Uncertainty propagation in vegetation distribution models based on ensemble classifiers

Ensemble learning techniques are increasingly applied for species and vegetation distribution modelling, often resulting in more accurate predictions. At the same time, uncertainty assessment of distribution models is gaining attention. In this study, Random Forests, an ensemble learning technique, is selected for vegetation distribution modelling based on environmental variables. The impact of two important sources of uncertainty, that is the uncertainty on spatial interpolation of environmental variables and the uncertainty on species clustering into vegetation types, is quantified based on sequential Gaussian simulation and pseudo-randomization tests, respectively. An empirical assessment of the uncertainty propagation to the distribution modelling results indicated a gradual decrease in performance with increasing input uncertainty. The test set error ranged from 30.83% to 52.63% and from 30.83% to 83.62%, when the uncertainty ranges on spatial interpolation and on vegetation clustering, respectively, were fully covered. Shannon’s entropy, which is proposed as a measure for uncertainty of ensemble predictions, revealed a similar increasing trend in prediction uncertainty. The implications of these results in an empirical distribution modelling framework are further discussed with respect to monitoring setup, spatial interpolation and species clustering.     

Monitoring and adaptive management: resolving social and organisational issues to improve information sharing in natural resource management

Adaptive, or 'learning by doing', approaches are often advocated as a means of providing increased understanding within natural resource management. However, a number of organisational and social issues need to be resolved if these approaches are to be used successfully. A case study in the South Island high country of New Zealand is used to review what is needed to support an ongoing community-based monitoring and adaptive management programme. First, the case study is described, paying attention to the social context of the resource management problem. The results of a workshop that explored this problem are then outlined, along with a proposed information flow suggested by participants. Requirements for future steps to resolve these problems (such as information protocols and a multi-stakeholder information system) are discussed. Finally, some broad lessons are drawn from this exercise that could help others developing similar approaches.     

Finding Hope in the Millennium Ecosystem Assessment

Abstract:  Over the past quarter century, a new scientific activity has emerged: collective assessments by large numbers of scientists from different disciplines combining their expertise to better understand human interrelations with nature and to inform policy. The Millennium Ecosystem Assessment exceeded all such assessments before it in both the breadth of its coverage and the depth of its analysis of socioecological system dynamics. The findings are not encouraging. Nearly all ecosystems are being degraded and will continue to be degraded for decades to come even if policy changes are initiated now. For scientists participating in the assessment, the MA had another disconcerting aspect. It clearly shows that our fragmented, disciplinary knowledges cannot simply be combined to form an understanding of a whole complex system. Counterbalancing the despair of the findings and scientific difficulties of aggregating specialized knowledges, the MA demonstrated the potential of a deliberative democratic approach to grappling with complex problems .     

基于链式多重插补的WOA-ELM煤与瓦斯突出预测模型

为了提高缺失数据下煤与瓦斯突出预测准确率,提出1种基于链式支持向量机多重插补(MICE＿SVM)的鲸鱼优化算法(WOA)-极限学习机(ELM)预测模型,以淮南朱集矿区为例,选取5个煤与瓦斯突出影响指标作为模型特征,采用提出的MICE＿SVM算法插补突出事故数据中缺失值,利用WOA优选ELM输入层权值及隐含层阈值,构建煤与瓦斯突出预测模型,将插补后数据用于WOA-ELM模型的训练与测试,并与其他模型的预测效果对比。研究结果表明：MICE＿SVM插补前、后的有突出数据预测准确率分别为83.02%,90.41%,MICE＿SVM显著提高了有突出预测准确率,对无突出和整体的预测准确率提高不明显;数据插补后WOA优化ELM对无突出、有突出和整体的预测准确率分别为97.94%,96.25%,96.48%,较优化前分别提高了5.79%,5.84%,5.55%,数据插补后WOA-ELM为最佳预测模型。     

基于SVR的含缺陷管道剩余强度研究

针对采用标准预测含缺陷管道剩余强度误差较大这一问题,在Matlab中建立基于SVR的含缺陷管道剩余强度预测模型,并基于60组含缺陷管道爆破试验数据进行训练测试,以验证模型的实际性能.结果表明:SVR模型预测测试集结果的最小相对误差为0.55％,最大相对误差为10.35％,平均相对误差为2.63％,预测结果的R2高达0....     

