数据故事化解释中分类型预测结果的反转点识别方法研究——基于LIME算法

[目的/意义]实现数据故事化中的反转点识别,有助于非专业人士理解分类型预测结果的产生原因,同时对于推动故事情节发展并使其快速到达故事高潮点具有促进作用。[方法/过程]提出故事点与反转点概念,基于LIME解释技术和反转点识别过程,设计了用于数据故事化中反转点识别的算法方案,并提出了面向分类模型的反转点识别流程。[结果/结论]将反转点识别算法应用到贷款数据集,证明此算法在数据故事化过程中寻找反转点的有效性,在获得用户期望的预测结果和快速识别反转点方面具有应用价值。

Multi-scale analysis of carbon mineralization in lime-treated soils considering soil mineralogy

Mineral carbonation is emerging as a reliable CO_(2) capture technology that can mitigate climate change.In lime-treated clayey soils,mineral carbonation occurs through the carbonation of free lime and cementitious products derived from pozzolanic reactions.The kinetics of the reactions in lime-treated clayey soils are variable and depend primarily on soil mineralogy.The present study demonstrates the role of soil mineralogy in CO_(2) capture and the subsequent changes caused by carbon mineralization in terms of the unconfined compressive strength(UCS)of lime-treated soils during their service life.Three clayey soils(kaolin,bentonite,and silty clay)with different mineralogical characteristics were treated with 4%lime content,and the samples were cured in a controlled environment for 7 d,90 d,180 d,and 365 d.After the specified curing periods,the samples were exposed to CO_(2) in a carbonation cell for 7 d.The non-carbonated samples purged with N2 gas were used as a benchmark to compare the mechanical,chemical-mineralogical,and microstructure changes caused by carbonation reactions.Experimental investigations indicated that exposure to CO_(2) resulted in an average increase of 10%in the UCS of limetreated bentonite,whereas the strength of lime-treated kaolin and silty clay was reduced by an average of 35%.The chemical and microstructural analyses revealed that the precipitated carbonates effectively filled the macropores of the treated bentonite,compared to the inadequate cementation caused by pozzolanic reactions,resulting in strength enhancement.In contrast,strength loss in lime-treated kaolin and silty clay was attributed to the carbonation of cementitious phases and partly to the tensile stress induced by carbonate precipitation.In terms of carbon mineralization prospects,lime-treated kaolin exhibited maximum carbonation due to the higher availability of unreacted lime.The results suggest that,in addition to the increase in compressive strength,adequate calcium-bearing phases and macropores determine the efficiency of carbon mineralization in lime-treated clayey soils.

Performance evaluation of laterite soil embankment stabilized with bottom ash,coir fiber,and lime

In tropical regions,heavy rainfall induces erosion and shallow landslides on road embankments.Cement-based stabilization methods,common in these regions,contribute to climate change due to their high carbon footprint.This study explored the potential application of coir fiber-reinforced laterite soil-bottom ash mixtures as embankment materials in the tropics.The objective is to enhance engineered embankment slopes'erosion resistance and stability while offering reuse options for industrial byproducts.This study examined various mix designs for unconfined compressive strength(UCS)and permeability,utilizing 30%bottom ash(BA)and 1%coir fiber(CF)with varying sizes ranging from 10 to 40 mm,6%lime,and laterite soil(LS),followed by microstructural analyses.The results demonstrate that the compressive strength increases as the CF length increases to 25 mm.In contrast,permeability increases continuously with increasing CF length.Lime-treated mixtures exhibit superior short-and long-term strength and reduce permeability owing to the formation of cementitious materials,as confirmed by microstructural analyses.A lab-scale slope box was constructed to evaluate the surface erosion of the stabilized laterite soil embankment.Based on the rainfall simulation results,the LS-BA-CF mixtures show better resistance to erosion and deformation compared to untreated LS,especially when lime is added to the top layer.This study provides insights into a sustainable and cost-effective approach for slope stabilization using BA and CF,offering a promising solution for tropical regions susceptible to surface erosion and landslides.

