基于改进粒子群优化的并行极限学习机*

为了提高极限学习机(ELM)网络的稳定性,提出基于改进粒子群优化的极限学习机(IPSO-ELM)。结合改进的粒子群优化算法寻找ELM网络中最优的输入权值、隐层偏置及隐层节点数。通过引入变异算子,增强种群的多样性,并提高收敛速度。为了处理大规模电力负荷数据,提出基于Spark并行计算框架的并行化算法(PIPSO-ELM)。基于真实电力负荷数据的实验表明,PIPSO-ELM具有更高的稳定性及可扩展性,适合处理大规模的电力负荷数据。     

铜厂矿区外围金矿勘查与开发

根据德兴铜厂地区所处区域地质特征及成矿地质条件,系统地分析了周边伴生金铜矿床和金矿床中矿体的控矿要素,结合铜厂矿区多年来对伴生金银工作的成果,简要地对铜厂周边金矿勘查与开发提出了见解,并阐述了部分成矿预测区和重点含金地质体。     

大坝变形预测的支持向量机模型

支持向量机在解决非线性及高维模式识别问题中表现出其特有的优势。针对大坝变形具有强非线性的特点以及传统神经网络模型预测时存在局部极小与过学习等问题,将支持向量机应用于大坝变形预测。以某大坝的变形监测数据为例,建立SVM预测模型,将支持向量机模型与神经网络模型进行比较分析。结果表明,该模型具有较高的精度,可以很好地应用于大坝变形监测分析。     

中间包内钢水温度预测模型

采用多元回归分析建立了60t中间包内钢水温度-T中间包/ ℃的预测模型,并对中间包内预测温度进行了统计分析.结果表明：该模型的钢水温度预测值与实测值非常吻合.     

粒子滤波参数估计方法在齿轮箱剩余寿命预测中的应用研究

针对非线性非高斯系统的剩余寿命（RUL）预测问题,本文提出了一种基于粒子滤波（PF）理论的设备剩余寿命预测方法。首先建立设备的非线性状态空间模型（含有未知的时变参数）,然后通过粒子滤波算法估计出设备状态的概率密度函数（PDF）,从而根据该PDF计算出设备的RUL。此外,计算设备RUL的期望值和95%置信区间,并对模型的预测效果进行评估,验证预测的有效性和准确性。最后通过齿轮箱的全寿命实验,对本文所提方法的有效性进行实例验证,将实验结果和传统的比例风险模型（PHM）预测结果对比分析,结果表明本文提出的剩余寿命预测方法要优于传统的PHM预测方法。     

A novel molecular interaction chemical model for silicate melts

A novel molecular interaction chemical model (MICM) for silicate melts has been suggested based on statistical thermodynamics. It can simultaneously predict activities of all components in the CaO-FeO-SiO₂ and CaO-Al₂O₃-SiO₂ melts using only four binary parameters for each binary melt which can be determined by fitting activities of its two components. The results indicate that the predicted values of activity of FeO and SiO₂ are in good agreement with the experimental data at 1823 K and 1873 K, and those of CaO, SiO₂ and Al₂O₃ are in reasonable agreement with the graphical integration data of the Gibbs-Duhem equation. This shows that the model is effective in which the physical interaction plays a main role and the chemical one does the auxiliary function.     

基于Fisher判别分析理论的地下开采安全评价模型

矿山安全是矿山企业稳定、持续、高速发展的根本保证。应用Fisher判别分析理论并结合矿山安全生产的标准规范,根据矿山人、机、环境、管理4个因素的实际特点,综合考虑到地下矿山的开采工艺和指标数据的可靠性,选取23项矿山安全评价因素作为评价影响因子,建立矿山地下开采安全评价的F isher判别分析模型(FDA)。对18组地下矿山的特征参数值作为学习样本进行训练和检验,回判估计的误判率为0。利用该模型对另外4组实测数据作为预测样本进行测试,预测结果与实际情况吻合较好。研究结果表明,该方法回判估计的误判率低,判别性能良好,对地下矿山安全管理现状及水平评价提供了可操作的方法,进而为矿山有关部门提供了科学安全管理的依据。     

人工神经网络在焊接中的应用研究和发展

本文重点综述了人工神经网络(ANN)技术在焊接工艺参数设计、焊接性能预测、焊接过程控制方面的应用研究现状,并指出了它在焊接应用研究中的问题以及未来的发展趋势。     

The stability of long-term trends in the number of traffic fatalities in a sample of highly motorised countries

This paper examines the stability of long-term trends in the number of traffic fatalities in eight highly motorised countries: Norway, Sweden, Denmark, Finland, The Netherlands, Great Britain, Australia and The United States. In all these countries, the number of traffic fatalities reached a peak around 1970-1972 and has since declined. The decline has, however, been irregular and fairly long periods of stagnation or even increase in the number of fatalities have occurred in all countries. A stable trend is defined in this paper as a trend that remains unchanged and therefore can be used to successfully predict the future number of fatalities by means of extrapolation. It is concluded that the trends towards fewer fatalities in the countries selected have not been stable and that even trend lines that fit past trends very closely are usually worthless for predictive purposes. An attempt was made to identify factors influencing long-term trends by fitting negative binomial regression models to fatality data. Although some of the models fitted the data marginally better than simple trend lines, these models are not necessarily more useful for predicting future trends than simple trend lines, since using the models for prediction requires that future changes in all explanatory variables must be predicted. The chief lesson is that past trends do not provide a reliable basis for predicting future developments with respect to the number of traffic fatalities.     

基于改进人工神经网络的LF钢水终点温度预报

采用改进的人工神经网络算法,开发了40t钢包炉精炼时钢水终点温度预报模型。与传统BP网络算法相比较,改进算法可提高预测速度和精度。生产现场实验表明,传统BP神经网络算法,钢水温度预测误差±5℃的炉次仅为77%,用改进的BP神经网络算法,其误差±5℃的炉次为90%。