基于BP神经网络的天线识别

本文比较研究了BP神经网络中的几种常用算法,针对这些不同算法下的BP神经网络进行训练,并得出了各自网络的性能。在此基础上,针对经典BP算法和LM算法进行对比研究,找到LM算法的改进之处。此外,在实际的应用中表明,不仅不同的BP算法对网络的运算速度、泛化能力等有较大的影响,而且BP神经网络对隐含层神经元数目也很敏感。我们希望在BP神经网络的基础上,构建一种合适的天线模型,来应用于天线的分类识别,这将具有很大的现实意义。     

基于深度学习的大数据空气污染预报

为了更好地反映环境污染变化趋势,为环境管理决策提供及时、全面的环境质量信息,预防严重污染事件发生,开展城市空气质量预报研究是十分必要的.本文针对环境大数据时代下的城市空气质量预报,提出了一种基于深度学习的新方法.该方法通过模拟人类大脑的神经连接结构,将数据在原空间的特征表示转换到具有语义特征的新特征空间,自动地学习得到层次化的特征表示,从而提高预报性能.得益于这种方式,新方法与传统方法相比,不仅可以利用空气质量监测、气象监测及预报等环境大数据,充分考虑污染物的时空变化、空间分布,得到语义性的污染物变化规律,还可以基于其他空气污染预测方法的结果(如数值预报模式),自动分析其适用范围、优势劣势.因此,新方法通过模拟人脑思考过程实现更充分的大数据集成,一定程度上克服了现有方法的缺陷,应用上更加具有灵活性和可操作性.最后,通过实验证明新方法可以提高空气污染预报性能.     

RESEARCH: Comparison of Temporal Trends in Ambient and Compliance Trace Element and PCB Data in Pool 2 of the Mississippi River, USA, 1985–1995

6+ ), copper (Cu), lead (Pb), mercury (Hg), nickel (Ni), selenium (Se), zinc (Zn), and polychlorinated biphenyls (PCBs). Water-column, bed-sediment, and fish-tissue (fillets) data collected by five government agencies comprised the ambient data set; effluent data from five registered facilities comprised the compliance data set. The nonparametric Mann-Kendall trend test indicated that 33% of temporal trends in all data were statistically significant (P < 0.05). Possible reasons for this were low sample sizes, and a high percentage of samples below the analytical detection limit. Trends in compliance data were more distinct; most trace elements decreased significantly, probably due to improvements in wastewater treatment. Seven trace elements (Cr, Cd, Cu, Pb, Hg, Ni, and Zn) had statistically significant decreases in wastewater and portions of either or both ambient water and bed sediment. No trends were found in fish tissue. Inconsistency in trends between ambient and compliance data were often found for individual constituents, making overall similarity between the data sets difficult to determine. Logistical differences in monitoring programs, such as varying field and laboratory methods among agencies, made it difficult to assess ambient temporal trends.     

HYDROLOGIC AND HYDRAULIC RESEARCH IN MOUNTAIN RWERS1

ABSTRACT: Although our current (1990) knowledge of hydrologic and hydraulic processes is based on many years of study, there are river environments where these processes are complex and poorly understood. One of these environments is in mountainous areas, which cover about 25 percent of the United States. Use of conventional hydrologic and hydraulic techniques in mountain-river environments may produce erroneous results and interpretations in a wide spectrum of water-resources investigations. An ongoing U.S. Geological Survey research project is being conducted to improve the understanding of hydrologic and hydraulic processes of mountainous areas and to improve the results of subsequent hydrologic investigations. Future hydrologic and hydraulic research needs in mountainous areas are identified.     

秸秆还田、覆膜和施氮对旱地麦田土壤质量的影响

通过5 a大田定位试验研究了秸秆还田、覆膜及氮肥对旱地土壤质量的影响.以双因素裂区设计,主处理为3种栽培模式:常规栽培(CT)、秸秆还田(SR)和覆膜栽培(FM);副处理为3种施氮水平:0、144和180 kg·hm-2.在第5季冬小麦收获后采集耕层0～20 cm土壤样品,测定容重(BD)、孔隙度(SP)、田间持水量(FC)、有机碳(SOC)、全氮(TN)、全磷(TP)、全钾(TK)、矿质氮(NO3--N + NH4+-N)、速效磷(AP)、速效钾(AK)、微生物生物量碳(MBC)、微生物生物量氮(MBN)、蔗糖酶(SA)、碱性磷酸酶(ALP)、脲酶(UA)和脱氢酶(DHA)等土壤理化性质及生物性质等指标,基于两个最小数据集(MDS),利用线性(LT)和非线性(NLT)评分模型计算土壤质量指数(SQI),通过与冬小麦产量拟合来寻找最优SQI,据此分析土壤质量对秸秆还田、覆膜及氮肥的响应.结果表明,SR处理的SP、FC、SOC、TK、SA、ALP、DHA、MBN和MBC分别较CT显著高10.1%、40.5%、28.3%、14.3%、22.0%、8.9%、48.3%、35.9%和 62.5%.FM处理的SP、FC、NO3--N、AP、AK、ALP和MBN分别较CT显著高3.1%、14.6%、729.0%、37.4%、28.2%、7.2%和17.7%.SR和FM处理冬小麦产量分别较CT显著高25.1%和18.6%.与N0相比,施用氮肥土壤SP、TN、NO3--N、NH4+-N、ALP和MBN显著增加3.4%～407.6%,BD、SOC和MBC/MBN显著下降3.0%～31.7%.在3种栽培模式下,施氮显著增加冬小麦产量,增幅为69.4%～79.1%,相同栽培模式两施氮水平间差异不显著.基于MDS2的非线性评分模型计算的SQI能够更准确地评价旱地麦田土壤质量.在NLT-SQI2中,SR、FM和氮肥均可增加土壤的SQI,在SR配施N180水平下SQI达到最大,FM次之,CT最小.综上,秸秆还田和覆膜配施适量氮肥能维持和提高土壤质量,秸秆还田优于覆膜.