AT the age of 32, I went back to school. I enrolled as a student in the Department of History at a night college attached to Beijing Normal University. My son was only two years old at the time. At first I went to nig...AT the age of 32, I went back to school. I enrolled as a student in the Department of History at a night college attached to Beijing Normal University. My son was only two years old at the time. At first I went to night classes just to earn a college diploma. I already had 15 years’ work experience, including four years of being a magazine editor. But if I wanted to keep my job as editor, I had to obtain a diploma. When I first sat down in the terraced, 300-seat classroom and touched my new textbooks, I was full of a craving for knowledge. I knew that my展开更多
农机备件是农机维修的重要基础,是农机故障及时维修和农业生产正常开展的必要保障,因此,对农机备件需求量的精准预测至关重要。然而,农机备件的需求量具有非平稳性、非线性、多零值、波动大等特点,使得预测任务变得困难。本文提出了一...农机备件是农机维修的重要基础,是农机故障及时维修和农业生产正常开展的必要保障,因此,对农机备件需求量的精准预测至关重要。然而,农机备件的需求量具有非平稳性、非线性、多零值、波动大等特点,使得预测任务变得困难。本文提出了一种基于卷积神经网络的时间序列高效卷积网络(Time series efficient convolution network,TECNet),用于农机备件需求量的预测。该模型首先利用快速傅里叶变换对原始一维序列进行周期性提取,然后根据周期性构建二维时间序列卷积模块进行特征提取,最后将二维特征重塑回一维特征,并通过线性变换得到预测值。利用某农机备件供应商4种不同备件类型的销售数据进行了评估验证,并引入均方根缩放误差作为衡量指标,以统一不同序列间的预测效果。试验结果表明,提出的模型预测效果显著优于其他参考模型,4种不同备件需求量预测的均方根缩放误差分别为0.775、1.349、0.822、0.205,均表现出良好的预测效果。该模型能有效考虑时间序列中的时间依赖关系,具有捕捉时间序列数据中非线性模式的能力,对不同农机备件类型的预测任务均能取得良好的效果,可为预测农机备件需求量提供参考。展开更多
文摘AT the age of 32, I went back to school. I enrolled as a student in the Department of History at a night college attached to Beijing Normal University. My son was only two years old at the time. At first I went to night classes just to earn a college diploma. I already had 15 years’ work experience, including four years of being a magazine editor. But if I wanted to keep my job as editor, I had to obtain a diploma. When I first sat down in the terraced, 300-seat classroom and touched my new textbooks, I was full of a craving for knowledge. I knew that my
文摘农机备件是农机维修的重要基础,是农机故障及时维修和农业生产正常开展的必要保障,因此,对农机备件需求量的精准预测至关重要。然而,农机备件的需求量具有非平稳性、非线性、多零值、波动大等特点,使得预测任务变得困难。本文提出了一种基于卷积神经网络的时间序列高效卷积网络(Time series efficient convolution network,TECNet),用于农机备件需求量的预测。该模型首先利用快速傅里叶变换对原始一维序列进行周期性提取,然后根据周期性构建二维时间序列卷积模块进行特征提取,最后将二维特征重塑回一维特征,并通过线性变换得到预测值。利用某农机备件供应商4种不同备件类型的销售数据进行了评估验证,并引入均方根缩放误差作为衡量指标,以统一不同序列间的预测效果。试验结果表明,提出的模型预测效果显著优于其他参考模型,4种不同备件需求量预测的均方根缩放误差分别为0.775、1.349、0.822、0.205,均表现出良好的预测效果。该模型能有效考虑时间序列中的时间依赖关系,具有捕捉时间序列数据中非线性模式的能力,对不同农机备件类型的预测任务均能取得良好的效果,可为预测农机备件需求量提供参考。
