汽车制造企业绿色制造模式评价研究

针对汽车制造企业的绿色制造模式建立了包括时间、质量、成本、服务、资源及环境评价因数集,并将其分解成25项亚指标评价体系;然后针对该体系引入一种基于遗传优化的BP人工神经网络评价方法,从而通过遗传算法确定最佳隐含层单元数、最佳学习速率和最佳动量因子;最后在已确定的这些参数条件下,通过BP人工神经网络实现了汽车制造企业绿色制造模式的评价,并对评价结果进行了分析。     

Simultaneous biohydrogen production and starch wastewater treatment in an acidogenic expanded granular sludge bed reactor by mixed culture for long-term operation

The biofilm-based expanded granular sludge bed (EGSB) reactor was developed to treat starch-containing wastewater and simultaneously recovery hydrogen by mixed microbial culture. Granular activated carbon (GAC) was used as the support media. Operating at the temperature of 30 °C for over 400 days (data not shown), the EGSB reactor presented high efficiency in hydrogen production and COD removal ability. The maximum hydrogen production rate (HPR) was found to be 1.64 L/L.d under the organic loading rate (OLR) of 1.0 g-starch/L.d, pH of 4.42 and HRT of 4 h. The hydrogen yield (HY) peaked at 0.11 L/g-COD, under the OLR of 0.5 g-starch/L.d, pH of 3.95 and HRT of 8 h. Hydrogen volume content was estimated to be 35–65% of the total biogas. The average COD removal rate was 31.1% under the OLR of 0.125 g-starch/L.d and HRT of 24 h. The main dissolved fermentation products were ethanol, acetate and butyrate. The average attached biofilm concentration was estimated to be 8.26 g/L, which favored hydrogen production and COD removal. It is speculated that the low pH operation in the present system would contribute significantly to lower the cost of alkaline amount required for pH control in the continuous operation, especially in the scale-up biohydrogen producing system. A model, built on the back propagation neural network (BPNN) theory and linear regression techniques, was developed for the simulation of EGSB system performance in the biodegradation of starch synthesis-based wastewater and simultaneous hydrogen production. The model well fitted the laboratory data, and could well simulate the removal of COD and the production of hydrogen in the EGSB reactor.     

基于CSSA-BPNN模型的胶结充填体动态抗压强度预测

充填采矿法二步骤回采时胶结充填体稳定性受爆破扰动而降低。为快速准确地获得充填体动态抗压强度,利用分离式霍普金森压杆(SHPB)进行了40组不同应变率的单轴冲击实验,以灰砂比、充填体密度、养护龄期和平均应变率作为输入参数,充填体动态抗压强度作为输出参数,建立了一种基于Logistic混沌麻雀搜索算法(CSSA)优化BP神经网络(BPNN)的预测模型,并与传统BPNN和麻雀搜索算法优化的BPNN进行了对比分析。研究结果表明：CSSA-BPNN模型的平均相对误差为4.11％,预测值与实测值之间拟合的相关系数均在0.96以上,模型预测精度高。CSSA-BPNN模型的均方根误差为0.3950 MPa,平均绝对误差为0.3592 MPa,决定系数为0.9952,均优于另外两种预测模型。实现了对充填体动态抗压强度的准确预测,可大幅减小物理实验量,为矿山胶结充填体的强度设计提供了一种新方法。     

BP Neural Network Classification of Remote Sensing Images based on DT\|CWT Decomposition

In order to solve the ambiguity and uncertainty of high resolution multi\|spectral remote sensing image classification and to better overcome the influence of noise,a new BPNN（Back Propagation Neural Network）classification method of multi\|spectral image,based on DT\|CWT decomposition,is presented in this paper.First,the NDVI and texture features of the image are extracted to reduce the classification uncertainty caused by the problem of different objects having the same spectrum and the same objects having different spectrum in the image,then,the original spectral band,NDVI and texture features of the image are decomposed by DT\|CWT to extract the Low\|frequency information of the image,as well as to reduce the image noise and the presence of “salt and pepper” in the classification.Finally,the extracted low\|frequency sub\|graphs are input to the BP neural network and classified according to the trained network to obtain the final classification result.The results of the comparison show that the proposed method with less miscellaneous points has stronger regional consistency,higher classification accuracy and better robustness.     