Prediction of lime utilization ratio of dephosphorization in BOF steelmaking based on online sequential extreme learning machine with forgetting mechanism

The machine learning models of multiple linear regression(MLR),support vector regression(SVR),and extreme learning ma-chine(ELM)and the proposed ELM models of online sequential ELM(OS-ELM)and OS-ELM with forgetting mechanism(FOS-ELM)are applied in the prediction of the lime utilization ratio of dephosphorization in the basic oxygen furnace steelmaking process.The ELM model exhibites the best performance compared with the models of MLR and SVR.OS-ELM and FOS-ELM are applied for sequential learning and model updating.The optimal number of samples in validity term of the FOS-ELM model is determined to be 1500,with the smallest population mean absolute relative error(MARE)value of 0.058226 for the population.The variable importance analysis reveals lime weight,initial P content,and hot metal weight as the most important variables for the lime utilization ratio.The lime utilization ratio increases with the decrease in lime weight and the increases in the initial P content and hot metal weight.A prediction system based on FOS-ELM is applied in actual industrial production for one month.The hit ratios of the predicted lime utilization ratio in the error ranges of±1%,±3%,and±5%are 61.16%,90.63%,and 94.11%,respectively.The coefficient of determination,MARE,and root mean square error are 0.8670,0.06823,and 1.4265,respectively.The system exhibits desirable performance for applications in actual industrial pro-duction.

Cement and Lime Stabilization Effect on the Evolutivity of an Expansive Overconsolidated Clay

This paper presents and analyzes the results of a series of compaction,fragmentability and damage tests performed on an expansive overconsolidated clay treated with cement and lime.This clay was obtained from the urban site of Sidi-Hadjrès city(wilaya of M'sila,Algeria),where significant damages frequently appears in the road infrastructures,roadway systems and light structures.Tests results obtained show that the geotechnical parameters values deduced from these tests are concordant and confirm the evolutivity of this natural clay treated with composed Portland cement or extinct lime and compacted under optimum Proctor conditions.

Sustainability of lime stabilized road subgrade in mountainous regions of Türkiye

In this paper the use of lime stabilized subgrade for low volume roads in two regions with high mountains and different frost penetration conditions in Türkiye was investigated in terms of design,performance,and cost.Pavements on unstabilized and stabilized subgrade were designed for two regions(Izmir and Van),covering all climate variations.The resilient modulus of the lime stabilized subgrade with different soil pulverization levels for non-freezing and freezing conditions were taken from a previous laboratory study.Frost effects were considered in pavement design using two different approaches,including limited subgrade frost penetration method and reduced subgrade strength method.Detailed application and evaluation were performed for all steps.Lime stabilized subgrades significantly reduced the thickness of base courses,and the benefit of lime stabilization was highly dependent on soil pulverization level.A detailed cost analysis on the unstabilized and stabilized cases found that the use of lime stabilization in the subgrade provided significant initial cost savings.Comparative analysis by using the AASHTO(1993)method and KENPAVE software,and quantity effect of soil pulverization level on the performance of low volume roads from a service life perspective,show that subgrade resilient modulus can be estimated.It is also possible to make correct performance estimation in the field.The results of the study show that lime stabilization is a good solution for low volume roads in the mountainous regions of Türkiye.

Preliminary Study on the Treatment Efficiency of Pasteurized Lime Thermal Alkaline Hydrolysis for Excess Activated Sludge and Reduction of Tetracycline Resistance Genes

Thermal alkaline hydrolysis is a common pretreatment method for the utilization of excess activated sludge(EAS).Owing to strict environment laws and need for better energy utilization,new methods were developed in this study to improve the efficiency of pretreatment method.Direct thermal hydrolysis(TH),pasteurized thermal hydrolysis(PTH),and alkaline pasteurized thermal hydrolysis(PTH+CaO and PTH+NaOH)methods were used to treat EAS.Each method was compared and analyzed in terms of dissolution in ammonium nitrogen(NH_(4)^(+)-N)and soluble COD(SCOD)in EAS.Furthermore,the removal of tetracycline resistance genes(TRGs)and class 1 transposon gene intI1 from EAS was investigated.The NH_(4)^(+)-N and SCOD concentrations in EAS treated by PTH were 1.24 and 2.58 times higher than those of TH.However,the removal efficiency of total TRGs and intI1 between the groups was comparable.The SCOD concentration of the PTH+NaOH group was 4.37 times higher than that of the PTH group,and the removal efficiency of total TRGs was increased by 9.52%compared with that by PTH.The NH_(4)^(+)-N and SCOD concentrations of the PTH+CaO group could reach 85.04%and 92.14%of the PTH+NaOH group,but the removal efficiency of total TRGs by PTH+CaO was 19.78%lower than that by PTH+NaOH.Thus,to reduce the financial cost in actual operation,lime(CaO)can be used instead of a strong alkali(NaOH),and pasteurized steam at 70℃ instead of conventional high-temperature heating to treat EAS.This study provides a reference for the development of alkaline hydrolysis under moderate temperatures along with the removal of TRGs in EAS.

