基于PSWE模型的土壤水盐运移与夏玉米生产效益模拟  

Simulation of Soil Salt-water Migration and Summer Maize Productivity Based on PSWE Model

作　　者：张万锋[1] 杨树青[1] 胡睿琦 鄂继芳 ZHANG Wanfeng;YANG Shuqing;HU Ruiqi;E Jifang(College of Water Conservancy and Civil Engineering,Inner Mongolia Agricultural University,Huhhot 010018,China;Australian Artificial Intelligence Institute,Sydney 2007,Australian)

机构地区：[1]内蒙古农业大学水利与土木建筑工程学院,呼和浩特010018 [2]澳大利亚人工智能研究所,悉尼2007

出　　处：《农业机械学报》2022年第6期359-369,共11页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(51669019);国家自然科学基金重点项目(51539005)。

摘　　要：为实现多因素影响下土壤水盐、作物生产效益间的双层递进因果关系模拟,基于深度学习理论及方法将分级长短期记忆网络(HLSTM)与批标准化多层感知机(BMLP)耦合,且将Dropout与Adam优化算法耦合作为面向收敛的改进算法,构建了递进水盐嵌入神经网络(Progressive salt-water embedding neural network,PSWE)模型。评估了PSWE模型的有效性,并开展了多因素协同秸秆深埋下不同灌水量的土壤水盐及夏玉米生产效益的模拟。结果表明,PSWE模型具有多因素整体协同性,有效地模拟了土壤水盐运移规律、夏玉米生产效益及各变量间的内在依存关系。模型平均均方根误差为0.031,平均绝对误差为0.569,平均决定系数为0.987。模拟结果表明,单次灌水60 mm的耕作层(0~40 cm)含水率随时间推移持续降低,单次灌水135 mm的耕作层含水率变幅较大,成熟期二者在秸秆隔层积盐率分别为49.2%和11.2%;单次灌水90 mm和120 mm的耕作层含水率保持在16%~24%之间,成熟期二者在大于40 cm土层含水率保持平稳,秸秆隔层有脱盐趋势,脱盐率为6.1%和5.9%;夏玉米单次理论灌水量为89.3~96.8 mm,耕作层理论含盐量为1.38~1.55 g/kg。综上,多因素协同秸秆深埋下适宜灌溉量可实现抑盐提效的目标,PSWE模型可有效模拟土壤水盐运移和作物生产效益,为深度学习理论及技术在土壤水盐运移模型上的应用提供参考。To realize simulation of the two-layer progressive causal relationship of soil salt-water and crop production benefits under the influence of multiple factors,based on deep learning theory and technology,the progressive salt-water embedding neural network(PSWE)model was constructed.In PSWE model,the time serialized data encoder framed by hierarchical long short-term memory(HLSTM)and decoder framed batch-normalized multi-layer perceptron(BMLP)were coupled,and the coupling between Dropout and Adam algorithm was optimized as an improved algorithm for convergence regression.The validity of PSWE model was evaluated,and the dynamic changes of soil water-salt of different irrigation amounts under multi-factors cooperative straw deep burial were simulated,and the production benefit of summer maize was predicted.The results showed that PSWE model had multivariable overall synergy,self-learning habit and high accuracy.PSWE model could effectively describe the law of soil salt-water migration under straw deep burial in Hetao Irrigation District,the internal dependence relationship between summer maize production benefit and various variables.The root mean square error of the PSWE model was 0.031,the mean absolute error was 0.569,and the determination coefficient was 0.987.Through the model simulation,along with the summer maize growth period,the moisture content of treatment of single irrigation 60 mm was reduced continuously in the tillage layer(0~40 cm),and affected the summer maize for normal growth,while the change of treatment of 135 mm was larger.In the mature stage,they produced salt accumulation in the straw inter-layer,and the salt accumulation rate was 49.2%and 11.2%.The water content in the tillage layer of single irrigation 90 mm and 120 mm were kept between 16%and 24%.At the end of the growth period,the water content in the soil layer over 40 cm was kept stable.The straw inter-layer showed a trend of desalting,and the desalting rate was 6.1%and 5.9%,respectively.It was suggested that the single irrigation amoun

关 键 词：水盐运移 夏玉米 秸秆深埋 长短期记忆网络 批标准化多层感知机 

分 类 号：S274[农业科学—农业水土工程] TV93[农业科学—农业工程]