore, by applying a correct regional noise structure, the KF data assimilation scheme reduces the prediction error from 25 to 10 ppm in our model, indicating an improvement of 60% in prediction accuracy.     

褐铁矿在烧结工艺中的优化配置

为了探究全进口矿条件下褐铁矿在烧结工艺中的合理配置,实现褐铁矿的高效利用以进一步提铁降本,针对S钢铁公司500 m2大型烧结机实际原燃料条件,基于试验用铁矿粉的常规理化性能和高温烧结基础特性开展了不同褐铁矿配比的烧结杯试验研究,结合Factsage 7.1热力学软件,模拟计算了不同褐铁矿配比条件下的黏附粉含量和理论液相生成量及性能,并采用矿相显微镜分析了烧结矿的显微结构,探明了褐铁矿与赤铁矿和磁铁矿的优化搭配规律。研究表明:澳大利亚褐铁矿具有粒度粗、矿化能力弱,同化温度低、黏结相强度差、吸液性强的特点,当褐铁矿质量分数由45%增加至55%时,提高磁铁精矿OD矿的质量分数至15%,同时降低OC矿质量分数至10%,烧结矿转鼓强度和低温还原粉化性能等指标达到最优,这是由于一方面提高磁铁精矿配比不仅具有增加黏附粉比例、改善液相生成数量和性能的作用,而且可以均匀液相分布,消除过熔现象;另一方面,增加磁铁精矿配比可以改善烧结料球的粒度组成,减少褐铁矿吸液量,提高烧结矿强度。因此,在高褐铁矿配比条件下,增加适宜的磁铁精矿配比有利于稳定烧结矿质量,全面改善烧结矿性能。      

Impact of Analysis-time Tuning on the Performance of the DRP-4DVar Approach

In this study we extend the dimension-reduced projection-four dimensional variational data assimilation (DRP-4DVar) approach to allow the analysis time to be tunable, so that the intervals between analysis time and observation times can be shortened. Due to the limits of the perfect-model assumption and the tangent-linear hypothesis, the analysis-time tuning is expected to have the potential to further improve analyses and forecasts.Various sensitivity experiments using the Lorenz-96 model are conducted to test the impact of analysistime tuning on the performance of the new approach under perfect and imperfect model scenarios, respectively. Comparing three DRP-4DVar schemes having the analysis time at the start, middle, and end of the assimilation window, respectively, it is found that the scheme with the analysis time in the middle of the window outperforms the others, on the whole. Moreover, the advantage of this scheme is more pronounced when a longer assimilation window is adopted or more observations are assimilated.     

Probabilistic Forecasting of Project Duration Using Kalman Filter and the Earned Value Method

The earned value method (EVM) is recognized as a viable method for evaluating and forecasting project cost performance. However, its application to schedule performance forecasting has been limited due to poor accuracy in predicting project durations. Recently, several EVM-based schedule forecasting methods were introduced. However, these are still deterministic and have large prediction errors early in the project due to small sample size. In this paper, a new forecasting method is developed based on Kalman filter and the earned schedule method. The Kalman filter forecasting method (KFFM) provides probabilistic predictions of project duration at completion and can be used from the beginning of a project without significant loss of accuracy. KFFM has been programmed in an add-in for Microsoft Excel and it can be implemented on all kinds of projects monitored by EVM or any other S-curve approach. Applications on two real projects are presented here to demonstrate the advantages of KFFM in extracting additional information from data about the status, trend, and future project schedule performance and associated risks.     

Ladle Furnace Liquid Steel Temperature Prediction Model Based on Optimally Pruned Bagging

 For accurately forecasting the liquid steel temperature in ladle furnace (LF), a novel temperature prediction model based on optimally pruned Bagging combined with modified extreme learning machine (ELM) is proposed. By analyzing the mechanism of LF thermal system, a thermal model with partial linear structure is obtained. Subsequently, modified ELM, named as partial linear extreme learning machine (PLELM), is developed to estimate the unknown coefficients and undefined function of the thermal model. Finally, a pruning Bagging method is proposed to establish the aggregated prediction model for the sake of overcoming the limitation of individual predictor and further improving the prediction performance. In the pruning procedure, AdaBoost is adopted to modify the aggregation order of the original Bagging ensembles, and a novel early stopping rule is designed to terminate the aggregation earlier. As a result, an optimal pruned Bagging ensemble is achieved, which is able to retain Bagging′s robustness against highly influential points, reduce the storage needs as well as speed up the computing time. The proposed prediction model is examined by practical data, and comparisons with other methods demonstrate that the new ensemble predictor can improve prediction accuracy, and is usually consisted compactly.     