YOLOv4目标检测算法在煤矿工人口罩佩戴监测工作中的应用研究*

为防止煤矿工人吸入过量粉尘而导致职业性尘肺病,基于Keras框架利用YOLOv4 (you only look once)目标检测算法对井下人员佩戴防尘口罩情况进行高精度且快速的检测与识别,并与MTCNN(Multi-task convolutional neural network)和FaceNet构成的人脸识别算法相结合,进行煤矿工人口罩佩戴监测的研究。结果表明:模型对井下人员口罩佩戴有较高的检测精度,识别已佩戴口罩的矿井下作业人员的平均精度达到92.78%,识别未佩戴防尘口罩检测的平均精度为91.63%,与其他主流算法相比算法具有更好的鲁棒性和检测效果。研究结果为预防煤矿工人职业性尘肺病提供1种有效的技术手段。     

基于卷积神经网络的冲击地压预警方法研究*

为探究深度学习在冲击地压预警方面的应用前景,以新疆某冲击地压矿井为研究背景,将深度学习和专家评判运用到微震数据分析中,基于卷积神经网络构建冲击地压预警模型。充分利用一维卷积神经网络对时序数据有较强特征提取能力的优势,以微震数据及其特征参数作为输入,以专家评判值作为标签,借助Python-Keras框架实现冲击地压预警模型的构建和训练。研究结果表明:模型预警效果并不随着训练迭代次数的增加而逐渐最优,存在最优迭代次数,对于所建模型当迭代次数为30时测试集的冲击危险预测结果与专家评判结果基本吻合,同时说明模型可以较好地学习专家评判经验实现冲击地压预警。研究表明所建模型对研究时段内发生的5次大能量矿震事件均进行预警,其准确度较高,具有现场实际应用价值。     

基于视频分析的空管员违规行为识别方法*

为利用视频数据对空管员违规行为进行智能化分析,降低不安全事件发生率,提出2阶段的违规行为识别模型（AR-ResNeXt）,基于实地调研构建空管员视频数据集,利用最小化动态多实例学习损失函数和中心损失函数,获得违规行为检测的判别特征表示,结合异常回归网络和ResNeXt网络,完成对空管员违规行为的时序区间检测与动作分类。研究结果表明:AR-ResNeXt模型在自制数据集中,其帧级AUC达到82.9%,分类准确率达到87.8%,可准确识别空管员发生违规行为的时序区间并进行分类,研究结果可为保障空中交通安全奠定基础。     

基于深度学习的工程结构损伤识别研究进展

为避免或减轻工程结构在建造和运营期间因结构振动产生不同程度损伤,造成安全隐患危及人们生命财产安全,针对结构振动损伤识别技术展开研究,探讨不同深度学习方法发展情况及其利弊,寻找更具可行性的损伤识别方法,并对其最新研究及应用现状进行全面综述。研究结果表明:应用深度学习开发新的结构损伤识别技术,无需冗余的数据预处理以及手工提取损伤特征,实现以较高精度实现损伤识别任务;一维卷积神经网络（1D-CNN）以其独特的应用优势,在数据样本有限条件下较二维卷积神经网络（2D-CNN）表现更为出色。研究结果可为数据驱动的结构损伤识别问题提供新思路,进一步完善土木结构健康监测研究体系。     

基于SA-ELM的爆破震动速度预测

针对爆破震动速度与其影响因素之间的复杂非线性,结合模拟退火算法(SA)的全局寻优性,提出了一种新的SA-ELM算法.以矿山周边建筑物爆破震动实测数据作为训练样本,选取总药量、最大段药量、测点与爆破点距离、地面震动特性、建筑物震动特性等8个影响因素作为输入变量,建立了爆破震动速度预测的SA-ELM模型.模型训练值和预测值与实测值的均方误差(MSE)分别为0.20和3.26,平均相对误差控制在5％以内,显示出该模型具有良好的训练精度和泛化能力.对比传统ELM模型,SA-ELM模型不但提高了精度和泛化能力,而且降低了隐层节点数变化对训练结果的影响,提高了模型的稳定性.