Comparison of artificial neural networks and prediction models for reference evapotranspiration estimation in a semi-arid region

Estimation of reference evapotranspiration (ET_o) is essential for determination of crop water requirements. In this research, Penman–FAO (P-FAO) and Penman–Monteith (PM) equations were calibrated and validated by lysimeter-measured ET_o with six and four weather parameters. Furthermore, two input structures (six and four weather parameters) to artificial neural networks (ANNs) were investigated. Results showed that the accuracy of the PM equation is greater than that of the P-FAO equation. An empirical equation was developed to estimate daily ET_o using mean daily temperature and relative humidity, and sunshine hours. The accuracy of the equation to estimate daily ET_o using smooth weather data is greater than that of an equation using original data. Furthermore, ANNs were able to estimate ET_o properly. The accuracy of ANNs with six inputs is higher than that obtained using the P-FAO equation and is similar to that determined using the PM equation. A decrease in number of inputs to ANNs generally decreased the accuracy of estimation, however, ANNs were able to estimate ET_o properly when wind speed and solar radiation were unavailable. Furthermore, the accuracy of ANNs, with four input parameters is greater than that obtained using the PM equation and is similar to that obtained with P–FAO and the developed empirical equations.     

Comparison between rice grain yield predictions using artificial neural networks and a very simple model under different levels of water and nitrogen application

It is important to model water and nitrogen requirements for rice yield in order to improve production. In this study, an artificial neural network (ANN) was used to predict rice grain yield under different water and nitrogen application. Grain yield was predicted based on five variables: nitrogen application rate, seasonal amount of applied irrigation water, plant population, and mean daily solar input before and after flowering. Furthermore, the ANN method was compared with a very simple model (VSM) for prediction of rice grain yield. Two approaches were considered for ANNs. In the first (local partitioning), rice grain yield and variable data from the south of Iran were used for training, and the network was then tested using independent data from the north of Iran. In another approach, the data for both experiments were mixed and randomized dividing was applied (stochastic partitioning). The results showed that stochastic partitioning networks are more accurate than local partitioning networks. Comparison between ANN and VSM results showed that using ANNs gives a more accurate prediction of grain yield. Therefore, ANNs with stochastic partitioning of data is an accurate method to predict rice grain yield using readily available inputs.     

基于多值神经元复数神经网络的土壤墒情预测

为指导节水灌溉策略的制定,利用基于多值神经元的复数神经网络(multilayer neural network with multi-valued neurons,MLMVN)方法,建立了土壤墒情多步预测模型。首先,利用均值法替换样本中的异常值并对缺失值进行补充,并由数据分析知土壤墒情数据为非平稳的非线性时间序列。然后,根据土壤墒情与环境因素(降雨量、气温和风速)的相关性分析结果选择降雨量为关键环境因素。最后将土壤墒情、降雨量及目标土壤墒情复数化,作为网络输入和期望输出建立MLMVN预测模型。结果表明,网络结构为240-15-1200-1时单步预测精度为0.883,采用循环预测法进行步长为72的多步预测,平均预测精度为0.853,比实数域误差反向传播神经网络BP提高了9.1%。研究表明,MLMVN模型多步预测误差累计小,预测结果可作为该地区节水灌溉策略制定的理论依据。     

基于灰色关联度与BP神经网络模型的日参考作物腾发量预测

参考作物腾发量是制定灌溉用水计划、水量分配计划最基本、最重要的内容之一,其精确预测可以提高灌溉预报的精度。采用灰色系统理论中的关联分析方法,对影响作物腾发量的各个气象因素进行关联度分析,挑选出影响作物腾发量的主要气象因子,并以这些主要气象因子为输入向量,以参考作物腾发量为输出向量,建立作物腾发量与主要气象因子之间的BP神经网络预测模型。通过实例证明,该方法简单可行,预测精度比较高,能够满足实际生产需要。     