CI Acid Orange 52 Dye Removal Using Natural and Formulated Clay-Lime Materials: Isotherm, Kinetic and Thermodynamic Studies

The main objective of the study is to improve the removal efficiency of Ourlago-kaolin (Kao), sodium montmorillonite (Na-MMT), and two formulated clay-lime (F13 and F23) towards CI Acid Orange 52 dye (AO52). F13 and F23 were obtained by chemical stabilization through thermal treatment at 300°C. Fourier Transform Infrared spectra showed different surface functional groups on the clay materials, X-ray diffraction patterns revealed the raw materials contain many crystalline phases, scanning electron microscopy micrographs showed the variation of the layered structures of different clay materials, energy dispersive X-Ray analysis micrographs revealed compositional information and thermogravimetric-differential scanning calorimetry curves indicated the higher weight loss of 11.26% and 11.38% were observed for F13 and F23 respectively. BET surface area analyzed gave 133.0071 m2•g−1 for F13 and 132.34803 m2•g−1 for F23. The optimum pH value was 2.0 for Kao and Na-MMT. The adsorption experiments indicated that F13 and F23 have the maximum uptake abilities of 7.8740 and 3.1645 mg•g−1, respectively, compared to Kao (0.8761 mg•g−1) and Na-MMT (2.6178 mg•g−1). The pseudo-second-order model well described the adsorption kinetic model of AO52 dye onto the overall samples;Langmuir and Freundlich’s isotherms appropriately described the uptake mechanism. The positive values of ∆G° and negative value ∆H° indicated that the adsorption process was spontaneous and endothermic for Na-MMT, and non-spontaneous and exothermic for Kao, F13, and F23 because of their positive values of ∆G° and negative value of ∆H°. The modified clays have higher adsorption capacities and better life cycles compared hence opening new avenues for efficient wastewater treatment.

基于LIME算法的低照度图像增强方法研究

由于低照度图像存在颜色退化、曝光不足等问题,影响图像质量与信息量,因此文章研究了一种基于LIME算法的低照度图像增强方法。通过合成正常曝光图像与低照度图像的数据集,提取了低照度图像的纹理特征与频域特征,并基于Retinex理论与深度学习建立了低照度图像增强模型。通过使用LIME算法提高模型的可解释性,同时减少模型复杂度,提高模型的透明度。在测试集LOL上,PSNR值为21.898、SSIM值为0.902,证明了该方法的可行性和可靠性。

基于前背景特性的逆光交通场景图像增强

由于逆光交通场景下采集到的图像存在前景亮度过低、背景亮度失真等特点,导致采集图像清晰度低、信息丢失严重和可识别性差,针对逆光图像前背景的不同特性,研究提出了一种分区域增强方法。对逆光图像使用最大类间方差法(OTSU法)分割图像的前景和背景;对逆光图像全局使用LIME方法,提升前景亮度的同时保持色彩的失真度;单独将背景部分RGB三个通道上的全局直方图均衡化结果一一映射到对应的限定区间内,提高了背景的对比度;使用Canny算子检测出前背景拼接处的黑边,根据黑边生成3个自适应的滤波模板对黑边进行分步均值滤波处理,消除了黑色边缘,提升了图像的视觉质量。在实验室自建的CHD_B数据集上,所提方法在4种常用客观评价指标上综合占优。实验结果表明,所提出的图像增强算法能有效地消除图像中的逆光现象。

Effects of Phosphate and Lime Application on Growth of Trifolium repens,Chamaecrista rotundifolia and Macroptilium atropurpureum in Red Soils

[Objective] The aim was to research effects of P fertilizer and lime on growth of Trrifolium repens, Chamaecrista rotundifolia and Macroptilium atropur- pureum, to provide references for cultivation of the three plants. [Method] Pot experiments were conducted with Trrifolium repens, Charnaecrista rotundifolia and Macroptilium atropurpureum in 2010 in order to research effects of lime and P fer-tilizer mixture on growth of the plants in southern hilly acidic red soils. [Result] With lime amount fixed, application of P fertilizer would enhance plant height, total tiller number and dry matter. When P fertilizer was not applied, however, plant height of the three plants achieved the peak by lime at 1.4 g/kg which proved best for improvement of acidity of red soils. With P fertilizer at 200 mg/kg was applied, biomass of Trifolium repens and Macroptilium atropurpureum achieved the highest by lime at 2.1 g/kg, but total biomass of Chamaecrista rotundifolia was the highest by lime at 1.4 g/kg. [Conclusion] The research provides references for planting and production of Trifolium repens, Chamaecrista rotundifolia and Macroptilium atropur-pureum in southern hilly regions.

