面向密闭空间内外温度的时序预测模型

目的 研究密闭空间条件下已知外部温度时间序列对内部实时温度的预测推理问题。方法 选取密闭空间内外温度时序预测典型场景,抽象为多变量时间序列预测问题,分析变量间的关联性和依赖性。借鉴特征融合、注意力机制、多任务模型等思路,结合物理机制与数据特征,基于长短期记忆网络基本网络单元,构建密闭空间内外温度时序预测模型,并在万宁、敦煌、漠河对某型密闭空间进行数据采样,基于三地数据集进行不同模型试验。结果 多变量模型比单变量模型具有更好性能,注意力机制对该场景没有显著性能提升,结合物理机制的模型结构设计充分考虑了变量之间的关联性和依赖性,能显著提升预测精度,双输入双输出的多变量时序预测模型具有相对最高的精度和最稳定的鲁棒性,是面向密闭空间内外温度时序预测的相对最优模型。结论 研究结论可指导密闭空间其他环境特征建模,研究思路可为其他多变量时序建模问题中变量之间的关联性、依赖性分析提供参考。     

Trends in Rainfall Exceedances in the Observed Record in Selected Areas of the United States1

Bonnin, Geoffrey M., Kazungu Maitaria, and Michael Yekta, 2011. Trends in Rainfall Exceedances in the Observed Record in Selected Areas of the United States. Journal of the American Water Resources Association (JAWRA) 47(6): 1173–1182. DOI: 10.1111/j.1752‐1688.2011.00603.x Abstract: Semantic differences have led to a gap in the understanding of the impacts of climate change on precipitation frequency estimates. There is popular perception that heavy rainfalls have become more frequent, and that this trend will increase with global warming. Most of the literature examines this question from the point of view of climatology using definitions of “heavy,”“very heavy,” or “extreme” rainfall, which are different from those commonly used by civil engineers. This article identifies the differences in meaning used by the climate and civil engineering communities and examines trends in the observed record in the frequency of exceedances (not trends in magnitudes). Using concepts recognized as the basis for design of the Nation’s civil infrastructure, we look at trends in the number of exceedances of thresholds for a variety of precipitation frequencies and event durations used by civil engineers. We found that the estimated trends in exceedances at one‐day and multiday durations were statistically significant and increasing for the Ohio River Basin and surrounding states but the reverse was true for the Semiarid Southwest (i.e., not significant and decreasing trends). In addition, we found the magnitude of the trends was small for all but the more frequent events and also small with respect to the uncertainty associated with the precipitation frequency estimates themselves.     

The business cycle and the incidence of workplace injuries: evidence from the U.S.A

Introduction

The current study explored the association between the business cycle and the incidence of workplace injuries to identify cyclically sensitive industries and the relative contribution of physical capital and labor utilization within industries.

Method

Bureau of Labor Statistics nonfatal injury rates from 1976 through 2007 were examined across five industry sectors with respect to several macroeconomic indicators. Within industries, injury associations with utilization of labor and physical capital over time were tested using time series regression methods.

Results

Pro-cyclical associations between business cycle indicators and injury incidence were observed in mining, construction, and manufacturing but not in agriculture or trade. Physical capital utilization was the highest potential contributor to injuries in mining while labor utilization was the highest potential contributor in construction. In manufacturing each effect had a similar association with injuries.

Conclusion

The incidence of workplace injury is associated with the business cycle. However, the degree of association and the mechanisms through with the business cycle affects the incidence of workplace injuries was not the same across industries.

Impact on Industry

The results suggest that firms in the construction, manufacturing, and mining industries should take additional precautionary safety measures during cyclical upturns. Potential differences among industries in the mechanisms through which the business cycle affects injury incidence suggest different protective strategies for those industries. For example, in construction, additional efforts might be undertaken to ensure workers are adequately trained and not excessively fatigued, while safety procedures continue to be followed even during boom times.     

