基于证据推理和置信规则库的涡扇发动机故障预测

针对某型涡扇发动机故障预测的问题,提出一种基于证据推理和置信规则库的涡扇发动机故障预测模型。为描述涡扇发动机的健康状态,利用证据推理算法融合发动机系统状态信息;结合先验知识建立混合驱动的置信规则库仿真预测模型;采用投影协方差自适应进化策略用于优化模型参数;通过实验验证了模型的有效性。研究结果表明：该方法不仅准确预测涡扇发动机故障风险概率,而且为故障诊断和维修保障提供了有力的支撑。     

GPS数据驱动的车辆跟驰行为建模与分析

为研究道路交通流特性,基于车载高精度GPS跟驰试验数据进行车辆跟驰建模研究,结合深度学习理论和数据驱动方法,构建了基于粒子群优化(particle swarm optimization, PSO)的长短期记忆(long short term memory, LSTM)车辆跟驰模型。首先,清洗和平滑车辆轨迹数据,并对驾驶特征行为参数及相关关系进行研究,如加速度、车头时距以及速度与跟驰距离特性关系等。在此基础上,制定跟驰状态筛选规则;其次,构建考虑时间序列的PSO-LSTM模型,识别跟驰数据样本集,将当前时刻的前车速度、车头间距和上一时刻的车头时距作为模型输入,预测当前时刻的跟驰车速度;接着,选用25辆车跟驰试验的高精度GPS数据验证PSO-LSTM模型性能;最后,为验证该模型的优越性,选用传统机器学习SVR(support vector regression)模型以及深度学习LSTM模型作为对比。结果表明,基于粒子群优化的长短期记忆模型预测精度高达0.993,整体预测效果高于SVR模型和LSTM模型,其中预测误差指标MAPE(mean absolute percentage error)较SVR和LSTM分别降低了60.02%、1.52%。PSO算法进行超参数优化后的PSO-LSTM模型,能更好地模拟车辆的跟驰行为。     

短期风电功率CEEMDAN-SMA-LSSVM预测模型研究

为了提高风力发电功率预测的准确性,建立了基于CEEMDAN分解的SMA算法优化LSSVM的短期风电功率组合预测模型。首先,采用完全集合经验模态分解(CEEMDAN)对原始风电功率数据进行分解与重构。随后,为了进一步优化最小二乘向量支持机模型(LSSVM)的参数,引入了黏菌算法(SMA)优化,通过调整惩罚参数和核参数来提高模型性能,最后,构建多种对比模型对比分析表明CEEMDAN-SMA-LSSVM模型预测精度最高,预测结果更接近真实值。研究可用于风电场短期风电功率预测使用。     

Prediction of decadal variability of sea surface temperature by a coupled global climate model FGOALS_gl developed in LASG/IAP

A decadal climate prediction was performed by a coupled global climate model FGOALS_gl developed by the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics(LASG) within the Institute of Atmospheric Physics(IAP),Chinese Academy of Sciences.First,an Incremental Analysis Updates(IAU) scheme was applied to assimilate surface and subsurface ocean temperature and salinity fields derived from oceanic objective analysis data,for the initialization of the ocean component of the model.Starting from the initialized states,hindcast integrations were performed with the specified historical solar cycle variations,concentrations of greenhouse gasses and sulfate aerosol,following the standard 20C3M scenario used in phase three of the Coupled Model Intercomparison Project(CMIP3).Based on the hindcast integrations,we performed forecast integrations under the radiative forcing of the A1B scenario in the CMIP3.Compared with the 20C3M run,the hindcast integrations have a much higher ability to simulate the decadal variability of SST(Sea Surface Temperature) in the tropical central-eastern Pacific and mid-latitude northeastern Pacific.This suggests that the ocean initialization is able to enhance the model skill in the regions with large decadal variability.The forecast integrations suggest that the SST in the tropical central-eastern Pacific has reached its trough phase,and will gradually increase in the following 10-15 years.Meanwhile,the global mean surface temperature predicted by the forecast integrations increases slower than that projected by the A1B scenario run over 2000-2010,but faster than the latter after that.     

