基于深度编码注意力的XLNet-Transformer汉-马低资源神经机器翻译优化方法

神经机器翻译(NMT)在多个领域应用中已取得显著成效,在大规模语料库上已充分论证其优越性。然而,在语料库资源不足的情形下,仍存在较大的改进空间。由于汉语-马来语(汉-马)平行语料的匮乏,直接导致了汉-马机器翻译的翻译效果不佳。为解决汉-马低资源机器翻译不理想的问题,提出了一种基于深度编码注意力和渐进式解冻的低资源神经机器翻译方法。首先,利用XLNet预训练模型重构编码器,在编码器中使用了XLNet动态聚合模块替代了传统编码层的输出方式,有效弥补了低资源汉-马语料匮乏的瓶颈;其次,在解码器中使用并行交叉注意力模块对传统编码-解码注意力进行了改进,提升了源词和目标词的潜在关系的捕获能力;最后,对提出模型采用渐进式解冻训练策略,最大化释放了模型的性能。实验结果表明,提出方法在小规模的汉-马数据集上得到了显著的性能提升,验证了方法的有效性,对比其他的低资源NMT方法,所提方法结构更为精简,并改进了编码器和解码器,翻译效果提升更加显著,为应对低资源机器翻译提供了有效的策略与启示。

深层煤层气水平井连续油管解卡工艺研究

针对深层煤层水平井连续油管较卡问题,从地质因素、工程因素两方面分析垢卡原因,并系统性总结解深层煤水平井连续油管煤垢卡的解卡过程及技术要点,为今后同类作业提供指导性建议。

RS码的Step-by-step译码及软件实现

RS码以其强大的纠正随机错误和突发错误的能力,被广泛地应用于各种数字通信系统中。其译码算法是实现RS码的核心,Step-by-step算法与传统的译码算法相比,不用计算错误位置多项式,可以大大地节省译码时间。文章在分析RS码的基本原理及Step-by-step译码算法的基础上,介绍了一种用软件来实现Step-by-step译码算法的方案,并且给出了流程图。

Effects of curing under step-by-step load on mechanical and deformation properties of cemented gangue backfill column

A step-by-step load was utilized to mimic the load history of the backfill column in the in-situ curing process.The inner damage of the specimen during curing and uniaxial compressive testing was monitored by electrical resistivity and ultrasonic equipment.Results show that:1)Uniaxial compressive strength(UCS)and elastic modulus(EM)of the samples curing under pressure are higher than those of the control samples without pressure,ranging in ratio from 0.5%to 20.2%and 7.1%to 52.3%,respectively,and are influenced by the initial loading age(ILA)and stress strength ratio(SSR).The SSR during curing should not exceed 80%.2)The earlier the ILA is,the higher the total strain becomes.The higher the SSR applies,the larger the total strain gets.The creep strain increases with the increase of SSR and can be described by Burger’s viscoelastic creep model.When SSR is less than 80%,the earlier the ILA is,the smaller the creep strain becomes after the last step-loading.3)The stability of the early age backfill column under pressure can be monitored based on the change of ultrasonic pulse velocity(UPV)and electrical resistivity.

Application of a Step-by-Step Fingerprinting Identification Method on a Spilled Oil Accident in the Bohai Sea Area

In recent years,oil spill accidents occur frequently in the marine area of China.Finding out the spilled oil source is a key step in the relevant investigation.In this paper,a step-by-step fingerprinting identification method was used in a spilled oil accident in the Bohai Sea in 2002.Advanced chemical fingerprinting and data interpretation techniques were used to characterize the chemical composition and determine the possible sources of two spilled oil samples.The original gas chromatography -flame ionization detec-tion (GC-FID) chromatogram of saturated hydrocarbons was compared.The gas chromatography-mass spectrometry (GC/MS) chromatograms of aromatic hydrocarbons terpane and sterane,n-alkane and poly-aromatic hydrocarbons (PAHs) were analyzed.The correlation analysis on diagnostic ratios was performed with Student’s t-test.It is found that the oil fingerprinting of the spilled oil (designated as sz1) from the polluted sand beach was identical with the suspected oil (designated as ky1) from a nearby crude oil refinery factory.They both showed the fingerprinting character of mixed oil.The oil fingerprinting of the spilled oil (designated as ms1) collected from the port was significantly different from oil ky1 and oil sz1 and was with a lubricating oil fingerprint character.The identification result not only gave support for the spilled oil investigation,but also served as an example for studying spilled oil accidents.