基于多因素的LSTM瓦斯浓度预测模型*

为解决煤矿瓦斯浓度预测问题,提出基于多因素的LSTM瓦斯浓度预测模型。模型首先对煤矿多源监测数据进行数据融合、缺失值处理;其次通过特征衍生、有监督化、无量纲化,融合各环境因素特征和时序数据的时间性特征,且衍生出更多交叉项特征和高次项特征;然后利用经验法和逐步试错法确定隐藏层维度;最后进行模型训练和测试。研究结果表明:基于多因素的LSTM瓦斯浓度预测模型的RMSE仅为0.021,MAE为0.01,比单因素LSTM模型、RNN模型预测效果好。因此,基于多因素的LSTM瓦斯浓度预测模型可更精准地进行瓦斯浓度多步预测,促进煤矿安全生产。     

An evaluation of risk assessment framework for industrial accidents in India

Due to rapid industrialization, with high population density and constraints of land, it is expected that level of risks arising from the hazardous industries will increase in India in the coming decades. However, 30 years after the Bhopal accident (1984), except a few discrete regulations, there is as yet no integrated system for assessing and managing risks arising out of these hazardous industries in India. The gravity of aspects related to the management of industrial risk still remains crucially important. In particular, there is no standard guideline on risk analysis methodology, acceptability or tolerability criteria, nor is there an accident database or a risk reduction strategy for the areas where risk levels are already high. On top of this, there are technical and legislative gaps in the institutional framework to implement any of the above mentioned issues. With the backdrop of the Bhopal gas tragedy, the objective of this paper is therefore to evaluate the effectiveness of a comprehensive risk assessment framework for the emerging economy of India, in order to control and/or to reduce the risk level that exists. In this context, regulations and policies pertaining to industrial risk assessment were reviewed.     

IAEA programme on the quality of marine radioactivity data

Society's growing interest in environmental issues requires the production of reliable information for policy-makers, stakeholders and society in general. This information must be based on accurate data produced by qualified laboratories and data need to be comparable between numerous laboratories for joint assessments, e.g. in International Conventions. The Marine Environment Laboratories of the International Atomic Energy Agency has been providing Quality Assurance services to laboratories involved in marine radioactivity studies for the past 30 years. This has included training, organization of laboratory intercomparison exercises, production of Reference Materials (RMs) and more recently, production of Certified Reference Materials (CRMs) that comply with relevant ISO standards. Here, the overall process of Certification of Reference Materials is reviewed, past work summarized and future needs of marine radioactivity laboratories briefly discussed.     

国六重型柴油车挥发性有机物排放特性

选取5辆典型国六重型柴油车,进行基于C-WTVC的挥发性有机物（VOCs）排放测试,分析了包括7种苯系物以及14种醛酮类物质的比排放量.结果表明,甲苯、苯、苯乙烯是国六重型柴油车苯系物排放的主体物质,占总量的60%~86%;而甲醛、乙醛、苯甲醛是醛酮污染物的主要物质,占总量的72%~87%.在同时包含市区、市郊和高速综合工况的试验车会产生较高的苯系物和醛酮类物质的排放,苯、苯乙烯在综合工况下的比排放量分别为处于市郊工况车型的1.25倍、1.45倍,市区占比为40%的货车具有最高的甲醛比排放量20.84mg/（kW×h）,有超甲醇车甲醛排放限值的风险;在市郊工况下车型的甲苯、乙醛排放平均分别为其余车型的1.65倍、2.1倍.测试车辆的臭氧潜势均值达到（249.53±10） mgO₃/（kW×h）.     

基于遗传算法的门限自回归模型在海温预测中的应用

提出了建立门限自回归模型（ＴＡＲ）的一套简便通用的方法。用作者提出的改进遗传算法，可同时优化门限值和自回归系数，成功地解决了ＴＡＲ建模过程所涉及的大量复杂寻优工作这一难题，为ＴＡＲ模型的广泛应用提供了强有力的工具。实例计算的结果说明了这套方法的可行性和有效性，同时也说明了，通过门限值的控制作用，ＴＡＲ模型可以有效地利用如海洋资源所隐含的时序分段相依性这一重要信息，限制了模型误差，从而保证了ＴＡＲ模型预测性能的稳健性，提高了预测精度。该方法具有通用性，在各种非线性时序预测中具有重要的理论意义和应用价值     

串联氧化塘中氮、磷去除的容积分配效应的研究

在人工模拟条件下,串联厌氧 好氧塘在总容积相同,不同的容积分配的情况下,污水中的氮、磷去除效率不同,存在最佳容积分配比：V_厌/V_总约为0.27。在最佳容积分配比处,COD、氮、磷的去除率可分别提高11.4％、26.0％、51.6％。随着负荷、污水C/N比、温度、光照时间等因素的变化,氮磷去除呈一定的变化规律。说明了串联厌氧 好氧塘氮、磷去除的容积分配效应是客观存在的,合理的容积分配对氮、磷去除尤其重要