Short-term travel flow prediction has been the core of the intelligent transport systems(ITS). An advanced method based on fuzzy C-means(FCM) and extreme learning machine(ELM) has been discussed by analyzing predictio...Short-term travel flow prediction has been the core of the intelligent transport systems(ITS). An advanced method based on fuzzy C-means(FCM) and extreme learning machine(ELM) has been discussed by analyzing prediction model. First, this model takes advantages of ability to adapt to nonlinear systems and the fast speed of ELM algorithm. Second, with FCM-clustering function, this novel model can get the clusters and the membership in the same cluster, which means that the associated observation points have been chosen. Therefore, the spatial relations can be used by giving the weight to every observation points when the model trains and tests the ELM. Third, by analyzing the actual data in Haining City in 2016, the feasibility and advantages of FCM-ELM prediction model have been shown when compared with other prediction algorithms.展开更多
To improve prediction accuracy of strip thickness in hot rolling, a kind of Dempster/Shafer(D_S) information reconstitution prediction method(DSIRPM) was presented. DSIRPM basically consisted of three steps to impleme...To improve prediction accuracy of strip thickness in hot rolling, a kind of Dempster/Shafer(D_S) information reconstitution prediction method(DSIRPM) was presented. DSIRPM basically consisted of three steps to implement the prediction of strip thickness. Firstly, iba Analyzer was employed to analyze the periodicity of hot rolling and find three sensitive parameters to strip thickness, which were used to undertake polynomial curve fitting prediction based on least square respectively, and preliminary prediction results were obtained. Then, D_S evidence theory was used to reconstruct the prediction results under different parameters, in which basic probability assignment(BPA) was the key and the proposed contribution rate calculated using grey relational degree was regarded as BPA, which realizes BPA selection objectively. Finally, from this distribution, future strip thickness trend was inferred. Experimental results clearly show the improved prediction accuracy and stability compared with other prediction models, such as GM(1,1) and the weighted average prediction model.展开更多
传统的知识图谱表示学习模型主要聚焦于三元组内部的结构信息,而未能充分利用外部语义增强嵌入表征能力,如没有充分考虑实体间的多步关系路径信息以及不同路径的重要程度,且没有利用实体描述信息增强上下文感知能力。为提升知识图谱的...传统的知识图谱表示学习模型主要聚焦于三元组内部的结构信息,而未能充分利用外部语义增强嵌入表征能力,如没有充分考虑实体间的多步关系路径信息以及不同路径的重要程度,且没有利用实体描述信息增强上下文感知能力。为提升知识图谱的应用效果,提出融合多步关系路径和实体描述信息的知识图谱表示学习(MPDRL)模型。首先,对两实体间的路径信息进行编码,并使用自注意力机制计算路径权重,从而获得关系路径信息的表示;其次,使用BERT(Bidirectional Encoder Representations from Transformers)模型对实体描述信息进行编码,并利用双向注意力机制计算实体描述信息嵌入与三元组关系嵌入之间的注意力权重,从而增强实体的语义信息;最后,将关系路径信息嵌入、实体描述信息嵌入和三元组结构嵌入融合起来进行训练。为评估模型性能,在公开数据集上针对所提模型和基准模型进行链接预测和三元组分类的实验。结果表明:在链接预测任务中,与融合关系路径与实体描述信息的知识图谱表示学习方法(PDRL)、多跳关系路径模型Att-ConvBiLSTM以及融合实体描述与关系路径信息的知识图谱嵌入模型TPKGE相比,所提模型在FB15k-237数据集上的Hit@10指标分别提高了5.7、2.9、2.5个百分点;在三元组分类任务上,所提模型在FB15k-237和WN18RR数据集上的准确率较最优基准模型PDRL分别提升了2.81和0.90个百分点。展开更多
目的:分析肺癌病人治疗期输液港发生医用粘胶相关皮肤损伤(medical adhesive related skin injury,MARSI)的危险因素,并建立风险预测模型,以期为临床护理干预提供参考。方法:回顾性收集2023年1月—2024年4月在某三级甲等综合医院呼吸与...目的:分析肺癌病人治疗期输液港发生医用粘胶相关皮肤损伤(medical adhesive related skin injury,MARSI)的危险因素,并建立风险预测模型,以期为临床护理干预提供参考。方法:回顾性收集2023年1月—2024年4月在某三级甲等综合医院呼吸与危重症医学科使用胸壁输液港的650例病人为调查对象,运用Logistic回归模型、决策树分类回归树(CART)模型和随机森林模型分别建立肺癌病人治疗期输液港医用粘胶相关皮肤损伤风险预测模型,通过比较3种模型的准确率、灵敏度、特异度、阳性预测值、阴性预测值、Kappa系数和受试者工作特征(ROC)曲线下面积(AUC)评价其性能。结果:Logistic回归模型、决策树CART模型和随机森林模型的准确率分别为84%、86%、86%,特异度为97%、98%、97%,灵敏度为54%、59%、61%,阳性预测值为54%、59%、61%,阴性预测值为97%、98%、97%,Kappa值为0.57,0.63,0.64,AUC为0.83,0.87,0.86。Logistic回归模型、决策树CART模型、随机森林的AUC比较差异均有统计学意义(P<0.05)。皮肤毒性为3种模型的共同预测因子。结论:决策树CART模型和随机森林模型相比Logistic回归模型在构建肺癌病人治疗期输液港医用粘胶相关皮肤损伤风险预测模型中具有更好的性能,可为临床护士预测肺癌病人输液港医用粘胶相关皮肤损伤发生风险提供参考。展开更多
