摘要： Determination of in-season N requirement for corn (Zea mays L.) is challenging due to interactions of genotype, environment, and management. Machine learning (ML), with its predictive power to tackle complex systems, may solve this barrier in the development of locally based N recommendations. The objective of this study was to explore application of ML methodologies to predict economic optimum nitrogen rate (EONR) for corn using data from 47 experiments across the US Corn Belt. Two features, a water table adjusted available water capacity (AWC(wt)) and a ratio of in-season rainfall to AWC(wt) (RAWC(wt)), were created to capture the impact of soil hydrology on N dynamics. Four ML models-linear regression (LR), ridge regression (RR), least absolute shrinkage and selection operator (LASSO) regression, and gradient boost regression trees (GBRT)-were assessed and validated using "leave-one-location-out" (LOLO) and "leave-one-year-out" (LOYO) approaches. Generally, RR outperformed other models in predicting both at planting and split EONR times. Among the 47 tested sites, for 33 sites the predicted split EONR using RR fell within the 95% confidence interval, suggesting the chance of using the RR model to make an acceptable prediction of split EONR is similar to 70%. When RR was used to test split EONR prediction with input weather features surrogated with 10 yr of historical weather data, the model demonstrated robustness (MAE, 33.6 kg ha(-1);R-2 = 0.46). Incorporating mechanistically derived hydrological features significantly enhanced the ability of the ML procedures to model EONR. Improvement in estimating in-season soil hydrological status seems essential for success in modeling N demand.

关键词： Toxicity prediction   machine learning   structure-activity relationship   prediction accuracy   model reliability   molecular descriptors  

摘要： In silico toxicity prediction plays an important role in the regulatory decision making and selection of leads in drug design as in vitro/vivo methods are often limited by ethics, time, budget, and other resources. Many computational methods have been employed in predicting the toxicity profile of chemicals. This review provides a detailed end-to-end overview of the application of machine learning algorithms to Structure-Activity Relationship (SAR)-based predictive toxicology. From raw data to model validation, the importance of data quality is stressed as it greatly affects the predictive power of derived models. Commonly overlooked challenges such as data imbalance, activity cliff, model evaluation, and definition of applicability domain are highlighted, and plausible solutions for alleviating these challenges are discussed.

关键词： 机器学习方法   随机森林算法   适宜性评价   农村居民点   分类和预测   房县  

摘要： 农村居民点适宜性评价是优化农村土地利用布局的基础性工作之一,解决该研究领域存在的问题,提高评价结果的科学合理性,使之更好地服务于农村土地资源的转型、优化利用,事关乡村振兴战略的实施。本文引入机器学习方法,通过对小规模已知样本的学习实现大规模未知数据的高精度分类与评价,从方法可行性、数模结合、优化策略和预测实现等方面进行探讨,应用随机森林算法对湖北省房县农村居民点利用的适宜性问题展开实证研究。结果显示:(1)房县农村居民点适宜性受其可通达能力、海拔高度、地块面积及当地农业生产活跃程度的影响较大,也与农民生活、收入水平紧密相关;(2)训练模型的性能受特征(评价因子)规模变化的影响较小,其中,16维特征集合模型的测试精度可达83.54%;(3)不同训练模型的预测结果均反映出房县的非适宜型居民点规模较大,该算法的鲁棒性强,评价结果稳定可靠。研究表明,机器学习方法能较好地支撑农村居民点适宜性评价工作,评价结果能为优化农村土地利用和基层治理提供数据基础。

关键词： 高斯过程回归   水温预测   气温   密苏里河   机器学习方法  

摘要： 通过收集密苏里河三个站点的多年实测水温、气温数据,建立了水温-气温的高斯过程回归模型.和传统的水温回归模型（线性、非线性及随机回归模型）相比,高斯过程回归模型的精度较高,各研究站点的相关系数值均较高（0.966 4-0.989 7）,均方根误差值相对较小（1.978 4-1.495 0）.高斯过程回归作为一种较为先进的机器学习方法,和其他机器学习方法相比,其突出的优点在于结合了许多机器学习任务,包括模型训练、不确定性分析及超参数估计等,该方法可以有效地应用于天然河流站点水温的预测.

关键词： 机器学习   决策树   LightGBM   气温   预报试验  

摘要： 利用2015—2017年湖北89个气象站地面观测温度、欧洲中心再分析资料和0～12 h预报资料回归模式输出要素与地面气温之间的关系,建立了LightGBM模型,并在2018年数据集上进行测试。结果表明,定时气温平均绝对误差由模式本身的1.8℃下降到1.1℃,2℃以内预报准确率由65.9%上升至86.6%,决定系数(拟合优度)高达0.97。该模型已经在武汉中心气象台业务化,初步选取定时气温中的极值进行2018年2—6月预报评分,24 h高、低温预报准确率分别为76.9%和91.4%,在客观产品中排名前列,较数值预报模式产品提升明显,低温预报准确率超过预报员水平。LightGBM作为一个年轻的机器学习框架,在气象要素预报方面具备良好的应用前景。