Position of the egocentric reference and directional arm movements in right-brain-damaged patients

An increasing number of economic evaluations are being conducted alongside clinical trials. While this practice offers the prospect of collecting comprehensive and accurate cost data, it requires considerable time and effort. In the case of clinical data, key analytic decisions such as which data to collect and sample size are often made with reference to smaller (pilot) trials. However, this approach is not normally followed in the case of economic evaluation. This study was based on a recently completed health technology assessment comparing conventional radiotherapy with continuous hyperfractionated accelerated radiotherapy (CHART) for patients with head and neck cancer or carcinoma of the bronchus. In the full health technology assessment, cost data were available for 526 head and neck patients (314 CHART and 212 conventional therapy) and 286 bronchus patients (175 CHART and 109 conventional therapy). In order to simulate a pilot study, data were extracted for the patients recruited to both trials in the first 3 months. These were then compared with the full data set in order to assess whether such a pilot study would have given useful guidance on: a) the usefulness of undertaking a full study; b) the sample size required; and c) the important resource items for which comprehensive data collection would be required. Pilot studies can be helpful in determining the likely advantages of undertaking full economic evaluations and in identifying important resource items. Therefore, it is important that clinical researchers and research funding bodies create the necessary time window to enable such studies to take place. However, formal sample size calculations are more difficult to perform on limited data, since they also require knowledge of the unit cost (or prices) to be attached to the resource items and the correlation between costs and clinical effects.     

基于AdaBoost.RS算法的LF炉钢水温度预报分析

LF炉钢水温度的精准控制有利于缩短钢的冶炼时间,从而节约其生产成本。而获得准确的LF炉钢水温度预报是钢水温度控制的先决条件。通过分析LF炉冶炼过程对钢水温度的影响因素,提出一种适用于LF炉钢水温度预报同时具有增量学习功能的AdaBoost.RS集成建模算法。该算法引入松弛变量和遗忘因子2个参数,在提高预测精度的同时,可以克服大噪声数据带来的干扰,同时增量学习可以降低早期生产数据对模型的影响。以福建三钢有限责任公司100tLF炉为研究对象,采用5个测试函数验证算法的抗噪性能,分别用静态数据和动态数据对钢水出站的终点温度进行预报。实验结果表明,预测的绝对误差小于10℃的样本数量超过了样本总数的90%,算法精度较高,有利于实际生产应用。     

基于卡尔曼滤波的迟滞神经网络风速序列预测

通过将迟滞特性引入神经元激励函数的方式,构造了一种前向型迟滞神经网络模型.结合卡尔曼滤波方法,将其应用于风速时间序列的预测分析中.在原始风速时间序列的基础上,构造出风速变化率序列.采用迟滞神经网络分别对两种序列进行预测分析,并将预测结果利用卡尔曼滤波方法进行融合,从而得到最优预测估计结果.仿真实验结果表明,迟滞神经网络具有更加灵活的网络结构,能够有效改善网络的泛化能力,预测性能优于传统神经网络.采用卡尔曼滤波方法对预测结果进行融合后能够进一步提高预测精度,降低预测误差.      

基于集成神经网络的剩余寿命预测

针对机器或设备的剩余寿命（Remaining useful life, RUL）预测精度低的问题,提出基于一维卷积神经网络（Convolutional neural network, CNN）和双向长短期记忆（Bidirectional long short-term memory, BD-LSTM）的集成神经网络模型。为了更好地抽取时间序列上的特征,以及产生更多的训练样本,采用滑动窗口对数据进行处理,同时采用卡尔曼滤波对数据进行降噪处理,将数据标准化以及设置RUL标签。与人工提取特征不同,利用一维CNN对数据进行特征提取,并舍弃了CNN中的池化层。然后将提取到的高维特征输入到BD-LSTM进行回归预测,并采用Bagging的方式对此神经网络进行集成来预测RUL。最后通过在NASA的数据集上验证该模型的有效性,以及相比于其他机器学习或者深度学习模型的优越性,实验表明所提模型在RUL预测方面更加准确。      

