基于贝叶斯网络的过度训练诊断     

马振龙 《计算机与农业》2007,(12):56-59

针对贝叶斯网络在具体应用时需根据实际应用问题来建立相应的贝叶斯网络模型的问题，采用Java语言设计并开发了一套贝叶斯网络预测诊断系统，用来实现因果推理和诊断推理。重点讨论了系统的网络数字化、网络学习、网络推理等关键问题。利用该平台建立了过度训练的诊断网络，通过试验证明得出，此套系统简单易用、通用性强，可应用于不同的研究领域。  相似文献   

改进Hargreaves模型估算川中丘陵区参考作物蒸散量   总被引：3，自引：2，他引：3  

李晨  崔宁博  魏新平  胡笑涛  龚道枝 《农业工程学报》2015,31(11):129-135

为提高Hargreaves-Samani(HS)模型参考作物蒸散量(ET0)计算精度,该文基于贝叶斯原理利用川中丘陵区1954-2002年逐日资料对其温度指数、温度系数和温度常数进行改进,并使用2003-2013年资料以Penman-Monteith(PM)模型为标准评价HS改进模型计算精度与适应性。结果表明:HS改进模型参数在川中丘陵区各区均小于联合国粮农组织推荐值,并呈现出随纬度上升而增大的趋势;与PM模型计算结果相比,HS改进模型计算的ET0相对误差在川中丘陵区北部从14.2%~60.9%降至-1.1%~33.4%、中部从40.6%~92.6%降至16.9%~61.1%、南部从31.3%~96.0%降至8.5%~64.4%、整个川中丘陵区从32.1%~82.7%降至9.5%~52.6%;相关性分析表明,HS改进模型和PM模型计算的ET0回归曲线的斜率更接近于1(北部1.16、中部1.02、南部0.99、全区1.13),决定系数均达到0.85(P0.01)以上;趋势分析表明,HS改进模型和PM模型计算的ET0变化一致,年内均呈开口向下的抛物线状,年际均呈微小上升趋势。因此,基于贝叶斯原理改进的HS模型在川中丘陵区不同区域变异性较小,适应性较强,具有较高的计算精度,可作为川中丘陵区参考作物蒸散量简化计算的推荐模型。  相似文献   

单变量动态Gamma分布模型及贝叶斯预测   总被引：2，自引：0，他引：2  

陈传勇 《仲恺农业技术学院学报》2004,17(1):49-52

给出了观测值服从单变量Gamma分布,并在自然参数与状态参数之间满足线性关系ωt=F'tθt的假设下,利用共轭分布给出了相应模型的修正递推及其预测公式.  相似文献   

广东大宝山矿区土壤侵蚀强度判别方法研究     

陈三雄  谢莉  廖建文  向慧昌  付秋豹 《中国水土保持》2007,(11):40-42

大宝山矿区土壤侵蚀主要由露天采矿及排弃的废渣土引起。通过对大宝山矿区范围内40个典型坡面及其中11个特征坡面的调查分析,认为坡长、坡高、侵蚀沟密度可以用来表征矿区不同坡面的土壤侵蚀特性及其差异。采用聚类分析、判别分析方法,构建了矿区土壤侵蚀强度判别模型,模型平均误判概率为0.07,具有较高的可信度。经判别,大宝山矿区现有水土流失面积324.48 hm2,占矿区总面积的48.8%,土壤侵蚀严重,由此引发了一系列环境问题,必须尽快进行整治。由本研究方法得到的侵蚀模数虽只是一个范围值,需要内插后才能得到确定值,但这对于编制水土保持方案、布置水土保持措施来讲,已完全能够满足要求。  相似文献   

Development and validation of numerical indexes integrating enzyme activities of soils   总被引：1，自引：0，他引：1  