基于暂态时-频特征差异的配电网高阻接地故障识别方法

高阻接地故障发生时,故障特征微弱,传统故障识别方法存在特征提取困难、阈值选取灵活性较差的技术瓶颈,导致极端故障场景下出现漏判。为此,提出基于暂态时-频特征差异的配电网高阻接地故障识别方法。首先,结合小波包香农熵量化分析高阻接地故障与正常扰动工况暂态信号的时频分布,发现二者存在显著差异:频域上,扰动工况信号的能量集中于低频,而高阻故障信号能量分布相对均匀;时域上,扰动工况信号能量集中于时间窗的前半段,高阻故障信号能量在整个时间窗内均匀分布。在此基础上,以暂态信号时-频域波形作为输入样本,将传统卷积神经网络(convolutional neural networks,CNN)模型中的softmax分类器改进为支持向量机(support vector machine,SVM)分类器,构建适应配电网高阻接地故障识别小样本场景下的CNN-SVM复合分类模型,以卷积层作为特征提取器,以SVM作为分类器,实现高阻接地故障识别。最后,为论证所提方法具有强适应性的内在原因,利用LIME可解释性分析算法可视化展现模型训练过程中的高关注度区域,从模型分类原理层面证明所提方法不受各种故障条件的影响,克服了传统故障识别方法在极端故障场景下出现漏判的缺陷,能准确识别配电线路末端10 kΩ高阻接地故障。

Study on Food-making Quality of Strong-gluten Wheat Varieties from Lime Concretion Black Soil Area in the Huanghuai Wheat Region

In this study, 13 strong-gluten wheat varieties screened by the Key Project of Modern Agricultural Industry Technology System ＂Study on Industrial Technology for Strong-gluten Wheat from Lime Concretion Black Soil Area in the Huanghuai Wheat Region＂ were used as experimental materials to investigate their bread-making quality, noodle-making quality and other related characteristics. The results showed that more than half of the wheat varieties had better bread-making quality; the bread made from wheat with longer dough mixing time than 3.0 min had better texture, lighter color, and better taste. All these 13 strong-gluten wheat varieties showed good noodle-making quality in color, appearance, smoothness and taste; the differences between varieties were mainly found in palatability and viscoelasticity. Jimai 20, Xinong 979, Zhengmai 7698, Ji＇nan 17 and Zhengmai 9023 exhibited excellent bread-making quality; Zhengmai 366, Jimai 20 and Xinong 979 displayed excellent noodle-making quality. Fresh dough sheets made from Zhengmai 366, Jimai 20 and Xinong 979 exhibited slight color variation within 24 h and high peak starch paste viscosity; dry and cooked noodles made from Zhengmai 366, Jimai 20 and Xinong 979 had good quality.

The Electrical Resistivity Characteristics of Cement-Soil and Flyash-Lime-Soil

The electrical resistivity characteristics of cement soil and flyash lime soil are investigated in the laboratory and the field. It is shown that the electrical resistivities of the cement soil and flyash lime soil are sensitive to water content, degree of saturation and unconfined strength. The cement soil and flyash lime soil with higher water content, greater degree of saturation, lower unconfined strength has lower electrical resistivity. Electrical resistivity is also correlated with additives. Based on the tests, it is concluded that the electrical resistivity method is available for checking the effectiveness of the soil improvement by the cement soil and flyash lime soil mixing pile in terms of engineering practice.

Biopolymer stabilization of clayey soil

This study investigates the efficacy of sodium alginate(SA),xanthan gum(XG),guar gum(GG)and chitosan(CS)d each applied at five different solid biopolymer-to-water mass ratios(or dosages)and cured for 7 d and 28 d d on the unconfined compressive strength(UCS)performance of a high plasticity clayey soil.Moreover,on identifying the optimum biopolymer-treatment scenarios,their performance was compared against conventional stabilization using hydrated lime.For a given curing time,the UCS for all biopolymers followed a riseefall trend with increasing biopolymer dosage,peaking at an optimum dosage and then subsequently decreasing,such that all biopolymer-stabilized samples mobilized higher UCS values compared to the unamended soil.The optimum dosage was found to be 1.5%for SA,XG and CS,while a notably lower dosage of 0.5%was deemed optimum for GG.Similarly,for a given biopolymer type and dosage,increasing the curing time from 7 d to 28 d further enhanced the UCS,with the achieved improvements being generally more pronounced for XG-and CS-treated cases.None of the investigated biopolymers was able to produce UCS improvements equivalent to those obtained by the 28-d soilelime samples;however,the optimum XG,GG and CS dosages,particularly after 28 d of curing,were easily able to replicate 7-d lime stabilization outcomes achieved with as high as twice the soil’s lime demand.Finally,the fundamental principles of clay chemistry,in conjunction with the soil mechanics framework,were employed to identify and discuss the clayebiopolymer stabilization mechanisms.

