基于环境参数和深度学习模型的毛竹液流密度预测  

Prediction of moso bamboo sap flow based on environmental parameters and deep learning models

作　　者：彭思乐 贾文丽 邓鹏飞 江朝晖[1] 方明刚 高健 PENG Sile;JIA Wenli;DENG Pengfei;JIANG Zhaohui;FANG Minggang;GAO Jian(College of Information and Artificial Intelligence,Anhui Agricultural University,Hefei 230036,China;Guangde Forestry Development Center,Guangde 242299,China;International Bamboo and Rattan Center,Beijing 100102,China)

机构地区：[1]安徽农业大学信息与人工智能学院,合肥230036 [2]广德市林业发展中心,广德242299 [3]国际竹藤中心,北京100102

出　　处：《农业工程学报》2025年第5期201-208,共8页Transactions of the Chinese Society of Agricultural Engineering

基　　金：“十三五”国家重点研发计划项目(2018YFD0600101);2022年中央财政林业科技推广示范资金项目(hx23350)。

摘　　要：为了揭示环境因素对毛竹液流的影响以及毛竹液流密度与主要环境因子的关系,研究利用竹林环境参数和深度学习方法对毛竹液流进行建模和预测。针对同步采集的土壤湿度、土壤温度、空气湿度、空气温度、二氧化碳浓度、光照强度和毛竹液流密度数据,采用灰色关联分析度量多维环境时序信号与液流之间的相关性,提出一种组合深度学习模型FCN-GRU-TPA进行建模和预测,并考察模型的泛化能力。试验结果显示,光照强度、土壤温度和空气温度对液流密度呈现较强的相关性;用全部6种环境因子可以较好地预测液流密度,用3种相关性强的环境因子预测性能略有下降;按8:2和5:5两种比例分配建模和预测数据,后者在6种参数建模时性能尚可,但是在3种参数建模时性能有所下降;4种试验模式下平均归一化均方根误差NRMSE均小于4.00%,决定系数R^(2)均大于0.90,且不同毛竹植株、不同时段数据的规律相同。研究表明,运用灰色关联分析和FCN-GRU-TPA模型,能够有效建立基于多种环境因子的毛竹液流预测模型,具有较高的预测精度和一定的稳健性。Sap flow is one of the important physiological parameters of moso bamboo,reflecting the plant's utilization of water and its response to the environment.To explore the impact of environmental factors on moso bamboo sap flow and the relationship between sap flow density and key environmental variables,this study employs bamboo forest environmental parameters and deep learning methods to model and predict moso bamboo sap flow.The data used for analysis includes measurements of soil moisture,soil temperature,air humidity,air temperature,CO2 concentration,light intensity,and moso bamboo sap flow density,which were all collected synchronously.Initially,the study applies grey relational analysis to evaluate the correlation between multi-dimensional environmental time-series signals and the sap flow signal.This allows for identifying the degree to which different environmental factors influence the sap flow.Subsequently,the study introduces a novel combined deep learning model called FCN-GRU-TPA for modeling and prediction.The FCN-GRU-TPA model is designed to capture temporal dependencies,select relevant cross-step length information,and perform parallel computation.These advantages theoretically lead to higher prediction accuracy compared to traditional models.Finally,the robustness and generalization capability of the model are is evaluated through four experimental modes,incorporating data from various time periods and multiple moso bamboo plants.The experimental results demonstrate that environmental factors such as air temperature and humidity,soil temperature and humidity,light intensity,and CO2 concentration are correlated with moso bamboo sap flow to varying degrees.The strongest correlation is observed with light intensity,followed by soil temperature and air temperature.When all six environmental factors are used as input variables,the model provides a highly accurate prediction of sap flow density.However,when only three of the most strongly correlated environmental factors are selected as inputs,the performa

关 键 词：毛竹液流 环境因子 建模与预测 灰色关联度 FCN-GRU-TPA模型 

分 类 号：S126[农业科学—农业基础科学]