E. Puglisi  M.A. Rao 《Soil biology & biochemistry》2006,38(7):1673-1681

Three numerical indexes, based on soil enzyme activities, which can discriminate between altered and unaltered soils, are presented and validated. Seven enzymes were measured at three different agricultural sites subjected to contamination by municipal and industrial wastes (site 1), intensive cultivation without either crop rotation or organic fertilization (site 2), and irrigation with brackish water (site 3). At each site neighbouring, unaltered soils were sampled and analysed as control soils.The three indexes were developed by means of canonical discriminant analysis (CDA) using data from two sites, while the third site was kept aside as an independent test. The first index is built up using all of the enzyme activities studied; the second, using the most discriminating enzyme activities only; and the third was developed for testing against data in published papers that deal with enzyme measurements in soils under different management regimes.The three indexes were able to discriminate between altered and control soils when applied to the data set from the third site, i.e., that not used for index development. When tested against published data, the third index was usually able to discriminate altered soils from controls by higher index scores. This index was successful in most cases: it was consistently able to classify soils according to their reported degree of alteration. Our results confirm that enzyme activities are suitable indicators of soil alteration and may be usefully integrated to develop soil alteration indexes suitable for monitoring soil status under different conditions.  相似文献   

The Application of Multivariate Statistical Methods for the Evaluation of Soil Profiles (8 pp)     

Kerstin Sielaff  Jürgen W. Einax 《Journal of Soils and Sediments》2007,7(1):45-52

Background, Aim and Scope   Contamination of soils does not only occur on their surface over large areas, but also in depth. Therefore a characterization of soil state after pollution demands a three-dimensional soil sampling, by what a large number of samples has to be analyzed. Analytical results could be evaluated by multivariate statistical methods, which have already been used for the evaluation of data sets containing results from soil sampling of two dimensions like areas or single profiles. In this case study, multivariate statistical methods were applied to investigate structure and interactions between features in a data set containing results of three-dimensional soil sampling. The investigated soil profiles were contaminated by emissions of a former cement and phosphate fertilizer plant. The aim of this study was to determine the remaining extent of contamination and to analyze whether pollutants are mobilized and vertically transported within the profiles. Materials and Methods: Three soil profiles were sampled in the surroundings of the plant. Grain size, organic and carbonatic bonded carbon, pH value, and the total contents of Ca, Cd, Co, Cu, F, Fe, K, Mn, Mg, Na, Ni, P, Pb, and Zn were determined. The resulting data set was evaluated by cluster analysis, linear discriminant analysis, and principal components analysis. The sequential extraction procedure according to Zeien and Brümmer was applied to analyze the binding properties of Ca, Cd, Cu, Na, Pb, and Zn from selected samples. Results: Cd was identified as contaminant of the top soils. The pH values of the bottom soils were determined to be in alkaline range, which is unnaturally high. Variables were clustered according to enrichment of variables in top soils. The samples were classified regarding their pollution state and their substrate by cluster analysis, which was confirmed by linear discriminant analysis. Geogenic and anthropogenic sources of variables as well as relationships between variables like the binding of heavy metals at organic matter were detected by using principal components analysis. The binding of heavy metals at organic matter in the top soils was confirmed by the results of the applied sequential extraction. A vertically altered distribution of Na binding was determined. Discussion: According to the current soil conditions, the uptake of heavy metals had probably occurred by the over ground part of plants during the deposition. The distribution of Na should likely result from the vertical transport of Na, which would also explain the high pH values of the bottom soils by ion exchange. Altogether, the main amount of deposited Ca, F, Na, P, and heavy metals is likely nearly insoluble bound in the top soils. Conclusions: Ten years after the end of production, the pollution of top soils in the surroundings of the former plant is still high. However, regarding the ecotoxicological relevance the now explored interactions between several soil features and elements strongly indicate that there is no short-term to medium-term risk of a mobilization of the deposited elements with the exception of Na. Recommendations and Perspectives: The results of this case study prove that multivariate statistical methods are powerful tools to explore interactions of variables and relationships in a data set derived from three dimensional soil sampling. The methods applied in this work can be highly recommended for evaluations of large data sets resulting from two- or three-dimensional samplings. Multivariate statistical methods enable the characterization of soils and their pollution state in a simple and economic way.  相似文献   

大数据挖掘在食品安全风险预警领域的应用     

王雅洁  杨冰  罗艳  何锦林  谭红 《安徽农业科学》2015,(8):332-334

食品安全综合评价与预警是食品安全的重难点.该研究着重介绍了大数据挖掘在食品安全风险预警领域的应用.首先对大数据的基本概念及3种典型的大数据挖掘技术(贝叶斯网络、决策树以及人工神经网络)概念进行分析,并探讨这3种大数据挖掘方式在食品安全行业的应用现状.之后比较3种大数据挖掘方式,提出将其中一种大数据挖掘方式BP神经网络运用于食品安全风险预警的构想.  相似文献   

