关键词： 分子动力学模拟   人工边界条件   机器学习   前馈神经网络  

摘要： 本文运用机器学习方法设计一维线性原子链的人工边界条件。该方法基于前馈神经网络,通过对一小部分数值解进行训练后得到人工边界条件。应用该法不需要较多先验知识,编程简单,实现速度快,算例表明数值反射较小。

关键词： 工业物联网   机器学习   神经网络   预测   工业智能   线性回归  

摘要： 随着传感器网络等技术在工业领域的不断拓展,工业物联网技术得到了长足发展。工业物联网及基于工业物联网的数据分析与预测是工业智能系统的基础。文中关注在工业互联网环境下,工业智能系统在训练阶段利用有监督的机器学习方法,将工艺参数以及环境参数作为输入,将生产过程关注的关键性能指标作为输出,从而对生产线复杂系统进行建模;在生产阶段利用训练阶段的系统模型及当前输入对关键性能指标进行预测。分别利用多元线性回归和非线性BP神经网络方法建模,并利用当前输入对输出结果进行预测,从而对关键性能指标进行预警及预处理操作。同时还利用真实的工业物联网数据进行了仿真和验证,证明了基于BP神经网络的预测方法性能优于线性预测技术。

关键词： 智能采油   大数据应用   产量预测   机器学习   特征选择   长短期记忆神经网络  

摘要： 油藏数值模拟是进行油田产量预测最为常用的方法,但其准确性建立在精确的地质模型和较高质量的历史拟合基础之上。为了克服数值模拟计算耗时长、成本高和所需数据资料多等缺点,建立了一种利用机器学习方法,根据现场广泛易得的油藏静态资料和开发动态参数实现油井日产油量的快速准确预测。传统的BP神经网络无法准确描述产量变化在时间维度上的相关性,因而基于长短期记忆神经网络(LSTM),建立能够考虑生产动态数据变化趋势和前后关联性的产量预测模型,是实现油井日产油量预测更为有效的途径。首先根据平均不纯度减少(MDI)方法,分析各个因素对单井产量的影响程度,基于特征参数的重要性进行数据降维,排除不相关的冗余特征,确定影响油井产量的主要因素。结合筛选出的特征参数和日产油量数据对LSTM模型进行训练和优化,建立最终的油井产量预测模型。利用实际油田数据对建立的模型进行验证和应用效果评价,结果表明基于LSTM模型的产量预测值与实际值高度一致,能准确反映产量的动态变化规律,为油井产量预测提供了一种新的方法。

关键词： Alzheimer's disease   Machine learning   Spectrogram   Speech recognition  

摘要： Most of the studies on Alzheimer's disease (AD) have been carried out using medical images. However, the acquisition of medical images data is difficult. The identification based on the patients speech data can effectively reduce the medical cost, and the speech data can be collected in a non-invasive manner so that the patients data can be collected in real-time and accurately. This paper proposes a new method that uses the spectrogram features extracted from speech data to identify AD, which can help families to understand the disease development of patients in an earlier stage, so that they can take measures in advance to delay the disease development. We use the speech data collected from the elderly that express the speech features displayed in the speech and used the machine learning methods for identifying AD. During the simulation and experiment, we collect a new speech dataset, which includes Alzheimer's disease patients and healthy control subjects. Then, we compare with the speech data made available by the Dem@Care project. Among the tested models, LogisticgressionCV model exhibited the best performance. It is shown that this method using extracted spectrogram features from speech data to identify AD is feasible. The credibility of the new dataset and feasibility of the used methods in this paper are demonstrated.

关键词： Risk assessment   Machine learning   Artificial neural network   Review  

摘要： The purpose of this article is to present a structured review of publications utilizing machine learning methods to aid in engineering risk assessment. A keyword search is performed to retrieve relevant articles from the databases of Scopus and Engineering Village. The search results are filtered according to seven selection criteria. The filtering process resulted in the retrieval of one hundred and twenty-four relevant research articles. Statistics based on different categories from the citation database is presented. By reviewing the articles, additional categories, such as the type of machine learning algorithm used, the type of input source used, the type of industry targeted, the type of implementation, and the intended risk assessment phase are also determined. The findings show that the automotive industry is leading the adoption of machine learning algorithms for risk assessment. Artificial neural networks are the most applied machine learning method to aid in engineering risk assessment. Additional findings from the review process are also presented in this article.

