基于注意力机制的人群计数方法

为准确预测固定场景中的人群计数,在人群分析领域,采用一种融合注意力机制的卷积神经网络(CNN)进行人群计数,该模块结合空间域注意力和通道域注意力,空间域注意力可以编码整个图像的像素级上下文信息,以更准确地表达像素级别的密度图,而通道域注意力可以在不同的通道中提取更多的区分特征使网络显著表达人群的局部区域,并在多个公开数...     

基于泛平均运算的矿井瓦斯浓度组合预测模型

为有效分析煤矿瓦斯监测数据以实现准确、可靠的瓦斯浓度预测,基于不等权泛平均运算模型,研究瓦斯浓度时间序列组合预测的方法,提出一种新的矿井瓦斯浓度组合预测模型,并证明最优组合预测模型是其特例。采用自回归(AR)模型和径向基函数(RBF)神经网络预测模型作为组合预测模型的单项预测模型;以遗传算法和最小二乘法确定新组合预测模型的参数,实现瓦斯浓度预测单项模型的最优组合。试验分析表明:新模型在平方和误差、平均绝对误差、均方误差、平均绝对百分比误差、均方百分比误差等评价指标上,均取得比自回归模型、径向基函数神经网络模型和最优组合预测模型更低的误差。     

The effect of contextual information on professional judgment: Reliability and biasability of expert workplace safety inspectors

Introduction: A critical aspect of occupational safety is workplace inspections by experts, in which hazards are identified. Scientific research demonstrates that expectation generated by context (i.e., prior knowledge and experience) can bias the judgments of professionals and that individuals are largely unaware when their judgments are affected by bias. Method: The current research tested the reliability and biasability of expert safety inspectors’ judgments. We used a two-study design (Study 1, N = 83; Study 2, N = 70) to explore the potential of contextual, task-irrelevant, information to bias professionals’ judgments. We examined three main issues: (1) the effect that biasing background information (safe and unsafe company history) had on professional regulatory safety inspectors’ judgments of a worksite; (2) the reliability of those judgments amongst safety inspectors and (3) inspectors’ awareness of bias in their judgments and confidence in their performance. Results: Our findings establish that: (i) inspectors’ judgments were biased by historical contextual information, (ii) they were not only biased, but the impact was implicit: they reported being unaware that it affected their judgments, and (iii) independent of our manipulations, inspectors were inconsistent with one another and the variations were not a product of experience. Conclusion: Our results are a replication of findings from a host of other professional domains, where honest, hardworking professionals underappreciate the biasing effect of context on their decision making. The current paper situates these findings within the relevant research on safety inspection, cognitive bias and decision making, as well as provides suggestions for bias mitigation in workplace safety inspection. Practical Application: Our results have implications for occupational health and safety given that inspection is an integral aspect of an effective safety system. In addition to our findings, this study contributes to the literature by providing recommendations regarding how to mitigate the effect of bias in inspection.     

CREAM方法中控制模式分类的模糊方法

认知可靠性与差错分析方法(CREAM)通过对任务环境进行分析从而直接确定人为差错发生概率。该方法提出控制模式的概念,认定不同的任务环境下可能对应不同的控制模式,而人为差错概率直接由控制模式确定。因此,确定控制模式是其中的关键环节。针对该问题,笔者提出了一种模糊方法。通过对性能影响因子进行打分对任务环境进行量化,利用专家打分的方式确定性能影响因子分值对于控制模式的隶属度;通过拟合分析确定性能影响因子的分值对控制模式的隶属函数,得到控制模式与分值之间的对应关系;将该方法应用到两个示例中,其结果与原有方法的结果基本一致。     

基于粒子群-支持向量机(PSO-SVM)的苯储罐泄漏事故风险预测

将支持向量机(SVM)模型运用于事故前苯储罐泄漏事故风险预测,为使模型性能最优, 用粒子群算法PSO优化SVM模型参数,建立了PSO-SVM风险预测模型。为验证模型风险预测性能,分别采用遗传算法(GA)和网格搜索法(GS)优化SVM参数,并比较测试集与PSO-SVM、GA-SVM、GS-SVM三种模型预测结果的均方误差及相关系数。然后进一步探讨模型中权重调整方式、种群规模对PSO-SVM模型预测性能的影响。研究发现,权重线性递减所建PSO-SVM预测值与测试集相关系数更高、均方误差更小、预测效果更好,种群规模没有影响PSOSVM模型预测值但会影响计算时间,这为危化品泄漏事故的风险预测提供了一种新的方法。     

基于改进的最小二乘法瓦斯涌出量预测

通过对基于绝对误差最小的传统最小二乘法与在此基础上考虑相对误差的改进最小二乘法的比较,得知改进的最小二乘法预测更为准确。现场应用于鹤壁四矿的相对瓦斯涌出量随深度变化的预测,建立了瓦斯涌出量预测模型,通过预测值与实际值比较,证明改进的最小二乘法预测鹤壁四矿瓦斯涌出量是可行的、有效的,对指导煤矿安全生产具有重要意义。     