3D Step-by-step inversion strategy for audio magnetotellurics data based on unstructured mesh

A three-dimensional(3D)step-by-step inversion strategy for audio magnetotellurics(AMT)is investigated in this study.The objective function is minimized by iteratively solving the Gauss-Newton normal equation,and the inversion region is discretized with unstructured tetrahedral elements.The inversion proceeds step-by-step from a coarse mesh to a fine mesh.In the inversion iteration process,a mesh is adaptively optimized according to the spatial gradient information about the model resistivity to fine delineate the boundaries of abnormal bodies.In the early stage of inversion execution,a coarse mesh is used for inversion,and the inversion stability is improved by reducing the number of inversion elements.In addition,mesh refinement is performed in the iterative inversion process.The inversion results obtained from the previous mesh are used as the reference and initial models for the next mesh iterative inversion.The step-by-step inversion strategy can ensure that the inversion is performed in the correct direction,improving the inversion stability and results gradually.Synthetic results show that the step-by-step inversion strategy with a Gauss-Newton method for 3D AMT inversion is stable and reliable,which lays a foundation for further practical 3D AMT data inversion.

Equidistant combination wavelength screening and step-by-step phase-out method for the near-infrared spectroscopic analysis of serum urea nitrogen

We applied near-infrared(NIR)spectroscopy with chemometrics for the rapid and reagent-fee analysis of serum urea nitrogen(SUN).The modeling is based on the average effect of multiple sample partitions to achieve parameter selection with stability.A multiparameter optimization platform with Norris derivative filter-partial least squares(Norris-PLS)was developed to select the most suitable mode(d=2,s=33,g=15).Using equidistant combination PLS(EC-PLS)with four parameters(initial wavelength I,number of wavelengths N,number of wavelength gaps G and latent variables LV),we performed wavelength screening after eliminating high-absorption wavebands.The optimal EC-PLS parameters were I=1228 nm,N=26,G=16 and LV=12.The root-mean square error(SEP),correlation coefficient(R_(p))for prediction and ratio of performance-to-deviation(RPD)for validation were 1.03 mmol L^(-1),0.992 and 7.6,respectively.We proposed the wavelength step-by-step phase-out PLS(WSP-PLS)to remove redun-dant wavelengths in the top 100 EC-PLS models with improved prediction performance.The combination of 19 wavelengths was identifed as the optimal model for SUN.The SEP,Rp and RPD in validation were 1.01 mmol L^(-1),0.992 and 7.7,respectively.The prediction effect and wavelength complexity were better than those of EC-PIS.Our results showed that NIR spectroscopy combined with the EC-PLS and WSP-PLS methods enabled the high-precision analysis ofSUN.WSP-PLS is a secondary optimization method that can further optimize any wavelength moc odel obtained through other continuous or discrete strategies to establish a simple and better model.

Step-by-step external fixation of unstable pelvis with separate anterior and posterior modules

Objective: To evaluate the treatment outcomes of patients with pelvic ring injury by applying step-by-step external pelvic fixation and circular external fixation device.Methods: A total of 28 patients suffering from disintegrated pelvic ring injury are involved in the study. Fourteen patients(the treatment group) underwent step-by-step external pelvic fixation by applying anterior(anti-shock) and posterior modules. For the rest 14 patients(the experimental group), the osteosynthesis were conducted by means of a circular external fixation device. The long-term outcomes were evaluated in a year after the injury.Results: The residual deformity of 5(4–7) mm was observed in 10 patients(71.4%) from the experimental group. In the treatment group, the residual deformity was evident only in 4(28.6%) cases being 2.5(2–3) mm(P = 0.000 319) on the average. The functional result(according to the Majeed scale) was statistically better in the treatment group(P = 0.000 319). Nine(64.3%) and five(35.7%) patients in treatment group showed excellent and positive results, respectively. The excellent result was demonstrated by 3patients(21.4%) of the experimental group, the positive outcomes were observed in 6 cases(42.9%) and the unsatisfactory one was displayed by 1 patient(7.1%) of the same group.Conclusions: The modular approach applied is the advantage of the transosseous osteosynthesis allowing for a separate anterior(anti-shock) fixation and final posterior reposition of the pelvic ring preceded by the stabilization of vital functions. The above mentioned method gives an opportunity to increase the amount of techniques applied for the pelvic external fixation in polytrauma cases.

COORDINATION OF THE MOTION PARAMETERS WITHIN STEP-BY-STEP INTEGRATION FOR DYNAMIC EQUATION

A method is presented that coordinates the calculation of the displacement, velocity and acceleration of structures within the time-steps of different types of step-by-step integration. The dynamic equation is solved using an energy equation and the calculating data of the original method. The method presented is better than the original method in terms of calculating postulations and is in better conformity with the system's movement. Take the Wilson-θ method as an example. By using the coordination process, the calculation precision has been greatly im proved (reducing the errors by approximately 90% ), and the greater part of overshooting of the calculation result has been eliminated. The study suggests that the mal-coordination of the motion parameters within the time-step is the major factor that contributes to the result errors of step-by-step integration for the dynamic equation.

