基于双阶段注意力机制和LSTM的鸡舍氨气浓度预测算法  

Ammonia concentration forecasting algorithm in layer house based on two-stage attention mechanism and LSTM

作　　者：郭祥云 连京华[2] 李慧敏[2] 孙凯[2] GUO Xiangyun;LIAN Jinghua;LI Huimin;SUN Kai(School of Information Management,Beijing Information Science and Technology University,Beijing 100192,China;Poultry Institute,Shandong Academy of Agricultural Science,Jinan 250023,China)

机构地区：[1]北京信息科技大学信息管理学院,北京100192 [2]山东省农业科学院家禽研究所,济南250023

出　　处：《中国农业大学学报》2021年第6期187-195,共9页Journal of China Agricultural University

基　　金：山东省家禽疫病诊断与免疫重点实验室开放课题基金(SDPDI201806)。

摘　　要：为寻求准确的鸡舍氨气浓度预测方法,构建基于双阶段注意力机制和长短时记忆神经网络(Long shortterm memory,LSTM)的鸡舍氨气浓度预测模型,将该模型应用于山东省商河县某蛋鸡养殖场,采集二氧化碳(CO_(2))、氧气(O_(2))和氨气(NH_(3))的体积分数,细颗粒物(PM_(2.5))质量浓度,温度,相对湿度时间序列数据对模型进行验证,并与支持向量回归(Support vector regression,SVR)、人工神经网络(Artificial neural network,ANN)模型和无注意力机制的LSTM模型对比研究。结果表明:1)不同时间窗口T下NH_(3)体积分数预测精度不同。T∈{2,3,4,8}时,均方根误差(Root mean square error,RMSE)分别为0.4334,0.3948,0.3799和0.4051μL/L,平均绝对误差(Mean absolute error,MAE)分别为0.2674,0.2629,0.2289和0.2724μL/L;2)基于双阶段注意力机制和LSTM的鸡舍NH_(3)浓度预测模型在RMSE和MAE评价指标框架下优于SVR、ANN和无注意力机制的LSTM模型。基于双阶段注意机制和LSTM的模型能较好地对鸡舍氨气浓度进行预测,可为鸡舍氨气浓度预测及调控提供技术支持。To precisely forecast NH_(3) concentration in layer house,this study proposes a NH_(3) concentration prediction model based on two-stage attention mechanism and long short-term memory(LSTM).In order to validate the model,time series of carbon dioxide(CO_(2)),oxygen(O_(2))and ammonia(NH_(3))volume fraction,particulate matter(PM_(2.5))mass concentration,temperature and relative humidity in layer house in Shanghe County Shandong Province are collected and input to two-stage attention mechanism and LSTM base model.The results show that the model could predict NH_(3) concentration in layer house accurately.Under different window size T∈{2,3,4,8},the root mean square error(RMSE)are 0.4334,0.3948,0.3799 and 0.4051μL/L,respectively;The mean absolute error(MAE)are0.2674,0.2629,0.2289 and 0.2724μL/L,respectively.The model performance in this research is better than the support vector regression(SVR),artificial neural network(ANN)and common LSTM without attention mechanism by RMSE and MAE showing that the model has great potential to be used in NH_(3) concentration prediction and control in layer house.

关 键 词：注意力机制 长短时记忆神经网络 循环神经网络 编码器-解码器 鸡舍 氨气浓度 

分 类 号：X513[环境科学与工程—环境工程]