集合卡曼滤波算法对FARSITE模型林火蔓延预测的修正效果研究

针对火源位置输入偏差导致的FARSITE林火行为模型火线预测不准确的问题,提出了一种基于集合卡曼滤波算法的动态修正方法。利用FARSITE对复杂工况下的林火蔓延过程进行数值模拟,以火线位置为待修正参量,以均方根误差(RMSE)为评价指标,对算法的可行性进行了验证,并研究了算法的集合元素个数,观测数据标准差及同化频率对FARSITE预测偏差的修正效果的影响。结果表明:算法能显著提高FARSITE火线预测精度;逐时同化时:集合元素个数为5 时,算法的修正效果并不理想,随着集合元素个数增大,样本误差减小,修正效果得到改善,但增大到30以上时, 修正能力的提升就不再明显;观测数据标准差大小与RMSE值呈正相关;给定条件下当同化频率由1 h/次降低至2 h/次,整个模拟时长内的误差仍能得到较好控制,RMSE曲线并不会过快增长。     

铀尾矿库WSN安全监控镜像约束域定位算法

铀尾矿库的WSN安全监控监测技术是一门新型领域技术,是实现铀尾矿库安全综合治理的重要前提,定位技术因其能够快速定位事件发生区域的地理位置而成为WSN的关键技术之一。RSSI测距定位技术因其易于实现、成本低等特点而被广泛应用,然而其在铀尾矿库中会受到多径、非视距等因素的影响,远距离得到的RSSI值测距误差较大。为此,提出一种镜像最小通信圆约束域定位算法,未知节点利用多个近距离RSSI测距值生成最小通信圆,以锚节点坐标镜像最小通信圆生成镜像圆,多条件约束形成定位区域进行区域定位。仿真结果表明:多最小通信圆定位效果优于单最小通信圆,且比传统RSSI和RSSI质心算法具有更高的定位精度和较小的误差均方差。     

Injury Severity Score alone predicts mortality when compared to EMS scene time and transport time for motor vehicle trauma patients who arrive alive to hospital

ABSTRACT

Objective: This study aims to identify the association, if any, between prehospital scene time, prehospital transport time, and Injury Severity Score (ISS) with in-hospital mortality.

Methods: A retrospective analysis was performed on patients at least 18 years of age who arrived to the hospital alive via emergency medical services (EMS) after a motor vehicle collision (MVC) between 1992 and 2016. These patients were divided into groups based on minutes spent at the scene and in transport. The ISS of the in-hospital mortalities, as well as the entire patient sample for each time frame, was collected. Patients without documented scene time, transport time, or ISS were excluded.

Results: Four thousand one hundred ninety-four patients were captured when analyzing scene time, though only 3,980 met inclusion criteria. In addition, 4,177 patients were captured when analyzing transport time, though only 3,979 met inclusion criteria. Scene time and transport time were not statistically significant predictors of in-hospital mortality (P = .31 and P = .458, respectively). ISS was found to be a statistically significant predictor of in-hospital mortality (P < .001).

Conclusions: ISS predicts mortality independent of scene time or transport time for patients who arrive to the hospital alive following an MVC at Guthrie Robert Packer Hospital. Limitations of our study include inability to capture prehospital deaths and inability to correlate ISS with prehospital injury severity scores.     

人因可靠性数据库基础架构研究

针对人因可靠性分析中的数据匮乏问题,在技能-规则-知识(SRK)模型的基础上建立层次化的人因可靠性数据分类体系,其中包括人因失误模式和人因失误影响因素。结合对一些实际人因失误数据的考察,以及可控实验,确定人因可靠性数据库中基准人因失误概率。在人因数据外推系统中,使用层次分析法(AHP)来定量评价人因失误所处的情境的等级,并使用概率方法将基准人因失误概率与情境进行叠加,从而得到人因失误概率。人因数据库有助于人因可靠性数据的搜集和分析,形式化的外推方法减少了对主观因素的依赖。     

Multi-level safety climate associations with safety behaviors in the fire service