基于BP神经网络的冬小麦耗水预测

该文根据中国科学院禹城农业试验站2003－2006年冬小麦季的气象资料和大型称重式蒸渗仪观测资料,把实测作物系数作为作物因子指标,建立了以日最高温度、日净辐射、实测表层60 cm土壤含水率、日序数和作物系数为输入因子,蒸渗仪实测蒸散量为输出因子的BP神经网络预测模型,神经网络拓扑结构为5-9-1,训练函数为Trainbr。检验结果表明冬小麦耗水量模型预测平均相对误差为13.1%,预测值和实测值的均方根误差为0.88 mm,模型预测Nash-Sutcliffe效率指数为0.865,预测效果较好,可满足生产需要。     

参考作物腾发量主成分神经网络预测模型

为解决采用神经网络模型预测参考作物蒸发蒸腾最Eto研究中预测能力不足的问题,将气象因子包括最高、最低和日平均温度、日照时数、气压、水汽压、相对湿度和风速进行主成分分析,提取主成分,建立了基于主成分的三层BP神经网络模型.选取崇川水利科学试验站2001年到2004年的旬气象资料,采用Matlab神经网络工具箱进行模型训练与预测,并以传统BP网络模犁作为对照.结果表明,主成分网络模型能够很好地反映诸多影响因子与Eto之间的关系,尤其对训练样本以外的验证样本,主成分网络模型具有显著优于传统BP网络模型的识别能力,取得更为可靠的预测结果.     

基于神经网络的水产品价格预测模型研究

以水产品中鲫鱼为例,选择时间、地理环境和经济条件因素作为输入层变量,价格作为输出单元,输入样本进行训练和仿真,对训练好的网络输入预测样本,将预测结果与市场实际价格进行比较,其相对误差均小于1%。结果证明,所构建的水产品价格预测模型具有良好的精确性和准确性,将神经网络应用于水产品价格预测是可行的。     

基于卫星光谱尺度反射率的冬小麦生物量估算

为探索基于光学卫星遥感数据的冬小麦地上生物量估算方法,本研究通过3年田间试验,获取冬小麦4个关键生育期（拔节期、抽穗期、开花期和灌浆期）和3种施氮水平下的地上生物量以及对应的近地冠层高光谱反射率数据。通过将高光谱数据重采样为具有红边波段的RapidEye、Sentinel-2和WorldView-2卫星波段反射率数据,构建任意两波段归一化植被指数。同时,将卫星波段反射率数据与6种机器学习和深度学习算法相结合,构建冬小麦生物量估算模型。研究结果表明：任意两波段构建的最佳植被指数在冬小麦开花期对生物量的敏感性最强（决定系数R2为0.50～0.56）。在不同施氮水平条件下,高施氮水平增强了植被指数对生物量的敏感性。Sentinel-2波段数据所构建的植被指数优于其他两颗卫星波段数据。对6种机器学习和深度学习算法,总的来说,基于深度神经网络（Deep Neural Networks,DNN）算法所构建的模型要优于其他算法。在单一生育期中,在拔节期（R2为0.69～0.78,归一化均方根误差为26%～31%）和开花期（R2为0.69～0.70,归一化均方根误差为24%～25%）的估算精度最高。Sentinel-2波段数据与DNN算法结合的估算精度最高,在全生育期中预测精度R2为0.70。施氮水平的提高同样增强了DNN模型的估算精度,3颗卫星波段数据在300 kg/hm2施氮条件下的预测精度R2都在0.71以上,均方根误差小于219 g/m2。研究结果揭示了光学卫星遥感数据在不同生育期和施氮条件下估算冬小麦生物量的潜力。     

Yield prediction in spring barley from spectral reflectance and weather data using machine learning