基于BP神经网络的透明转发卫星功放预失真

研究卫星通信功放性能优化问题,传统的预失真技术通常用来补偿地面功放的非线性失真或仅考虑卫星功放的失真补偿,线性化性能有限.为解决上述问题,提出了一种适合透明转发卫星的星地一体BP神经网络预失真算法.改进算法的学习结构同时考虑了卫星地球站固态功放和透明转发卫星功放的记忆非线性特性,利用带抽头延迟的BP神经网络作为预失真器,并结合收敛速度较快的Levenberg-Marquardt算法对其权值和阈值矢量进行自适应更新.仿真结果表明,经过神经网络预失真的星座图误差矢量幅度改善了84.67％,输出信号功率谱带外再生抑制提升近了13 dB,线性化效果十分显著.     

改进PSO-BPNN的电力变压器故障诊断与模式识别

为了优化反向传播网络相关学习参数,提出一种粒子群优化辅助BP神经网络(BPNN)的新方法.以变压器油中气体体积分数百分比构造故障特征,将BP网络的初始权值和阈值进行实数编码,以对应PSO中的粒子,实现BP网络的离线训练与在线分析,对变压器故障模式做出判断.结果表明,该算法更合理地更新了粒子的位置和速度,最优地设置了全局极值,有效克服了粒子的早熟收敛,获得的故障诊断准确率高达91%,并大大提升了BP网络的收敛速度.该算法为此类设计提供了有效的模型参考.     

基于半监督机器学习的滑坡易发性预测建模

为了克服滑坡编录样本不足、扩充滑坡样本较困难、主观随机选择的非滑坡样本准确性较低等缺点,以江西省南康区为例,拟用半监督卡方自交互侦测决策树(SSCHAID)和半监督反向传播神经网络(SSBPNN)进行滑坡易发性预测(LSP), 在已知滑坡样本和随机选取的非滑坡样本基础上,用全监督机器学习将初始LSP划分成不同级别;将高分辨率遥感影像和初始滑坡易发性图中的极高易发区叠加,筛选一定数量的潜在滑坡栅格单元扩充滑坡样本;从极低易发区选取非滑坡栅格单元组合成新的输出变量;将新的输出变量导入全监督机器学习,获得最终LSP并评价其精度. 结果表明：半监督机器学习的LSP精度远高于全监督机器学习的LSP精度.     

一种搜索编码法及其在监督分类中的应用

纠错输出码作为监督分类领域中的一个新的研究方向,是提高分类器泛化能力的一种有效方法,但目前还没有通用的确定性编码方法.分析了现有纠错输出码的性质,提出一种搜索编码法,该方法通过对整数空间的顺序搜索,获得满足任意类别数目与最小汉明距离要求的输出码;然后探讨了基于搜索编码的监督分类技术.对简单贝叶斯与BP神经网络算法进行实验,结果表明,搜索编码法可作为一种通用的编码方法用于提高监督分类器的泛化能力.     

电子鼻技术对猪肉挥发性盐基氮的预测研究

为预测不同肥瘦配比猪肉的新鲜度,对4℃恒温贮藏条件下的新鲜猪肉进行挥发性盐基总氮(Total Volatile Basic Nitrogen,TVB-N)检测和营养成分检测,同时利用电子鼻技术检测挥发性气味的信息。以传感器阵列特征值为自变量建立蛋白质、脂肪的回归预测模型,分别对不同肥瘦配比的猪肉样本建立不分类和分类2种TVB-N神经网络预测模型。结果表明:先分类再建立神经网络模型预测的效果更优,将样本进行二分类建立2个模型后,模型训练组的相关系数达0.994、0.985(p<0.01),预测组的相关系数达到0.984、0.979(p<0.01);模型的绝对误差小而且分布区间集中,训练组和预测组各有86%、62.6%的样本的绝对误差在0~1之间;训练组中没有绝对误差大于2.5的样本,预测组中仅有8.5%的样本绝对误差大于2.5。电子鼻传感器特征信号与TVB-N数据具有很强的相关性,电子鼻可以快速预测出不同肥瘦配比猪肉在贮藏期间TVB-N含量的变化,进而无损的评价猪肉的新鲜度。     

结合ERDAS的基于BPNN遥感图像分类

基于改进BPNN的遥感图像分类,先利用ERDAS Imagine的Classifier模块对原始遥感图像进行聚类分析并对比,确定分类类别数.后在原始图像上采集各类别的训练样本和测试样本,通过输入训练样本集训练BP神经网络分类器.最后输入整幅遥感图像到训练好的分类器,得到分类图像.试验表明,其精度优于传统统计分类方法.