突发性地质灾害研究现状

近年来,突发性地质灾害的频繁发生,给人们的生命财产造成的损失越来越大,对突发性地质灾害的研究已经成为国际社会关注的热点领域.为了及时总结该领域已有的研究成果,为防灾减灾提供科学依据,本文从对突发性地质灾害的成因、危险性评价、预测预警和防治等方面进行了现状综述,分析讨论了突发性地质灾害及防治存在的问题与进一步研究的方向.     

IPNet:帧间预测校验的多域卷积神经网络目标跟踪算法

针对现有卷积神经网络分类预测MDNet算法在线更新机制容易将错误样本引入网络模型,导致跟踪算法失效的问题,提出一种基于帧间预测校验的MDNet目标跟踪改进算法IPNet。该算法运用视频压缩领域的帧间预测方法和聚类算法,在前一帧目标跟踪位置的基础上,计算前后连续两帧中目标的相似度,估计出目标下一帧可能出现的候选区域,实现目标位置预测,达到校验跟踪结果的目的。IPNet算法能有效减少更新样本导致的跟踪失效问题,改善了在目标旋转、快速运动以及背景混杂等情况下的跟踪效果,提升了算法的跟踪性能。     

脉冲噪声滤波的无监督鲁棒递归最小二乘方法

在未知期望信号的条件下,提出一种能够抑制脉冲噪声的鲁棒递归最小二乘自适应滤波方法.与传统最小二乘法的代价函数不同,通过引入饱和非线性约束,降低可能出现的脉冲噪声对滤波器权值更新的影响.此外,提出一种多步预测器来重构滤波器的输入信号,通过比较判断滤波器输入信号可能受到脉冲噪声干扰时,采用预测值来替代原始观测信号.实验结果表明,提出的无监督鲁棒递归最小二乘自适应滤波方法在未受到脉冲噪声干扰时与传统的递归最小二乘法具有相近的收敛性能;在脉冲噪声条件下,传统递归最小二乘法和其他的无监督自适应滤波方法性能都变得很差,但本文提出的方法几乎未受到脉冲噪声的影响.     

特征分析法在最小预测区优选中的应用

特征分析法在矿产预测中有着极为广泛的应用;在研究当曲式铁矿矿床地质特征、成矿规律的基础上,将特征分析方法应用到最小预测区优选中。根据成矿模型优选出含矿建造、断裂构造、航磁异常等8个预测变量,在12个最小预测区中圈定出了7个一级最小预测区和4个二级最小预测区,指出了进一步找矿方向。     

Bioinformatic prediction of selenoprotein genes in the dolphin genome

Selenium (Se), an essential trace element in vivo, is present mainly as selenocystein (Sec) in various selenoproteins. The Sec residue is translated from an in-frame TGA codon, which traditionally functions as a stop codon. Prediction of selenoprotein genes is difficult due to the lack of an effective method for distinguishing the dual function of the TGA codon in the open reading frame of a selenoprotein gene. In this article a eukaryotic bioinformatic prediction system that we have developed was used to predict selenoprotein genes from the genome of the common bottlenose dolphin, Tursiops truncatus. Sixteen selenoprotein genes were predicted, including selenoprotein P and glutathione peroxidase. In particular, a type II iodothyronine deiodinase was found to have two Sec residues, while the type I iodothyronine deiodinase gene has two alternative splice forms. These results provide important information for the investigation of the relationship between a variety of selenoproteins and the evolution of the marine-living dolphin.     

复杂产品项目工作分解绩效预测的系统动力学模型

基于复杂产品项目工作分解的特征和系统动力学在解决系统本质行为和机理方面的优势, 建立系统动力学模型, 用以分析对产品复杂度的把握、核心技术的创新能力、人员动因、 合作伙伴和外部环境对工作分解绩效的发展影响, 并对特大型船舶制造项目进行动态仿真, 分析和预测工作分解的绩效和可能存在的问题, 结果表明: 对产品复杂度的把握和核心技术的创新能力是工作分解绩主要压力, 因此导致的企业投入分配比例问题将是工作分解高效进行的主要制约因素, 主要影响因素相互协调配合才能有效提升工作分解绩效.