关键词： Agriculture   Machine learning   Actual yield   Yield Gap   Remote sensing   National scale   Regression  

摘要： Closing the yield gap between actual and potential wheat yields in Australia is important to meet the growing global demand for food. The identification of hotspots of the yield gap, where the potential for improvement is the greatest, is a necessary step towards this goal. While crop growth models are well suited to quantify potential yields, they lack the ability to provide accurate large-scale estimates of actual yields, owing to the sheer quantity of data they require for parameterisation. In this context, we sought to provide accurate estimates of actual wheat yields across the Australian wheat belt based on machine-learning regression methods, climate records and satellite image time series. Out of nine base learners and two ensembles, support vector regression with radial basis function emerged as the single best learner (root mean square error of 0.55 t ha(-1) and R-2 of 0.77 at the pixel level). At national scale, this model explained 73% of the yield variability observed across statistical units. Benchmark approaches based on peak Normalised Difference Vegetation Index (NDVI) and on a harvest index were largely outperformed by the machine-learning regression models (R-2 < 0.46). Climate variables such as maximum temperatures and accumulated rainfall provided additional information to the 16-day NDVI time series as they significantly improved yield predictions. Variables observed up to and around the flowering period had a particularly high predictive power with additional information gained from data during grain filling. We further showed that, while all models were sensitive to a reduction of the training set size, a large majority had not reached saturation with a data set of 125 fields (2000 pixels). This indicates that additional training data are likely to further improve the skill of the models. We estimated that observations from 75 fields (1200 pixels) are required for the best single model to reach an R-2 of 0.7. We contend that machine-learning

关键词： landslide susceptibility (LSM)   alternative decision tree (ADTree)   Fisher's linear discriminant function (FLDA)   machine learning approach   R program  

摘要： This analysis aims to generate landslide susceptibility maps (LSMs) using various machine learning methods, namely random forest (RF), alternative decision tree (ADTree) and Fisher's Linear Discriminant Function (FLDA). The results of the FLDA, RF and ADTree models were compared with regard to their applicability for creating an LSM of the Gallicash river watershed in the northern part of Iran close to the Caspian Sea. A landslide inventory map was created using GPS points obtained in a field analysis, high-resolution satellite images, topographic maps and historical records. A total of 249 landslide sites have been identified to date and were used in this study to model and validate the LSMs of the study region. Of the 249 landslide locations, 70% were used as training data and 30% for the validation of the resulting LSMs. Sixteen factors related to topographical, hydrological, soil type, geological and environmental conditions were used and a multi-collinearity test of the landslide conditioning factors (LCFs) was performed. Using the natural break method (NBM) in a geographic information system (GIS), the LSMs generated by the RF, FLDA, and ADTree models were categorized into five classes, namely very low, low, medium, high and very high landslide susceptibility (LS) zones. The very high susceptibility zones cover 15.37% (ADTree), 16.10% (FLDA) and 11.36% (RF) of the total catchment area. The results of the different models (FLDA, RF, and ADTree) were explained and compared using the area under receiver operating characteristics (AUROC) curve, seed cell area index (SCAI), efficiency and true skill statistic (TSS). The accuracy of models was calculated considering both the training and validation data. The results revealed that the AUROC success rates are 0.89 (ADTree), 0.92 (FLDA) and 0.97 (RF) and predication rates are 0.82 (ADTree), 0.79 (FLDA) and 0.98 (RF), which justifies the approach and indicates a reasonably good landslide prediction. The results of the S

关键词： bioinformatics   personalized medicine   oncology   chemotherapy   machine learning   omics profiling  