面向网络欺凌文本检测模型的算法解释及其故事化呈现研究

[目的/意义]利用可解释性技术与故事化方法研究网络欺凌检测有助于识别欺凌内容、参与网络言论治理以及净化网络生态。[方法/过程]详细分析LIME解释算法的选择依据与文本解释原理,并提出LIME算法在网络欺凌检测模型中的“钻石结构式”应用流程,进一步构建了“数据层—模型层—解释层—叙事层”的故事化呈现框架,最后通过实验验证了该框架的有效性。[结果/结论]可解释性技术辅助网络欺凌检测模型有利于提高模型应用价值与可信度,权衡模型准确性与可解释性二者的关系,而基于数据分析及解释结果的故事化呈现方法为网络信息平台数据治理提供可信、可靠、可理解依据。

Explainable Deep Fake Framework for Images Creation and Classification

Deep learning is a practical and efficient technique that has been used extensively in many domains. Using deep learning technology, deepfakes create fake images of a person that people cannot distinguish from the real one. Recently, many researchers have focused on understanding how deepkakes work and detecting using deep learning approaches. This paper introduces an explainable deepfake framework for images creation and classification. The framework consists of three main parts: the first approach is called Instant ID which is used to create deepfacke images from the original one;the second approach called Xception classifies the real and deepfake images;the third approach called Local Interpretable Model (LIME) provides a method for interpreting the predictions of any machine learning model in a local and interpretable manner. Our study proposes deepfake approach that achieves 100% precision and 100% accuracy for deepfake creation and classification. Furthermore, the results highlight the superior performance of the proposed model in deep fake creation and classification.

基于可解释关系模型的SCR入口气温预测

新能源在全社会发电中占比不断增高,为了能够辅助电网进行削峰填谷,火电机组需要增加在低负荷范围内的运行时间,这会对选择性催化还原(selective catalytic reduction,SCR)的脱硝效果产生负面影响。因此,对SCR入口烟气温度(SCR gas temperature,SCRIGT)进行准确预测非常重要。首先,采用极限梯度提升(extreme gradient boosting,XGBOOST)模型,以电厂的运行参数作为输入,功率与SCRIGT的比值作为输出进行预测,然后计算得到SCRIGT。结果显示,对于两个不同锅炉类型的火电机组,平均绝对百分比误差(mean absolute percentage error,MAPE)分别为3.07%和2.49%。其次,采用不可知模型的局部可解释(local interpretable model-agnostic explanations,LIME)算法分析XGBOOST模型的预测结果,显示功率和功率与SCRIGT的比值之间存在线性关系,且R2(R-squared)为0.994,基于此构建了一种可解释关系模型进行SCRIGT预测。最后,比较分析显示,对于试验中的两个机组,可解释关系模型预测结果的MAPE分别改进至0.68%和0.97%。

25CrMo4合金疲劳寿命预测局部可解释性分析

在使用机器学习的方法对25CrMo4合金进行疲劳寿命预测时,特征变量对疲劳寿命的决策行为一直是关注的重点。基于此,以89组25CrMo4合金高周循环对称疲劳试验样本作为研究对象,使用XGBoost算法预测其疲劳寿命,再使用LIME(Local Interpretable Model-agnostic Explanations)及其改进方法(ElasticNet-LIME)分析各特征变量在不同数值范围内对疲劳寿命的正负增益。结果表明,经过微喷丸强化后25CrMo4合金表面残余应力减小,表面粗糙度增大,疲劳寿命增大,ElasticNet-LIME对此判断比LIME更为合理。研究结果可为25CrMo4合金疲劳性能优化提供依据。

利用LIME对脓毒症预测模型进行可解释性分析

针对机器学习应用于脓毒症预测存在预测准确率低和可解释性不足的问题,提出了利用LIME对基于机器学习的脓毒症预测模型进行可解释性分析。模型由预测和解释两部分组成:预测部分使用XGBoost和线性回归(LR),首先通过XGBoost进行特征提取,再利用LR对提取到的特征进行分类;解释部分使用LIME模型提取出关键的预测指标对模型进行解释。实验结果表明,通过XGBoost+LR模型进行脓毒症预测的准确率为99%,受试者工作特征曲线下面积(AUROC)为0.984,优于单独使用XGBoost(准确率:95%,AUROC:0.953)和LR(准确率:53%,AUROC:0.556)或者LGBM(准确率:90%,AUROC:0.974),同时通过LIME能有效地提取出前10个最重要的指标,对脓毒症预测模型进行可解释性分析,提高了模型的可信度。