针对现有的归纳关系预测方法中大多只考虑实体之间的关系路径,未考虑关系上下文包含的头尾实体的性质,提出一种融合关系路径和关系上下文的归纳关系预测(inductive relation prediction fusing relation path and context,IRP-RPC)模型...针对现有的归纳关系预测方法中大多只考虑实体之间的关系路径,未考虑关系上下文包含的头尾实体的性质,提出一种融合关系路径和关系上下文的归纳关系预测(inductive relation prediction fusing relation path and context,IRP-RPC)模型,将关系上下文作为关系路径的补充来进行归纳关系预测.该方法仅依赖于关系语义信息,因此能够自然地推广到完全归纳的设置.先使用随机行走寻径策略获取关系路径和关系上下文,再设计并实现一个层次化的融合了门控网络的Transformer架构来统一聚合关系路径和关系上下文,以捕获实体之间的联系和实体的内在属性,并采用这些组件的自适应加权组合来做出最终预测.在公开的FB15K-237和NELL-995的8个版本归纳数据集上进行实验,与9个基线模型相比,IRP-RPC模型在精确率-召回率曲线下的面积(area under the precision-recall curve,AUC-PR)和hits@10指标上均取得了优异的性能,验证了其有效性和可推广性.研究表明,IRP-RPC模型通过融合关系路径和关系上下文,能够更全面地建模实体间的语义联系与结构信息,在解决传统归纳关系预测方法中路径信息与上下文信息利用不足的问题上具有显著优势.展开更多
基金Project(2016YFB0100906)supported by the National Key R&D Program in ChinaProject(2014BAG03B01)supported by the National Science and Technology Support plan Project China+1 种基金Project(61673232)supported by the National Natural Science Foundation of ChinaProjects(Dl S11090028000,D171100006417003)supported by Beijing Municipal Science and Technology Program,China
文摘Short-term travel flow prediction has been the core of the intelligent transport systems(ITS). An advanced method based on fuzzy C-means(FCM) and extreme learning machine(ELM) has been discussed by analyzing prediction model. First, this model takes advantages of ability to adapt to nonlinear systems and the fast speed of ELM algorithm. Second, with FCM-clustering function, this novel model can get the clusters and the membership in the same cluster, which means that the associated observation points have been chosen. Therefore, the spatial relations can be used by giving the weight to every observation points when the model trains and tests the ELM. Third, by analyzing the actual data in Haining City in 2016, the feasibility and advantages of FCM-ELM prediction model have been shown when compared with other prediction algorithms.
基金Projects(61174115,51104044)supported by the National Natural Science Foundation of ChinaProject(L2010153)supported by Scientific Research Project of Liaoning Provincial Education Department,China
文摘To improve prediction accuracy of strip thickness in hot rolling, a kind of Dempster/Shafer(D_S) information reconstitution prediction method(DSIRPM) was presented. DSIRPM basically consisted of three steps to implement the prediction of strip thickness. Firstly, iba Analyzer was employed to analyze the periodicity of hot rolling and find three sensitive parameters to strip thickness, which were used to undertake polynomial curve fitting prediction based on least square respectively, and preliminary prediction results were obtained. Then, D_S evidence theory was used to reconstruct the prediction results under different parameters, in which basic probability assignment(BPA) was the key and the proposed contribution rate calculated using grey relational degree was regarded as BPA, which realizes BPA selection objectively. Finally, from this distribution, future strip thickness trend was inferred. Experimental results clearly show the improved prediction accuracy and stability compared with other prediction models, such as GM(1,1) and the weighted average prediction model.