摘要： (1) Background: Machine learning (ML) methods are rarely used for an omics-based prescription of cancer drugs, due to shortage of case histories with clinical outcome supplemented by high-throughput molecular data. This causes overtraining and high vulnerability of most ML methods. Recently, we proposed a hybrid global-local approach to ML termed floating window projective separator (FloWPS) that avoids extrapolation in the feature space. Its core property is data trimming, i.e., sample-specific removal of irrelevant features. (2) Methods: Here, we applied FloWPS to seven popular ML methods, including linear SVM, k nearest neighbors (kNN), random forest (RF), Tikhonov (ridge) regression (RR), binomial naive Bayes (BNB), adaptive boosting (ADA) and multi-layer perceptron (MLP). (3) Results: We performed computational experiments for 21 high throughput gene expression datasets (41-235 samples per dataset) totally representing 1778 cancer patients with known responses on chemotherapy treatments. FloWPS essentially improved the classifier quality for all global ML methods (SVM, RF, BNB, ADA, MLP), where the area under the receiver-operator curve (ROC AUC) for the treatment response classifiers increased from 0.61-0.88 range to 0.70-0.94. We tested FloWPS-empowered methods for overtraining by interrogating the importance of different features for different ML methods in the same model datasets. (4) Conclusions: We showed that FloWPS increases the correlation of feature importance between the different ML methods, which indicates its robustness to overtraining. For all the datasets tested, the best performance of FloWPS data trimming was observed for the BNB method, which can be valuable for further building of ML classifiers in personalized oncology.

关键词： Bayesian methods   Site characterization   Spatial variability   Sparse data  

摘要： Identification of subsurface stratification and characterization of spatially varying soil properties profiles in multiple soil layers are indispensable in geotechnical site investigation. Subsurface soils are often stratified first through visual inspection of soil samples obtained from boreholes or soil behavior type index from cone penetration tests. Then, the soil property profile within each layer is characterized via interpolation of the data points measured within the corresponding layer. Although this stratification first procedure is commonly used in engineering practice, it is difficult to apply when measured data points within each layer are sparse and limited (e.g., only a few data points within each layer), a scenario often encountered in geotechnical site characterization. When the number of measurements within each layer is small, it is difficult to properly interpolate spatially varying soil property profiles in each layer. To address this difficulty and increase the number of data points for enabling proper interpolation, an interpolation first procedure is proposed which utilizes measurements from all layers as input for interpolation, followed by soil stratification using the interpolation results. The proposed procedure includes two key elements: (1) a Bayesian supervised learning method for interpolation of non-stationary data from multiple layers, and (2) an unsupervised machine learning method (e.g., clustering) for soil stratification. An index is also proposed to determine when the proposed method is beneficial and performs better than the stratification first procedure in engineering geology practice. Both numerical and real-life data are used to illustrate the proposed method.

关键词： acid solubility   artificial neural network (ANN) TensorFlow   Extreme Gradient Boosting (XGBoost)   mid-infrared spectra   multivariate data analysis (MVDA)  

摘要： Interests in the use of chemometric and data science methods for laboratory techniques have grown rapidly over the last 10 years, for the reason that they are cheaper and faster than traditional analytical methods of material testing. This study uses 888 rock samples collected from the exploration and production (E&P) sector of the oil industry. Based on the Fourier-transform infrared (FT-IR) spectra of these rock samples their solubility predictions have been developed and investigated with nine methods including both linear and non-linear ones. Two of these methods such as Partial Least Squares Regression (PLSR) and Support Vector Regression (SVR) are available in a commercial software package and the other seven methods, Extreme Gradient Boosting (XGBoost), Ridge Regression (RR), k-nearest neighbours (k-NN), Decision Tree (DT), Multilayer Perceptron (MLP), Support Vector Regression (SVR), Artificial Neural Network (ANN) with TensorFlow (TF), were coded by the authors based either on commercial applications or open source libraries. The investigation starts with spectral data pre-processing carried out by standard normal variate (SNV), baseline correction and feature selection methods creating the feature set for all machine learning (ML) applications. The accuracy of predictions has been evaluated with mean squared error as a performance metric for each investigated method. The comparisons of predicted values to real data of test samples have shown that mineral solubility in acids can be well predicted in the range of the uncertainties of real laboratory measurements, therefore it can be used to improve the response time of these investigations and reduce the risk in industrial applications. In those cases, where the unknown samples have got some out of the range features, the limitations in the accuracy of predictions have become clear. We have also identified the limitations in the methodology and planned steps to further improve the prediction capabilities. Th