关键词： 质子转移反应质谱(PTR-MS)   挥发性有机物(VOCs)   数据预处理   机器学习  

摘要： 质子转移反应质谱(PTR-MS)法是一种用于检测挥发性有机物(VOCs)的分析技术。它具有检测限低、响应速度快、无需样品前处理、实时分析等特点,在大气化学、环境化学、食品、生物医学等领域得到广泛应用。随着PTR-MS应用的扩展和样品种类的增加,如何从复杂的质谱数据中提取特征,并寻找内在规律,对分析算法的处理能力提出了更高的要求。本工作从数据预处理技术和机器学习方法两方面展开论述,归纳了具有PTR-MS特点的数据预处理技术,总结了不同机器学习算法在PTR-MS数据分析中的应用,并讨论了它们的优点和不足。

摘要： The statolith in cephalopods has a stable morphology and contains important ecological information. Influenced by genetic structures and environmental variability, statolith shapes often vary among different stocks and are ideal indices for stock discrimination. In the present study, wavelet analysis was used to explore the statolith shape variations in Dosidicus gigas (D'Orbigny, 1835 in 1834-1847) among four geographic stocks obtained by Chinese jigging fleets in the eastern tropical Pacific Ocean (ETP). In addition, machine learning methods were compared with traditional classification methods to improve the stock classification results of D. gigas. According to our analyses, statolith shapes of D. gigas differed significantly among the four stocks. Wavelet coefficients extracted from the statolith images by computer software were used to reconstruct the mean statolith shape for every stock. The rostrum and wing of the statolith are two main components determining the variances among stocks. Canonical analysis of principal coordinates dearly separated Costa Rican from other stocks. Machine learning methods performed better than the traditional method of statolith shape dassification. The results of our study supported the geographical separation of D. gigas stocks (Costa Rican and equatorial stock in the northern hemisphere, and Peruvian and Chilean stock in the southern hemisphere) reported in previous studies. Wavelet analysis is an appropriate method for stock classification and machine learning methods can effectively improve the classification accuracy and is a promising method for determining the stock structure.

关键词： Machine learning   Ore grade estimation   Ordinary kriging   Indicator kriging   Neural networks   Random forests   Gaussian processes  

摘要： In this study, machine learning methods such as neural networks, random forests, and Gaussian processes are applied to the estimation of copper grade in a mineral deposit. The performance of these methods is compared to geostatistical techniques, such as ordinary kriging and indicator kriging. To ensure that these comparisons are realistic and relevant, the predictive accuracy is estimated on test instances located in drill holes that are different from the training data. The results of an extensive empirical study in the Sarcheshmeh porphyry copper deposit in Southeastern Iran illustrate that specially designed Gaussian processes with a symmetric standardization of the spatial location inputs and an anisotropic kernel yield the most accurate predictions. Furthermore, significant improvements are obtained when, besides location, information on the rock type is included in the set of predictor variables. This observation highlights the importance of carrying out detailed studies of the geological composition of the deposit to obtain more accurate ore grade predictions.

关键词： fractional vegetation cover   VIIRS surface reflectance   machine learning method   global  

摘要： Fractional vegetation cover (FVC) is an essential input parameter for many environmental and ecological models. Recently, several global FVC products have been generated using remote sensing data. The Global LAnd Surface Satellite (GLASS) FVC product, which is generated from Moderate Resolution Imaging Spectroradiometer (MODIS) data, has attained acceptable performance. However, the original MODIS operation design lifespan has been exceeded. The Visible Infrared Imaging Radiometer Suite (VIIRS) onboard the Suomi National Polar-Orbiting Partnership (S-NPP) satellite was designed to be the MODIS successor. Therefore, developing an FVC estimation algorithm for VIIRS data is important for maintaining continuous FVC estimates in case of MODIS failure. In this study, a global FVC estimation algorithm for VIIRS surface reflectance data was proposed based on machine learning methods, which investigated the performances of back propagating neural networks (BPNNs), general regression networks (GRNNs), multivariate adaptive regression splines (MARS), and Gaussian process regression (GPR). The training samples were extracted from the GLASS FVC product and corresponding reconstructed VIIRS surface reflectance in 2013 over the global sampling locations. The VIIRS reflectances of red and near infrared (NIR) bands were the input variables for these machine learning methods. The theoretical performances and independent validation results indicated that the four machine learning methods could achieve similar and reliable FVC estimates. Regarding the FVC estimation accuracy, the GPR method achieved the best performance (R-2 = 0.9019, RMSE = 0.0887). The MARS method had the obvious advantage of computational efficiency. Furthermore, the FVC estimates achieved good spatial and temporal continuities. Therefore, the proposed FVC estimation algorithm for VIIRS data can potentially generate reliable global FVC data for related applications.

关键词： Multiuser data transmission   Spatial coupling   Constellation rotation   Power allocation   Coordinate ascent algorithm   Genetic algorithm  

摘要： The rapid development of mobile Internet triggered massive data exchange among multiple sources, which then results in challenges in the efficiency of multi-user data transmission. Spatial coupling multiple access, where different users share the same resource blocks by superimposing data streams with different time offsets, has been proposed to obtain higher spectral efficiency. The optimization of the system can be carried out through for example constellation rotation or power allocation, which mitigates the interferences and distinguishes users/data streams better. The optimization aims for the maximization of the average mutual information between the transmitted data symbols and the received signal. To solve the optimization problem, we employ machine learning based methods to locate the possibly global optimal solution. Simulation results show that the optimization clearly contributes to the performance improvement of the system.