针对新生儿疼痛表情识别任务中由于有类别标签样本数量不足而导致分类准确率不高的问题,提出了一种基于图的半监督深度学习(Graph-based Semi-supervised Deep Learning,GSDL)方法。首先,使用训练集中少量有类别标签的样本对深度神经网...针对新生儿疼痛表情识别任务中由于有类别标签样本数量不足而导致分类准确率不高的问题,提出了一种基于图的半监督深度学习(Graph-based Semi-supervised Deep Learning,GSDL)方法。首先,使用训练集中少量有类别标签的样本对深度神经网络模型进行初步训练,得到初始模型;然后,利用初始模型提取有类别标签样本与无类别标签样本的特征向量,并使用提取的特征向量构建一个邻接矩阵,进而构建一个图,在构建的图上通过标签传播算法推测出无类别标签样本的伪标签;最后,使用所有样本及其标签对深度神经网络模型进行微调,得到最终的新生儿疼痛表情识别分类模型。在新生儿疼痛表情数据集上的实验结果表明,在使用相同数量的有类别标签样本情况下,文中提出的GSDL模型的分类准确率优于传统的有监督深度学习模型,也高于现有的半监督深度学习模型(Mean-Teachers,MT),验证了GSDL方法在新生儿疼痛表情识别中的有效性。展开更多
Aeolian sandy soil in mining areas exhibits intense evaporation and poor water retention capacity.This study was designed to find a suitable biochar application method to improve soil water infiltration and minimize s...Aeolian sandy soil in mining areas exhibits intense evaporation and poor water retention capacity.This study was designed to find a suitable biochar application method to improve soil water infiltration and minimize soil water evaporation for aeolian sand soil.Using the indoor soil column method,we studied the effects of three application patterns(A(0-20 cm was a mixed sample of mixed-based biochar and soil),B(0-10 cm was a mixed sample of mixed-based biochar and soil and 10-20 cm was soil),and C(0-10 cm was soil and 10-20 cm was a mixed sample of mixed-based biochar and soil)),four application amounts(0%(control,CK),1%,2%,and 4%of mixed-based biochar in dry soil),and two particle sizes(0.05-0.25 mm(S1)and<0.05 mm(S2))of mixed-based biochar on water infiltration and evaporation of aeolian sandy soil.We separately used five infiltration models(the Philip,Kostiakov,Horton,USDA-NRCS(United States Department of Agriculture-Natural Resources Conservation Service),and Kostiakov-Lewis models)to fit cumulative infiltration and time.Compared with CK,the application of mixed-based biochar significantly reduced cumulative soil water infiltration.Under application patterns A,B,and C,the higher the application amount and the finer the particle size were,the lower the migration speed of the wetting front.With the same application amount,cumulative soil water infiltration under application pattern A was the lowest.Taking infiltration for 10 min as an example,the reductions of cumulative soil water infiltration under the treatments of A2%(S2),A4%(S1),A4%(S2),A1%(S1),C2%(S1),and B1%(S1)were higher than 30%,which met the requirements of loess soil hydraulic parameters suitable for plant growth.The five infiltration models well fitted the effects of the treatments of application pattern C and S1 particle size(R2>0.980),but the R2 values of the Horton model exceeded 0.990 for all treatments(except for the treatment B2%(S2)).Compared with CK,all other treatments reduced cumulative soil water infiltration,except for B4%(S2).With the same application amount,cumulative soil water evaporation difference between application patterns A and B was small.Treatments of application pattern C and S1 particle size caused a larger reduction in cumulative soil water evaporation.The reductions in cumulative soil water evaporation under the treatments of C4%(S1),C4%(S2),C2%(S1),and C2%(S2)were over 15.00%.Therefore,applying 2%of mixed-based biochar with S1 particle size to the underlying layer(10-20 cm)could improve soil water infiltration while minimizing soil water evaporation.Moreover,application pattern was the main factor affecting soil water infiltration and evaporation.Further,there were interactions among the three influencing factors in the infiltration process(application amount×particle size with the most important interaction),while there were no interactions among them in the evaporation process.The results of this study could contribute to the rational application of mixed-based biochar in aeolian sandy soil and the resource utilization of urban and agricultural wastes in mining areas.展开更多
随着“双碳目标”的提出和可再生能源渗透率的增加,确保电力系统灵活运行变得愈加重要。针对源、荷、储等多类分布式灵活性资源,构建了电力市场环境下计及风光不确定性和共享储能的虚拟电厂(virtual power plant,VPP)运行优化和双层效...随着“双碳目标”的提出和可再生能源渗透率的增加,确保电力系统灵活运行变得愈加重要。针对源、荷、储等多类分布式灵活性资源,构建了电力市场环境下计及风光不确定性和共享储能的虚拟电厂(virtual power plant,VPP)运行优化和双层效益分配模型。首先,构建了含分布式风光机组、共享储能、柔性负荷的VPP系统结构,并建立了两级电力市场交易机制。其次,利用场景生成与削减的方法处理了风光出力不确定性,在此基础上以VPP运行收益最大化为目标,构建了考虑日前-实时交易的VPP运行优化模型。再次,为保障参与主体收入公平性和合理性,基于双层合作博弈Owen值法,分别建立了VPP参与主体收益分配策略和共享储能投资主体收益分配策略。最后,通过算例分析验证了所提模型的有效性。算例结果表明含共享储能的VPP能够有效降低风光不确定性造成的干扰,提高VPP在日前-实时两级市场中的收益;通过Owen值法对各投资主体进行效益分配,有助于保障系统的公平合理性。展开更多