Accurate preharvest yield estimation is an important issue for agricultural planning purposes and precision farming. Machine learning (ML) based on readily obtained information on the cropping system, typically including spectral reflectance measurements, is an essential approach for achieving practical solutions. We tested in a 9-year soil compaction experiment the accuracy of ML-based yield predictions made up to 2 months before harvest from a Ratio Vegetation Index (RVI) and recordings of precipitation and reference evapotranspiration. The applied data set comprises 224 combinations of plots and years with measured grain yields in the range of 4.22–9.34 Mg/ha. The best ML model [i.e., with the smallest mean absolute error (MAE)] was selected automatically by the AutoML interface included in the R program package H₂O. Its cross-validated predictions made on June 30 more than 1 month before harvest showed an MAE of 0.38 Mg/ha when trained on all data from all years except the one under consideration. MAE increased to about 0.68 Mg/ha when determined 3 weeks earlier on June 10. MAE values in the range of 0.32–0.42 Mg/ha were obtained for predictions made on June 30 when based on data from at least six consecutive years; however, MAE showed no generally decreasing trend with the number of years. Yield estimations were robust towards a considerable soil variation observed within the experimental area due in part to the experimental treatments. The results show a potential of making yield predictions in barley 1–2 months before harvest, which, however, is not sufficiently early to support decisions on top-dress N fertilization.     

温室封闭式栽培自适应灌溉方法

为了改进和提高温室封闭式栽培精细灌溉控制方法,针对利用Penman-Monteith(P-M)公式和传感器数据信息相结合进行灌溉控制中因为涉及参数较多而使用不便、需要近似计算导致建立的作物蒸腾模型精度不够等问题,该文根据封闭式栽培可以回收并循环利用多余灌溉水的特点,利用灌溉量与排出量的差值和温室小气候环境数据建立相对精确的作物蒸腾量计算模型,并在此基础上利用人工神经网络算法实现了温室封闭式栽培自适应灌溉控制,结果表明,在10 d内灌溉用水量为实际蒸腾量的97.8%,基本实现了按照作物需水量进行灌溉。研究对于实现按照作物蒸腾量进行准确的水分供给、节约灌溉用水量、提高水分利用效率具有一定的现实意义。     

黄土区不同土地经营方式径流量神经网络模拟

基于黄土坡面降雨—径流关系的复杂性和非线性,引用3层前馈型BP网络模型,对不同土地经营方式(草灌地、刈割地、翻耕地)径流量进行模拟,以植被盖度、降雨强度、坡度、土壤前期含水率和土壤容重5个因子作为输入层变量,次降雨下径流量作为输出层变量。利用野外人工模拟降雨试验所得到的不同降雨强度下各类土地经营径流小区的径流量实测资料,对网络进行模拟训练并预测,径流量平均误差不超过10%,且径流量较大的翻耕地训练精度及预测结果较草灌地、刈割准确性更高些。与传统回归统计方法进行了误差比较,结果表明,该模型能更好地预测次降雨的径流量。     

基于GEP和地理位置信息的湘鄂地区月参考作物腾发量模拟计算

参考作物腾发量(ET0)是计算植被蒸散发的关键因子,准确估算ET0对水资源管理、灌溉制度设计等具有重要意义。本研究利用湘鄂地区46个气象站点1955—2005年的逐月气象数据,包括月最高气温、最低气温、平均风速、日照时数以及相对湿度,用FAO-56 Penman-Monteith法计算各站的逐月ET0值。然后结合基因表达式编程(GEP)算法挖掘公式的能力,以各站点的地理位置信息(纬度、经度、海拔)及月序数为输入,以多年逐月平均ET0值为输出,建立基于地理位置信息的月ET0模型,并与传统ET0模型的计算结果进行比较。结果表明,所建立的模型具有足够的精度,校正、检验阶段的决定系数(R2)和均方根误差(RMSE)分别为0.934、0.951和10.050 mm、8.628 mm;与Hargreaves和Priestley-Taylor法相比,基于地理位置信息建立的GEP模型的结果均方根误差最小,变化范围为8.628~9.967 mm。本研究所建立的月ET0模型具有明确的表达式,简单易用,在湘鄂地区仅利用地理位置信息计算逐月ET0是可行的,可以利用该模型进行月尺度的灌溉制度设计和植被蒸散发的估算。     