文摘传统的知识图谱表示学习模型主要聚焦于三元组内部的结构信息,而未能充分利用外部语义增强嵌入表征能力,如没有充分考虑实体间的多步关系路径信息以及不同路径的重要程度,且没有利用实体描述信息增强上下文感知能力。为提升知识图谱的应用效果,提出融合多步关系路径和实体描述信息的知识图谱表示学习(MPDRL)模型。首先,对两实体间的路径信息进行编码,并使用自注意力机制计算路径权重,从而获得关系路径信息的表示;其次,使用BERT(Bidirectional Encoder Representations from Transformers)模型对实体描述信息进行编码,并利用双向注意力机制计算实体描述信息嵌入与三元组关系嵌入之间的注意力权重,从而增强实体的语义信息;最后,将关系路径信息嵌入、实体描述信息嵌入和三元组结构嵌入融合起来进行训练。为评估模型性能,在公开数据集上针对所提模型和基准模型进行链接预测和三元组分类的实验。结果表明:在链接预测任务中,与融合关系路径与实体描述信息的知识图谱表示学习方法(PDRL)、多跳关系路径模型Att-ConvBiLSTM以及融合实体描述与关系路径信息的知识图谱嵌入模型TPKGE相比,所提模型在FB15k-237数据集上的Hit@10指标分别提高了5.7、2.9、2.5个百分点;在三元组分类任务上,所提模型在FB15k-237和WN18RR数据集上的准确率较最优基准模型PDRL分别提升了2.81和0.90个百分点。
文摘目的:分析肺癌病人治疗期输液港发生医用粘胶相关皮肤损伤(medical adhesive related skin injury,MARSI)的危险因素,并建立风险预测模型,以期为临床护理干预提供参考。方法:回顾性收集2023年1月—2024年4月在某三级甲等综合医院呼吸与危重症医学科使用胸壁输液港的650例病人为调查对象,运用Logistic回归模型、决策树分类回归树(CART)模型和随机森林模型分别建立肺癌病人治疗期输液港医用粘胶相关皮肤损伤风险预测模型,通过比较3种模型的准确率、灵敏度、特异度、阳性预测值、阴性预测值、Kappa系数和受试者工作特征(ROC)曲线下面积(AUC)评价其性能。结果:Logistic回归模型、决策树CART模型和随机森林模型的准确率分别为84%、86%、86%,特异度为97%、98%、97%,灵敏度为54%、59%、61%,阳性预测值为54%、59%、61%,阴性预测值为97%、98%、97%,Kappa值为0.57,0.63,0.64,AUC为0.83,0.87,0.86。Logistic回归模型、决策树CART模型、随机森林的AUC比较差异均有统计学意义(P<0.05)。皮肤毒性为3种模型的共同预测因子。结论:决策树CART模型和随机森林模型相比Logistic回归模型在构建肺癌病人治疗期输液港医用粘胶相关皮肤损伤风险预测模型中具有更好的性能,可为临床护士预测肺癌病人输液港医用粘胶相关皮肤损伤发生风险提供参考。
文摘针对现有的归纳关系预测方法中大多只考虑实体之间的关系路径,未考虑关系上下文包含的头尾实体的性质,提出一种融合关系路径和关系上下文的归纳关系预测(inductive relation prediction fusing relation path and context,IRP-RPC)模型,将关系上下文作为关系路径的补充来进行归纳关系预测.该方法仅依赖于关系语义信息,因此能够自然地推广到完全归纳的设置.先使用随机行走寻径策略获取关系路径和关系上下文,再设计并实现一个层次化的融合了门控网络的Transformer架构来统一聚合关系路径和关系上下文,以捕获实体之间的联系和实体的内在属性,并采用这些组件的自适应加权组合来做出最终预测.在公开的FB15K-237和NELL-995的8个版本归纳数据集上进行实验,与9个基线模型相比,IRP-RPC模型在精确率-召回率曲线下的面积(area under the precision-recall curve,AUC-PR)和hits@10指标上均取得了优异的性能,验证了其有效性和可推广性.研究表明,IRP-RPC模型通过融合关系路径和关系上下文,能够更全面地建模实体间的语义联系与结构信息,在解决传统归纳关系预测方法中路径信息与上下文信息利用不足的问题上具有显著优势.