A STEP-BY-STEP INTEGRAL METHOD FOR ACTIVELY CONTROLLED STRUCTURES

A new active control method was proposed, in which the analytical control law was deduced by using a step by step integral method to differential equation of motion under the condition of static error being zero. This control law is terse in mathematical expression and convenient for practical use. The simulation results demonstrate that the proposed method can provide much more remarkable peak response reduction of seismically excited structures than the classical LQR method.

The step-by-step CFD design method of pressure-compensating emitter

In order to improve the design and research and development (R & D) efficiency of the pressure- compensating drip irrigation emitter,a step-by-step computational fluid dynamics (CFD) design method was proposed based on CFD theory combined with the finite element method. By analyzing its hydraulic performance through the step-by-step CFD method,the prediction pressure-flow curve(p-Q curve) of the pressure-compensating emitter was obtained. Then the test samples were fabricated using rapid prototype and manufacturing(RP & M) technology. The emitters' hydraulic performance experiment was carried out and the experimental p-Q curve was obtained. The step-by-step CFD design method was verified by comparing the experimental p-Q curve with the prediction values,which showed that the prediction values met the experimental results well within the normal range of the emitter's working pressure. On this basis,the effect of the emitter structure on its pressure-compensating performance was studied,which showed that the height of the pressure-compensating region had significant effects on the emitter's pressure-compensating performance. Series products of the pressure-compensating emitter could be designed by changing the region's height.

Simple Program for Step-by-Step Time Integration in Chemical Kinetics, Applied to Simple Model for Hydrogen Combustion

A simple algorithm is proposed for step-by-step time integration of stiff ODEs in Chemical Kinetics. No predictor-corrector technique is used within each step of the algorithm. It is assumed that species concentrations less than 10-6 mol·L-1 do not activate any chemical reaction. So, within each step, the time steplength Δt of the algorithm is determined from the fastest reaction rate maxR by the formula Δt = 10-6mol·L-1/max R. All the reversible elementary reactions occur simultaneously;however, by a simple book-keeping technique, the updating of species concentrations, within each step of the algorithm, is performed within each elementary reaction separately. The above proposed simple algorithm for Chemical Kinetics is applied to a simple model for hydrogen combustion with only five reversible elementary reactions (Initiation, Propagation, First and Second Branching, Termination by wall destruction) with six species (H2, O2, H, O, HO, H2O). These five reversible reactions are recommended in the literature as the most significant elementary reactions of hydrogen combustion [1] [2]. Based on the proposed here simple algorithm for Chemical Kinetics, applied to the global mechanism of proposed five reversible elementary reactions for hydrogen combustion, a simple and short computer program has been developed with only about 120 Fortran instructions. By this proposed program, the following are obtained: 1) The total species concentration of hydrogen combustion, starting from the sum of initial reactants concentrations [H2] + [O2], gradually diminishes, due to termination reaction by wall destruction, and tends to the final concentration of the product [H2O], that is to the 2/3 of its initial value, in accordance to the established overall stoichiometric reaction of hydrogen combustion 2H2 + O2 → 2H2O. 2) Time-histories for concentrations of main species H2, O2, H, H2O of hydrogen combustion, in explosion and equilibrium regions, obtained by the proposed program, are compared to corresponding ones obtained by accurate computational studies of [3]. 3) In the first step of the algorithm, the only nonzero species concentrations are those of reactants [H2], [O2]. So, the maximum reaction rate is that of the forward initiation reaction max R = Rif = kif[H2] [O2], where the rate constant kif is very slow. Thus, the first time steplength Δt1 = 10-6mol·L-1/max R results long in sec. After the first step, the sequences of all the following Δt’s are very short, in μsec. So, the first time steplength Δt1 can be considered as ignition delay time. 4) It is assumed that explosion corresponds to ignition delay time Δt1 t1 = 10 sec., can be considered as explosion limit curve. This curve is compared to the corresponding one obtained by the accurate computational studies of [2].

国内水平井修井工艺技术现状

水平井是油气田高效开发的一项重要技术,随着钻井技术水平的提高,提高油气井产量、采收率、降低油气田开发成本的需要,水平井的数量和类型越来越多,水平井的应用越来越广泛。另一方面,水平井由于其井身轨迹的特殊性,水平井的油套管更易出现磨损、腐蚀等情况,也更易发生故障,水平井修井面临更加严峻复杂的挑战。国内水平井修井起步较晚,工艺技术相对滞后,需要针对性地开展研究,不断探索,不断实践,以提高水平井修井工艺水平,满足油气田高效开发中水平井修井数量和难度不断增长的需要。