Potential of differential somatic cell counts as indicators of mastitis in quarter milk samples from dairy cows     

U Emanuelson  P Wever 《Acta veterinaria Scandinavica》1989,30(4):475-481

Bacteriological status, somatic cell counts and proportions of lymphocytes, granulocytes and monocytes were determined in 1,659 quarter milk samples from 39 dairy cows. Discriminant analysis was performed in order to assess the ability of total and differential somatic cell counts and combinations of total somatic cell count and each of the differential cell counts, to discriminate between infected and pathogen-free quarters, as well as between quarters infected with minor pathogens and quarters infected with major pathogens. Total somatic cell count classified 82.9% of all quarters correctly with respect to bacteriological status. Differential somatic cell count was less effective than total somatic cell count in discriminating between infected and pathogen-free quarters, as well as between quarters infected with minor vs. major pathogens. Combination of total and differential somatic cell counts did not improve the rate of correctly classified quarters. Inclusion of demographic data into the discriminant function increased the number of quarters correctly classified, mainly through an increase in the proportion of correctly classified infected quarters.  相似文献   

Seroprevalences for ovine enzootic abortion in Switzerland     

Borel N  Doherr MG  Vretou E  Psarrou E  Thoma R  Pospischil A 《Preventive veterinary medicine》2004,65(3-4):205-216

Our aim was to assess the seroprevalence of Chlamydophila (Cd) abortus (Chlamydia psittaci serotype 1), denoted ovine enzootic abortion (OEA), in the Swiss sheep population. A competitive enzyme-linked immunosorbent assay (cELISA) was adapted for the investigation of pooled serum samples (pool approach) and receiver-operator characteristic (ROC) analysis was applied to define the cut-off of the pool approach. At a cut-off value of 30% inhibition, the flock-level pooled sensitivity and specificity were 92.9% and 97.6% when compared to classifying the flock based on individual-animal samples.

Subsequently, sera from 775 randomly selected flocks out of 11 cantons of Switzerland were investigated using the pool approach. The cantons included in the study represented 72% of the Swiss sheep flocks and 76% of Swiss sheep population. Antibodies against Cd. abortus were found in almost 19% (144) of the 775 examined sheep flocks. Test prevalences were adjusted for the imperfect test characteristics using the Rogan–Gladen estimator and Bayesian inference. Seroprevalence was highest (43%) in the canton Graubünden. In the remaining 10 cantons the seroprevalence ranged from 2 to 29%. The cELISA in combination with testing pooled sera and statistical methods for true prevalence estimation provided a good survey tool at lower costs and time when compared to other approaches.  相似文献   

Forecasting clinical disease in pigs: comparing a naive and a Bayesian approach     

Baadsgaard NP  Højsgaard S  Gröhn YT  Schukken YH 《Preventive veterinary medicine》2004,64(2-4):85-100

In veterinary practice the clinician often evaluates and predicts herd health status over time according to clinical criteria. In this paper, we modeled three different clinical signs among pigs based on longitudinal clinical observations in 15 pig herds. We compared and discussed the outputs from two different approaches for making clinical forecasts in a herd: a naive approach using a simple time series model with previous disease observations as predictors and a Bayesian state space models approach, in which the time lag variable entered into the random component of the model. We used the Markov chain Monte Carlo technique to calculate posterior distributions of the forecasts. For the herd specific forecasts the results showed that there were only minor differences between the forecasts from the simple time series model and the median forecasts from the Bayesian model. However, the credibility intervals from the Bayesian model were wider than the forecasts from the simple model and, therefore the Bayesian model encompassed the variability in the forecasts better. Compared to the statistical model, the simple time series would be easier to implement in a practical setting. However, the latter lacks the inherent “generality” from the statistical model that allows the user to make statements about the distribution of the herds and to predict disease status based on the “average” correlation among the herds. The applicability of the Bayesian approach within a clinical decision-making framework was discussed, with special emphasis on the use of prior information and clinical forecasting.  相似文献