Introduction: Limited research associated with safety climate has been completed within the fire service. Given this dearth of information, the present study sought to identify a valid and reliable measure of safety climate at both the workgroup and organizational levels within the fire service. Methods: Researchers surveyed 994 firefighters in two large metropolitan fire departments. Preliminary analyses including psychometrics, confirmatory factor analyses, and shared perception analyses were completed. A linear mixed model analysis was then completed to assess the relationships between workgroup safety climate, organizational safety climate, and safety behaviors, including both safety compliance and safety citizenship behaviors. Results: Measures of safety climate at the workgroup (WGSC) and organizational levels (OSC) were derived. WGSC factors include supervisor support (α = 0.92), vertical cohesion (α = 0.89), and horizontal cohesion (α = 0.94). OSC factors include management commitment (α = 0.91), safety programs/policies (α = 0.89), perceived fairness (α = 0.86) and incident command (α = 0.90). Confirmatory factor analyses confirmed our multi-factor models were a good fit to the data. The linear mixed model analysis found that WGSC positively predicted safety compliance behavior (B = 0.13, p < .001) and safety citizenship behavior (B = 0.22, p < .001) and OSC positively predicted safety compliance behavior (B = 0.16, p < .001) and safety citizenship behavior (B = 0.15, p < .001). Conclusions: This work presents reliable and valid measures of both workgroup and organizational safety climate, which have positive relationships with safety behavior outcomes. Practical application: The measures, which were developed through an extensive multi-method process, provide a means for researchers and practitioners to assess safety climate in the fire service and provides guidance for future safety climate research, including informing intervention research, which could potentially bolster safety climate and enhance safety in the fire service.     

基于IEEMD样本熵分析的管道泄漏定位

为降低城市管道泄漏定位误差,提出1种改进的集合经验模态分解（IEEMD）样本熵分析的管道多点泄漏定位方法。首先通过在EEMD中添加自相关函数计算和EMD算法,得到IEEMD;然后应用IEEMD可将原始泄漏信号直接去噪并分解为真实信号分量和冗余分量,经样本熵分析计算剔除冗余分量,获得有效泄漏信号;最后根据互相关时延计算和声发射时差定位法精确计算泄漏点位置。结果表明:该方法泄漏信号提取效果好、计算效率更高,有效提高了信号的信噪比,降低了信号的均方误差;该方法将管道泄漏定位误差降低至4.06%,较大程度提高了管道泄漏定位精确度。     

组织定向的人因失误因果模型及影响关系研究

为了更好地从源头上预防人因失误,首先,基于系统理论发展一个组织定向人因失误的"结构-行为"因果模型,包括组织子模型、情境状态子模型、个体因素子模型以及人因失误子模型。然后,分析模型中各子模型之间以及其因素类别之间的直接和间接影响关系,如模型中各层级之间的因果关系:组织因素→情境状态因素→直接触发人因失误的个体因素→人的认知行为失误。最后,基于事件报告分析、专家意见和文献资料,并依据建立的4种影响类型(I,C,A,N)识别具体情境环境与人因失误或认知行为之间的影响关系。分析结果表明,情境环境因素对人因失误的影响非常复杂,不同的情境环境因素对同一人因失误可能产生不同类型的影响,同一情境环境因素对不同的人因失误可能产生不同类型的影响。     

An illusion of objectivity in workplace investigation: The cause analysis chart and consistency,accuracy, and bias in judgments

Introduction: Investigation tools used in occupational health and safety events need to support evidence-based judgments, especially when employed within biasing contexts, yet these tools are rarely empirically vetted. A common workplace investigation tool, dubbed for this study the “Cause Analysis (CA) Chart,” is a checklist on which investigators select substandard actions and conditions that apparently contributed to a workplace event. This research tests whether the CA Chart supports quality investigative judgments. Method: Professional and undergraduate participants engaged in a simulated industrial investigation exercise after receiving a file with information indicating that either a worker had an unsafe history, equipment had an unsafe history, or neither had a history of unsafe behavior (control). Participants then navigated an evidence database and used either the CA Chart or an open-ended form to make judgments about event cause. Results: The use of the CA Chart negatively affected participants' information seeking and judgments. Participants using the CA Chart were less accurate in identifying the causes of the incident and were biased to report that the worker was more causal for the event. Professionals who used the CA Chart explored fewer pieces of evidence than those in the open-ended condition. Moreover, neither the open-ended form nor the structured CA Chart mitigated the biasing effects of historical information about safety on participants' judgments. Conclusion: Use of the CA Chart resulted in judgments about event cause that were less accurate and also biased towards worker responsibility. The CA Chart was not an effective debiasing tool. Practical application: Our results have implications for occupational health and safety given the popular nature of checklist tools like the CA Chart in workplace investigation. This study contributes to the literature stating that we need to be scientific in the development of investigative tools and methods.     

Trends in aggressivity and driver risk for cars,SUVs, and pickups: Vehicle incompatibility from 1989 to 2016

Abstract

Objective: When 2 vehicles of different sizes collide, the occupants of the smaller vehicle are more likely to be injured than the occupants of the larger vehicle. The larger vehicle is both more protective of its own occupants and more aggressive toward occupants of the other vehicle. However, larger, heavier vehicles tend to be designed in ways that amplify their incompatibility with smaller, lighter vehicles (by having a higher ride height, for example). A 2012 study by the Insurance Institute for Highway Safety (IIHS) concluded that fatalities caused by design incompatibility have decreased in recent years. The current study was conducted to update the 2012 IIHS analysis and to explore trends in vehicle incompatibility over time.