高效震击解卡工艺研究

目前常用的震击钻具组合面对线卡、面卡等复杂卡管柱现象时,由于震击力不足,常常导致解卡失败。为此,对震击钻具组合及其施工参数进行优化设计。优化后的钻具组合为:钻杆+加速器+钻铤+震击器+安全接头+打捞工具,其中的加速器有很好的储能效果,配合加速器使用比不用加速器能使震击效果提高近1倍,同时还有一定的减震效果。在卫77-9井的现场应用中,经过连续震击2.5 h便顺利打捞出机械内割刀,随后又下井3次,连续震击解卡9 h,并配合套铣,共震击360余次,打捞进尺24 m,顺利打捞出已切割断的部分套铣筒、油管反循环阀、喷砂器及封隔器等全部井下落物。

一种新型活动式井下节流器

为了解决老式节流器在使用中存在的打捞困难及密封皮碗与油管之间的摩擦问题,同时大幅度降低节流器打捞时的上提力,研制了新型活动式井下节流器。该节流器主要由滤砂管、外套管、密封皮碗、卡瓦、剪切销钉和投送接头等组成,保留了老式节流器定位准确、密封良好、耐压差高的优点,同时也克服了老式节流器打捞难的缺点。苏里格苏75区块10口井的应用表明,与同类产品相比,该节流器打捞性能及节流降压效果均更好,值得大面积推广应用。

解卡打捞技术在小井眼井中的应用

通过对大井眼井 (套管规范为Φ139 7mm)和小井眼井 (套管规范为Φ10 1 6mm和Φ114 3mm)落物在井内遇卡、解卡和打捞特点的分析 ,给出了在小井眼井中对遇卡事故的处理方法 ,即 :对于砂卡、盐水和矿物的结晶卡、胶件卡、封隔器卡、小件落物卡等主要采用震击方法 ;对于水泥凝固、化学物质凝固、电缆、钢丝绳类落物堆积卡主要采取钻铣磨解卡方法。提出了一套适合小井眼井解卡打捞的配套工艺和自行研制的打捞工具 。

基于知识元的动态知识管理模型研究

[目的/意义]以知识元相关理论为基础,构建基于知识元的动态知识管理模型,剖析动态知识管理过程。[方法/过程]界定知识元概念,回顾和分析现有知识管理模型,基于知识元组合—知识分解—知识元再组合范式,探究动态知识管理模型,并从知识元组合、分解和再组合三个层面解释知识管理过程。[结果/结论]研究表明:动态知识管理模型分为组合阶段、分解阶段、再组合阶段。组合阶段是基于任务导向的知识元融合;分解阶段是知识解冻的过程,实现知识元自由流动;再组合阶段是基于新任务导向的知识元重新组合和配置。

甘蓝型油菜花粉超低温保存及其花粉活力的研究

研究超低温方法保存油菜花粉过程中的预冻和解冻处理方式对油菜花粉的形态、大小及其活力的影响。结果表明:采用0℃(12h)→-4℃(12h)→-20℃(12h)→-80℃(12h)变温预冻处理和用-80℃(1h)→-20℃(1h)→4℃(1h)逐步解冻后对油菜花粉形态大小和活力影响最小。而经过25℃室温解冻法和42℃快速解冻法处理后油菜花粉出现破裂,破裂率分别达到7.6%和9.1%。同时液氮保存油菜花粉的时间长短对花粉的大小及活力影响不大。

体外培养1d对玻璃化解冻卵裂期胚胎移植效果的影响

目的 探讨体外培养1 d对玻璃化解冻卵裂期胚胎移植效果的影响。方法 选择本中心2012年1~7月进行玻璃化冷冻并复苏的277周期卵裂期胚胎作为研究对象,根据解冻后是否多培养1 d分为对照组和实验组。对照组在解冻当天移植,共123周期。实验组解冻后多培养1 d,共154周期。实验组又分为培养后含桑葚胚组（A组）及无桑葚胚组（B组）。观察延长体外培养时间对妊娠结局的影响以及培养后胚胎发育情况与妊娠结局的关系。结果 实验组的妊娠率和着床率显著高于对照组,差异有统计学意义（P〈0.05）。A组的妊娠率与着床率显著高于B组,差异有统计学意义（P〈0.05）。结论 体外培养1 d不影响玻璃化解冻后卵裂期胚胎的移植效果,延长培养时间有利于选择发育良好的胚胎并显著提高临床妊娠率和着床率。

煤的化学族组成研究(Ⅱ)煤的可溶化体系的优化

煤的可溶化体系是煤化学族组成研究的前提和关键，通过从高抽提率、溶剂回收性、普遍适应性、快捷省时、方便准确等方面综合考察溶剂选择、溶剂组合方式、溶剂抽提方式等因素对煤可溶化的影响，提出了一套较为优化的可溶化体系，以满足煤化学族组成研究的需要．