摘要
土壤作为农作物生长的主要营养来源,氮是植物生长的重要元素,有效评价土壤氮素含量可以促进配方施肥的发展。提出主成分分析、注意力机制和长短时记忆神经网络相结合的模型(PCA-Attention-LSTM)来监测土壤的氮素含量。采用PCA(主成分分析)对数据进行处理,提取影响土壤氮含量的关键影响因子,降低模型向量输入的维数,利用注意机制突出预测中的关键输入特征。在Keras深度学习框架的基础上搭建PCA-Attention-LSTM的网络模型,实现对未来2 h土壤氮含量的精监测。最后,以黑龙江省依安甜菜养植基地的数据对土壤氮含量进行训练和验证。结果表明,与RNN等其它网络模型相比,该模型的效果更好,基于PCA-Attenlion-LSTM网络模型的平均绝对误差,均方根误差和平均绝对百分误差分别为0.119、0.020、0.156。该模型预测精度高,泛化能力强,可以应用于土壤氮含量的监测。
Soil is the main nutrient source for crop growth,and nitrogen is an important element for plant growth.Effective evaluation of soil nitrogen content can promote the development of formula fertilization.A model(PCA-Attention-LSTM)combining principal component analysis(PCA-Attention-LSTM)with neural network of long-and long-term memory was proposed to monitor soil n content.PCA(principal component analysis)was used to process the data,extract the key influencing factors of soil nitrogen content,reduce the dimension of model vector input,and highlight the key input characteristics in the prediction by using the attention mechanism.B ased on the Keras deep learning framework,a PCA-Attention-LSTM network model was established to achieve the precise monitoring of soil nitrogen content in the next 2 hours.Finally,the soil nitrogen content was trained and verified with the data of Yi’an Beet cultivation base in Heilongjiang Province.The results show that compared with other network models such as RNN,this model has a better effect.Based on the average absolute error of PCA-Attention-LSTM network model,the root mean square error and average absolute percentage error are 0.119,0.020 and 0.156 respectively.The model has high prediction accuracy and strong generalization ability,and can be used to monitor soil nitrogen content.
作者
刘慧敏
甄佳奇
刘勇
解洪富
许文超
Liu Huimin;Zhen Jiaqi;Liu Yong;Xie Hongfu;Xu Wenchao(Heilongjiang University,Haerbin,150000,China;Shandong University of Science and Technology,Qingdao,266000,China;Heilongjiang East Water-saving Equipment Co.,LTD.,Haerbin,150000,China)
出处
《中国农机化学报》
北大核心
2020年第9期190-197,共8页
Journal of Chinese Agricultural Mechanization
基金
国家自然科学基金(61501176)
黑龙江省自然科学基金项目(F2018025)。
关键词
土壤氮含量监测
主成分分析
Attention机制
LSTM神经网络
monitoring of nitrogen in soil
principal component analysis(PCA)
attention mechanism
long short-term memory neural networks