利用人工神经网络以及相关地形属性绘制数字土壤地图

Detailed soil surveys involve costly and time-consuming work and require expert knowledge. Since soil surveys provide information to meet a wide range of needs, new methods are necessary to map soils quickly and accurately. In this study, multilayer perceptron artificial neural networks (ANNs) were developed to map soil units using digital elevation model (DEM) attributes. Several optimal ANNs were produced based on a number of input data and hidden units. The approach used test and validation areas to calculate the accuracy of interpolated and extrapolated data. The results showed that the system and level of soil classification employed had a direct effect on the accuracy of the results. At the lowest level, smaller errors were observed with the World Reference Base (WRB) classification criteria than the Soil Taxonomy (ST) system, but more soil classes could be predicted when using ST (7 soils in the case of ST vs. 5 with WRB). Training errors were below 11% for all the ANN models applied, while the test error (interpolation error) and validation error (extrapolation error) were as high as 50% and 70%, respectively. As expected, soil prediction using a higher level of classification presented a better overall level of accuracy. To obtain better predictions, in addition to DEM attributes, data related to landforms and/or lithology as soil-forming factors, should be used as ANN input data.     

新安江模型和人工神经网络的耦合应用

提出了一种集成人工神经网络的概念性水文模型,该模型是一种半分布式概念性水文模型,考虑了降雨的空间变异性,流域特征的不均匀性等因素对径流过程的影响。采用遗传算法进行概念性模型参数优选,同时考虑雨量站的空间分布,利用GIS和DEM数据进行流域单元划分;对于每个子流域,考虑模型输入参数和降雨资料的空间分布特性,进行产流计算;在径流演算过程中,利用人工神经网络的非线性映射方法代替传统概念模型中线性叠加方法计算整个流域的出口流量过程。以半湿润的淮河上游支流的大坡岭流域为例,对模型的可行性进行验证,并与单一的新安江模型的结果进行了比较。验证结果表明,集成人工神经网络技术和新安江模型的耦合模型有较好的模拟和预报结果。     

空气悬架系统动态载荷的识别

针对空气悬架系统主动控制中神经辨识器的离线训练问题,利用BP神经网络实现从空气悬架系统非簧载质量振动加速度空间到其动态载荷空间的映射。建立带有空气悬架系统的1/4工程车辆动态模型,通过仿真得出了工程车辆空气悬架系统的非簧载质量振动加速度和动态载荷数据,以空气悬架系统的非簧载质量振动加速度数据作为神经网络的输入,动态载荷数据作为神经网络的输出,训练BP神经网络,并对训练好的BP神经网络进行泛化能力的测试,路面输入采用幅值为0.01 m,频率为1 rad/s正弦波时,识别误差率在30%以内的点占总数的82.95%;以幅值为0.02 m,频率为2 rad/s的正弦波作为系统的路面输入,识别误差率在30%以内的点占总数的77.94%。结果表明BP神经网络能够对不同的路面输入具有较好的适应性。     

基于竞争学习网络的田间籽棉图像分割

为了正确识别田间籽棉,将籽棉和铃壳、绿叶、根茎、土地等自然背景视为二个类别,基于竞争学习网络进行了图像分割。从多幅典型的籽棉图像中选取10000个像素作为训练样本,并为它们贴上类别标签,在HSI、Lab、Ohta、RGB颜色空间下对训练样本的颜色特征及其组合进行K-均值聚类,选取了误分率普遍较低的RGB颜色空间,其B值的误分率尤其低。在RGB颜色空间下,用训练样本的R、G、B组合或B值一次性地训练了竞争学习网络,将图像的全部像素输入网络进行测试,同时与K-均值聚类比较,形态学滤波去噪后的结果表明,基于B值的竞争学习网络较优,用907幅籽棉图像对其进行仿真的精度达92.94%。该方法结合了有监督的学习算法,避免了传统K-均值聚类的反复迭代和过拟合现象,提高了图像分割的效率和精度。     

谷物识别中对神经网络的优化

主要讨论了在谷物纹理识别中对神经网络的优化。通过比较优化神经网络和非优化神经网络的输入、输出之间的非线性联系,可知优化神经网络能够更迅速、准确地进行纹理识别。同时,该文还评价了优化方法的有效性。     