冷轧不锈钢连续退火在线检测

为快速反馈不锈钢带退火质量,减少退火工艺调整时间,降低取样费用,根据奥氏体不锈钢和铁素体不锈钢的特点,用超声波和剩磁法在线检测退火后晶粒尺寸,有效控制连续退火酸洗线带钢退火性能,目前已在国内外多条冷轧不锈钢连续退火酸洗线上得到了成功应用。     

面向材料数据的主动回归学习方法

材料的生产环境和测量条件不同,导致用于机器学习的材料数据的噪声较大.对材料数据进行标注需要一定的专业知识和专业技能,因此标注成本也相对较高.这两方面的因素给机器学习应用于材料领域带来了巨大挑战.为应对这个挑战,提出了一个主动回归学习方法,由离群点检测模块、贪婪采样模块和最小变化采样模块组成.同其他主动学习方法相比,该方法整合了离群点检测机制,选取高质量样本的同时有效地排除了噪声数据的影响,避免了沉没成本.在公开数据集和非公开数据集上与最新的主动回归学习方法进行了对比实验,实验结果表明本文方法在相同的数据量下训练的任务模型性能指标相比于其他模型平均提高15%,且只需30%～40%的数据量作为训练集就可以达到甚至超过使用全部数据训练任务模型的精度.     

Searching for one or many targets: Effects of extended experience on the runs advantage.

Pigeons searched for any of several targets in experiments designed to explore attentional priming or search-image formation. Three experiments assessed the runs advantage, seen in improved search efficiency during single-target trial sequences compared with search efficiency during mixed-target sequences. Experiment 1 varied the number of items in the search display and the target set. Well-practiced pigeons showed the runs advantage only for a display size of 36 and a target set of 8. Experiments 2 and 3 followed the runs effect over training. For a target set of 4, the runs advantage was present initially but diminished with practice. For a target set of 12, the runs advantage persisted. The data suggest that learning permits a shift from controlled to automatic processing of memory, but constrains the number of items that can be addressed in parallel. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Enhanced Criterion for Base Soil Retention in Embankment Dam Filters

In effective filters, potentially erodible base particles are transported to the filter and retained to form a stable self-filtration layer. At any given time, the mass proportion of the filter and the base materials in this layer depends on the initial porosity of the filter and the subsequent porosity of the self-filtration layer. In this paper, an analytical procedure is given to obtain the particle size distribution (PSD) of the self-filtration layer by combining the PSDs of the filter and the base soil modified by Dc95, where 95% of filter constrictions are finer than the size denoted by Dc95. The assessment of internal stability of the PSD of the self-filtration layer forms a rational model to successfully identify the effective filters from their ineffective counterparts. The proposed model is verified by large-scale laboratory tests carried out by the writers in addition to other published data. The model performance is acceptable in relation to various base and filter materials, and provides an alternative and rigorous design approach by eliminating most limitations of the conventional particle based criteria (e.g., D15/d85 ratio).     

铁矿粉与CaO同化反应性能的表征方法

为解决在研究铁矿粉与CaO进行同化反应时通常采用最低同化温度这一终点体系指标来表征,而没有考虑同化反应速度和升温速率的缺陷以及采用多个指标来表征同化反应温度和速度在使用上的不便,本文充分利用铁矿粉在高温同化反应过程中获得的数据和范特霍夫规则,提出了量纲为1的同化反应特征数.该特征数包括同化反应速度、升温速率、同化温度和温度对同化反应速度的影响,可用来表征铁矿粉与CaO的同化过程中铁矿粉的同化温度、同化反应时间、同化反应结束温度等性能,并以烧结过程中的实际温度作为基准.该特征数综合了铁矿粉与CaO在同化反应过程中各重要信息,可以更全面地反映铁矿粉与CaO的同化反应性能.对两种方法得到的结果进行了比较,发现仅考虑最低同化温度这一终点指标时得到的结果相近,而采用同化反应特征数可以进一步细分出不同铁矿粉同化反应性能的差异.