Methods: Analyses examined deaths in crashes involving 1- to 4-year-old passenger vehicles from 1989 to 2016 collected from the Fatality Analysis Reporting System (FARS). Trends in driver risk were examined by comparing driver death rates per million registered vehicle years across vehicle type and size. Trends in aggressivity were examined by comparing partner driver death rates across vehicle type and size.

Results: Cars and SUVs have continued their trend toward reduced incompatibility. In 1989–1992, SUVs were 132% more likely to kill the driver in a partner car compared with when a car crashed with another car. By 2013–2016, this value had dropped to 28%. Pickups and cars remain just as incompatible in 2013–2016 as they were in 1989–1992, however (159% vs. 158%). Remaining pickup incompatibility may be largely due to excess curb weight rather than to shape or design features, because light pickups were just 23% more likely to kill the driver in a partner car compared with when a car crashed with another car.

Conclusions: The trend toward reduced fleet incompatibility has continued in the latest crash data, particularly for cars and SUVs. Although pickup–car incompatibility has also decreased over time, pickups remain disproportionately aggressive toward other vehicles, possibly due to their greater average curb weight. Reducing the weight of some of the heaviest vehicles and making crash avoidance technology fitment more widespread may be promising means to reduce remaining fleet incompatibility. Identifying the source of remaining incompatibility will be important for safety improvements going forward.     

Measurement Accuracy of the Electrosensitive Protective Device Response Time When Using the Double Penetration Method

The Double Penetration Method (DPM) method of measuring ESPD (Electrosensitive Protective Device) response time was presented by Dzwiarek (1997). Calibrating the measuring equipment is a crucial stage of the procedure. Experimental verification of theoretical predictions is also crucial. For calibration purposes, a device simulating real ESPD operation thus enabling a correct setting of the response time was designed (Dzwiarek, 1997).

Theoretical analysis has shown that measuring ESPD response time with the DPM is subject to localisation errors made in the localisation of the detection zone border, rod position measurement errors made during highspeed penetration, and time delay measurement errors. The values of all those components of the total error have been determined experimentally using the calibrating device. Measurements have been taken under conditions as close to real ones as possible proving that the total measurement error is really enclosed within the assumed limits.     

基于Hankel-DMD的城市交通事故风险时空预测*

为解决城市交通事故风险时空分布预测任务中时空关联性捕捉困难的问题,提出基于动态模态分解（DMD）的城市交通事故分析时空预测模型,模型利用总最小二乘法去除交通事故数据中的噪声,应用结合Hankel矩阵的动态模态分解模型（Hankel-DMD）捕捉交通事故风险的时空关联性,对交通事故风险的时空分布进行预测。研究结果表明:DMD框架能够为高维预测任务提供低秩解决方案,从高维数据中捕捉时空关联性;Hankel-DMD模型在预测评价指标平均绝对误差和均方根误差方面的表现明显优于统计学及机器学习等方法;Hankel-DMD模型产生的动态模态和特征值,对事故风险系统的时空动态特征具有一定的可解释性,同时验证Hankel-DMD模型的适用性。     

民航安全信息管理人员能岗匹配测评研究*

为了给大数据环境下的民航企事业单位招聘、选拔和评价安全信息管理人员提供科学依据,基于胜任力内涵、数据生命周期理论与信息处理层次结构理论,构建以数据科学知识等为支撑、民航安全数据挖掘能力等为核心、民航安全信息管理态度为导向的民航安全信息管理人员胜任力概念模型与胜任力指标体系。运用贝叶斯优化BPNN的超参数,构建民航安全信息管理人员能岗匹配测评模型,并从数值仿真、趋势拟合方面进行模型验证,从均方根误差与算法耗时方面与单一BPNN和随机搜索优化BPNN对比性能。研究结果表明:所构建的测评模型可行性强、误差更小、收敛速度更快,能够更加客观准确地反映安全信息管理人员能岗匹配的实际情况。     

钢筋混凝土矩形梁的当量耐火时间

为了正确评估钢筋混凝土矩形梁在实际火灾条件下的承载力,以一般室内火灾轰燃后的房间平均温度一时间曲线为火作用,以热传导和建筑结构耐火理论为基础,以梁在实际和标准火灾下的承载力相等为等效准则,用数值计算方法研究梁的当量耐火时间,并用最小二乘法导出其计算公式.火灾房间火灾荷载越大,开口因子越小,梁的当量耐火时间越长;反之越短.所给梁的当量耐火时间可用于建筑结构性能化耐火设计与评估:当火灾荷载密度较大,开口因子较小时加大梁的截面参数以获得安全性,反之,减小截面参数以获得经济性.