植物电子病历(EMR)以结构化和非结构化的形式记录了大量关于疾病症状、环境特征以及诊断开方的信息,为病害的智能诊断提供了优质知识来源,但是其样本量少、公开数据集缺乏和多种类型数据并存的特点给相关研究带来困难。根据植物EMR多类...植物电子病历(EMR)以结构化和非结构化的形式记录了大量关于疾病症状、环境特征以及诊断开方的信息,为病害的智能诊断提供了优质知识来源,但是其样本量少、公开数据集缺乏和多种类型数据并存的特点给相关研究带来困难。根据植物EMR多类型数据混合的特点,提出了一种基于BERT-MPL数据融合与注意力机制优化的作物病害诊断模型(BERT-MPL data fusion model based on attention mechanism,BM-Att)。首先采用BERT预训练语言模型抽取电子病历中非结构化部分的文本语义特征;其次通过one-hot编码和多层感知机(MLP)对结构化数据进行编码和向量维度的扩增;最后在特征融合阶段采用注意力机制强调关键特征,利用多层全连接层实现病害诊断。构建了番茄、黄瓜、生菜和西瓜4种作物的15种病害数据集验证模型的效果并进行消融实验,并且对比了CNN、RCNN、AttRNN、FastText、Transformer、BERT和ERNIE等处理文本数据的常见模型,以及BERT-ALEX、BERT-1dCNN、BERT-1dLSTM、BERT-1dAttLSTM、BERT-MLP、ERNIE-ALEX、ERNIE-1dCNN、ERNIE-1dLSTM、ERNIE-1dAttLSTM、ERNIE-MLP等不同数据融合策略。结果表明,BM-Att取得最优结果,在测试集的准确率、精确率、召回率和F1值宏平均值分别达到95.82%、96.38%、95.48%和95.85%,能够实现作物病害的有效诊断。在特征融合阶段添加注意力机制的策略将模型F1值宏平均值提高1.47个百分点,显著提升了模型对生菜霜霉病、西瓜线虫等小样本病害的分类效果。该研究可为电子病历数据挖掘及实现智能辅助病害诊断提供参考。展开更多
The addition of effective additives can effectively improve the pyrolysis performance of oil sludge(OS)and have a great potential to reduce pyrolysis costs.In the present study,co-pyrolysis performance of OS with diff...The addition of effective additives can effectively improve the pyrolysis performance of oil sludge(OS)and have a great potential to reduce pyrolysis costs.In the present study,co-pyrolysis performance of OS with different proportions of additives at a heating rate of 10°C/min was conducted in a thermal analyzer.Walnut shell,Fe_(2)O_(3),K_(2)CO_(3),PVC and the pyrolysis char produced from OS at the final pyrolysis temperature of 700℃were selected as the additives.TG results showed that the OS pyrolysis with and without additives can be divided into five reaction stages,which include volatilization of free water,the escape of light components,the cleavage of heavy components,carbon decomposition and inorganic minerals decomposition.The addition of additives decreased the maximum weight loss rate when the blending ratio was 5 wt%during OS pyrolysis.Kinetic analysis revealed that the overall activation energy of pyrolysis reaction was lower during pyrolysis of OS with the addition of walnut shells and pyrolysis char.The activation energy of three main reaction stages all decreased during co-pyrolysis of OS with K_(2)CO_(3)and PVC.展开更多
文摘植物电子病历(EMR)以结构化和非结构化的形式记录了大量关于疾病症状、环境特征以及诊断开方的信息,为病害的智能诊断提供了优质知识来源,但是其样本量少、公开数据集缺乏和多种类型数据并存的特点给相关研究带来困难。根据植物EMR多类型数据混合的特点,提出了一种基于BERT-MPL数据融合与注意力机制优化的作物病害诊断模型(BERT-MPL data fusion model based on attention mechanism,BM-Att)。首先采用BERT预训练语言模型抽取电子病历中非结构化部分的文本语义特征;其次通过one-hot编码和多层感知机(MLP)对结构化数据进行编码和向量维度的扩增;最后在特征融合阶段采用注意力机制强调关键特征,利用多层全连接层实现病害诊断。构建了番茄、黄瓜、生菜和西瓜4种作物的15种病害数据集验证模型的效果并进行消融实验,并且对比了CNN、RCNN、AttRNN、FastText、Transformer、BERT和ERNIE等处理文本数据的常见模型,以及BERT-ALEX、BERT-1dCNN、BERT-1dLSTM、BERT-1dAttLSTM、BERT-MLP、ERNIE-ALEX、ERNIE-1dCNN、ERNIE-1dLSTM、ERNIE-1dAttLSTM、ERNIE-MLP等不同数据融合策略。结果表明,BM-Att取得最优结果,在测试集的准确率、精确率、召回率和F1值宏平均值分别达到95.82%、96.38%、95.48%和95.85%,能够实现作物病害的有效诊断。在特征融合阶段添加注意力机制的策略将模型F1值宏平均值提高1.47个百分点,显著提升了模型对生菜霜霉病、西瓜线虫等小样本病害的分类效果。该研究可为电子病历数据挖掘及实现智能辅助病害诊断提供参考。
基金supported by Natural Science Foundation of Shandong Province(ZR2020QE199)State Key Laboratory of Pollution Control and Resource Utilization research(No.PCRRF19023)Key Research and Development Program of Liaoning Province(No.2020JH2/10300099)。
文摘The addition of effective additives can effectively improve the pyrolysis performance of oil sludge(OS)and have a great potential to reduce pyrolysis costs.In the present study,co-pyrolysis performance of OS with different proportions of additives at a heating rate of 10°C/min was conducted in a thermal analyzer.Walnut shell,Fe_(2)O_(3),K_(2)CO_(3),PVC and the pyrolysis char produced from OS at the final pyrolysis temperature of 700℃were selected as the additives.TG results showed that the OS pyrolysis with and without additives can be divided into five reaction stages,which include volatilization of free water,the escape of light components,the cleavage of heavy components,carbon decomposition and inorganic minerals decomposition.The addition of additives decreased the maximum weight loss rate when the blending ratio was 5 wt%during OS pyrolysis.Kinetic analysis revealed that the overall activation energy of pyrolysis reaction was lower during pyrolysis of OS with the addition of walnut shells and pyrolysis char.The activation energy of three main reaction stages all decreased during co-pyrolysis of OS with K_(2)CO_(3)and PVC.