文摘针对新生儿疼痛表情识别任务中由于有类别标签样本数量不足而导致分类准确率不高的问题,提出了一种基于图的半监督深度学习(Graph-based Semi-supervised Deep Learning,GSDL)方法。首先,使用训练集中少量有类别标签的样本对深度神经网络模型进行初步训练,得到初始模型;然后,利用初始模型提取有类别标签样本与无类别标签样本的特征向量,并使用提取的特征向量构建一个邻接矩阵,进而构建一个图,在构建的图上通过标签传播算法推测出无类别标签样本的伪标签;最后,使用所有样本及其标签对深度神经网络模型进行微调,得到最终的新生儿疼痛表情识别分类模型。在新生儿疼痛表情数据集上的实验结果表明,在使用相同数量的有类别标签样本情况下,文中提出的GSDL模型的分类准确率优于传统的有监督深度学习模型,也高于现有的半监督深度学习模型(Mean-Teachers,MT),验证了GSDL方法在新生儿疼痛表情识别中的有效性。
基金supported by the State Key Laboratory of Water Resource Protection and Utilization in Coal Mining,Open Foundation Ecological Self-Repair Mechanism and Promotion Technology in Shendong Mining Area,China(GJNY-18-73.19)the National Key Research and Development Program of China(2020YFC1806502)。
文摘Aeolian sandy soil in mining areas exhibits intense evaporation and poor water retention capacity.This study was designed to find a suitable biochar application method to improve soil water infiltration and minimize soil water evaporation for aeolian sand soil.Using the indoor soil column method,we studied the effects of three application patterns(A(0-20 cm was a mixed sample of mixed-based biochar and soil),B(0-10 cm was a mixed sample of mixed-based biochar and soil and 10-20 cm was soil),and C(0-10 cm was soil and 10-20 cm was a mixed sample of mixed-based biochar and soil)),four application amounts(0%(control,CK),1%,2%,and 4%of mixed-based biochar in dry soil),and two particle sizes(0.05-0.25 mm(S1)and<0.05 mm(S2))of mixed-based biochar on water infiltration and evaporation of aeolian sandy soil.We separately used five infiltration models(the Philip,Kostiakov,Horton,USDA-NRCS(United States Department of Agriculture-Natural Resources Conservation Service),and Kostiakov-Lewis models)to fit cumulative infiltration and time.Compared with CK,the application of mixed-based biochar significantly reduced cumulative soil water infiltration.Under application patterns A,B,and C,the higher the application amount and the finer the particle size were,the lower the migration speed of the wetting front.With the same application amount,cumulative soil water infiltration under application pattern A was the lowest.Taking infiltration for 10 min as an example,the reductions of cumulative soil water infiltration under the treatments of A2%(S2),A4%(S1),A4%(S2),A1%(S1),C2%(S1),and B1%(S1)were higher than 30%,which met the requirements of loess soil hydraulic parameters suitable for plant growth.The five infiltration models well fitted the effects of the treatments of application pattern C and S1 particle size(R2>0.980),but the R2 values of the Horton model exceeded 0.990 for all treatments(except for the treatment B2%(S2)).Compared with CK,all other treatments reduced cumulative soil water infiltration,except for B4%(S2).With the same application amount,cumulative soil water evaporation difference between application patterns A and B was small.Treatments of application pattern C and S1 particle size caused a larger reduction in cumulative soil water evaporation.The reductions in cumulative soil water evaporation under the treatments of C4%(S1),C4%(S2),C2%(S1),and C2%(S2)were over 15.00%.Therefore,applying 2%of mixed-based biochar with S1 particle size to the underlying layer(10-20 cm)could improve soil water infiltration while minimizing soil water evaporation.Moreover,application pattern was the main factor affecting soil water infiltration and evaporation.Further,there were interactions among the three influencing factors in the infiltration process(application amount×particle size with the most important interaction),while there were no interactions among them in the evaporation process.The results of this study could contribute to the rational application of mixed-based biochar in aeolian sandy soil and the resource utilization of urban and agricultural wastes in mining areas.
文摘随着“双碳目标”的提出和可再生能源渗透率的增加,确保电力系统灵活运行变得愈加重要。针对源、荷、储等多类分布式灵活性资源,构建了电力市场环境下计及风光不确定性和共享储能的虚拟电厂(virtual power plant,VPP)运行优化和双层效益分配模型。首先,构建了含分布式风光机组、共享储能、柔性负荷的VPP系统结构,并建立了两级电力市场交易机制。其次,利用场景生成与削减的方法处理了风光出力不确定性,在此基础上以VPP运行收益最大化为目标,构建了考虑日前-实时交易的VPP运行优化模型。再次,为保障参与主体收入公平性和合理性,基于双层合作博弈Owen值法,分别建立了VPP参与主体收益分配策略和共享储能投资主体收益分配策略。最后,通过算例分析验证了所提模型的有效性。算例结果表明含共享储能的VPP能够有效降低风光不确定性造成的干扰,提高VPP在日前-实时两级市场中的收益;通过Owen值法对各投资主体进行效益分配,有助于保障系统的公平合理性。