基于人工神经网络的数据处理和多目标群方法用土壤传递函数(PTFs)估算保水性

Pedotransfer functions(PTFs) have been developed to estimate soil water retention curves(SWRC) by various techniques.In this study PTFs were developed to estimate the parameters(θ s,θ r,α and λ) of the Brooks and Corey model from a data set of 148 samples.Particle and aggregate size distribution fractal parameters(PSDFPs and ASDFPs,respectively) were computed from three fractal models for either particle or aggregate size distribution.The most effective model in each group was determined by sensitivity analysis.Along with the other variables,the selected fractal parameters were employed to estimate SWRC using multi-objective group method of data handling(mGMDH) and different topologies of artificial neural networks(ANNs).The architecture of ANNs for parametric PTFs was different regarding the type of ANN,output layer transfer functions and the number of hidden neurons.Each parameter was estimated using four PTFs by the hierarchical entering of input variables in the PTFs.The inclusion of PSDFPs in the list of inputs improved the accuracy and reliability of parametric PTFs with the exception of θ s.The textural fraction variables in PTF1 for the estimation of α were replaced with PSDFPs in PTF3.The use of ASDFPs as inputs significantly improved α estimates in the model.This result highlights the importance of ASDFPs in developing parametric PTFs.The mGMDH technique performed significantly better than ANNs in most PTFs.     

锡林河流域长时间序列蒸散量遥感监测及其相关因子

蒸散的准确估算对于草地干旱监测、水资源分布及利用等具有重要的参考价值。选择锡林河流域为研究区,基于地表能量平衡原理,利用遥感方法对2000—2014年每年7月、2000年、2007年、2010—2014年每年4—9月的MODIS影像数据进行处理,结合同期气象资料估算出流域日蒸散量,按所占日数加权得到月蒸散量。运用FAO推荐式进行了验证,平均相对误差为16.678%,在误差允许范围之内,说明该遥感方法有一定的可用性。结果表明,在时间分布上,2000—2014年这15年流域蒸散量的时间变化没有固定的趋向,基本与各年7月份降水量的趋势相一致,一年中蒸散量的最大值主要出现在6—7月份。将月蒸散量与月日均气温、风速、大气相对湿度、水汽压和月降水量作了单因子相关分析,表明,与蒸散量最为相关的气象因子是降水量,说明降水量是影响蒸散量大小的主要限制因子;由于气温季节变化明显,因此气温也是影响蒸散量的主要因子,但在每年的同一时间段(如7月),气温变化不明显时,气温就不再是影响蒸散量大小的主要因子了。     

基于改进BP神经网络与支持向量机的土壤墒情预测方法研究

为提高土壤墒情预测精度,提出了一种基于遗传算法（GA）、改进粒子群算法（IPSO）、误差反向传播（BP）神经网络和支持向量机（SVM）的土壤墒情组合预测模型（GA_IPSO_BP-SVM）。该模型首先在BP神经网络的权阈值选择中同时引入GA和IPSO构成GA_IPSO_BP模型,然后对GA_IPSO_BP和SVM模型分别进行训练和数据仿真,最后利用建立的加权模型对GA_IPSO_BP和SVM模型的土壤墒情预测结果进行组合。以安庆市8个监测站某时段内农田土壤墒情数据为例,分别按隔日、两日后和三日后三种时间跨度进行土壤墒情预测,并对照BP、GA-BP、PSO-BP、IPSO-BP、GA_IPSO_BP和SVM模型,验证和比较提出的GA_IPSO_BP-SVM模型的土壤墒情预测精度。结果表明,GA_IPSO_BP-SVM模型的土壤含水量预测相对误差平均值最小。GA_IPSO_BP与SVM模型组合的GA_IPSO_BP-SVM模型提高了土壤墒情的预测精度,更适合于土壤墒情的短期预测,该方法可为农业节水灌溉方案的制定提供技术支撑。