Convolutional neural networks for automated microseismic detection in downhole distributed acoustic sensing data and comparison to a surface geophone array

Distributed acoustic sensing is a growing technology that enables affordable downhole recording of strain wavefields from microseismic events with spatial sampling down to ∼1 m. Exploiting this high spatial information density motivates different detection approaches than typically used for downhole geophones. A new machine learning method using convolutional neural networks is described that operates on the full strain wavefield. The method is tested using data recorded in a horizontal observation well during hydraulic fracturing in the Eagle Ford Shale, Texas, and the results are compared to a surface geophone array that simultaneously recorded microseismic activity. The neural network was trained using synthetic microseismic events injected into real ambient noise, and it was applied to detect events in the remaining data. There were 535 detections found and no false positives. In general, the signal-to-noise ratio of events recorded by distributed acoustic sensing was lower than the surface array and 368 of 933 surface array events were found. Despite this, 167 new events were found in distributed acoustic sensing data that had no detected counterpart in the surface array. These differences can be attributed to the different detection threshold that depends on both magnitude and distance to the optical fibre. As distributed acoustic sensing data quality continues to improve, neural networks offer many advantages for automated, real-time microseismic event detection, including low computational cost, minimal data pre-processing, low false trigger rates and continuous performance improvement as more training data are acquired.     

Unscented Kalman Filter trained neural network control design for ship autopilot with experimental and numerical approaches

In the recent decades, the application and research of unmanned surface vessels are experiencing considerable growth, which have caused the demands of intelligent autopilots to grow along with the ever-growing requirements. In this study, the design of an autopilot based on Unscented Kalman Filter (UKF) trained Radial Basis Function Neural Networks (RBFNN) was presented. In particular, in order to provide satisfactory control performance for surface vessels with random external disturbances, the modified UKF was utilised as the weights training mechanism for the RBFNN based controller. The configurations of the newly developed free running scaled model, as well as the online signal processing method, were introduced to enable the experimental studies. The experimental and numerical tests were carried out through using the physical scaled model and corresponding mathematical model to validate the capability of the designed control system under various sailing conditions. The results indicated that the UKF RBFNN based autopilot satisfied the functionalities of course keeping, course changing and trajectory tracking only using the rudder as the actuator. It was concluded that the developed control scheme was effective to track the desired states and robust against unpredictable external disturbances. Moreover, in comparison with Back-Propagation (BP) RBFNN and Proportional-Derivative (PD) based autopilots, the UKF RBFNN based autopilot has the comparable capability in the aspects of providing smooth and effective control laws.     

FAM人工神经网络模型在震后趋势判断中的应用

介绍了FAM模糊人工神经网络的原理、学习方法与规则,提出了用于地震震后序列3种震型的判断标准及现场预报规则,建立了基于FAM人工神经网络模型的地震现场预报手段与准则.     

The application of the self-tuning neural network PID controller on the ship roll reduction in random waves

In this paper, we present a mathematical model including seakeeping and maneuvering characteristics to analyze the roll reduction for a ship traveling with the stabilizer fin in random waves. The self-tuning PID controller based on the neural network theory is applied to adjust optimal stabilizer fin angles to reduce the ship roll motion in waves. Two multilayer neural networks, including the system identification neural network (NN1) and the parameter self-tuning neural network (NN2), are adopted in the study. The present control technique can save the time for searching the optimal PID gains in any sea states. The simulation results show that the present developed self-tuning PID control scheme based on the neural network theory is indeed quite practical and sufficient for the ship roll reduction in the realistic sea.     

中国县域农村贫困的空间模拟分析

以中国县级行政区划为研究单元,从自然和社会经济因素中选取贫困的影响因子,建立评价指标体系,利用GIS空间分析和 BP人工神经网络,模拟各县域的自然致贫指数和社会经济消贫指数,并在分析贫困内在形成原因的基础上,明晰了空间贫困的分布特征。结果显示：自然因素是现阶段中国县域主要的致贫原因,全国县域自然致贫指数的分布呈现出明显随纬度和经度地带性分布的规律,自北而南、自西而东逐次呈带状排列分布。社会经济因素对贫困起到一定的缓解作用,全国县域社会经济消贫指数的空间分布较为破碎,各省区内部县域社会经济消贫指数的变异系数均大大高于自然致贫指数的变异系数。全国贫困压力指数以“黑河-百色”一线为界,东中西差异显著,呈现“大分散、小聚集”的空间分布格局。本文识别的贫困县与国家确定的重点扶贫县在空间上具有较高的重合性。     

Fusion of PolSAR and PolInSAR data for land cover classification

The main research goal of this study is to investigate the complementarity and fusion of different frequencies (L- and P-band), polarimetric SAR (PolSAR) and polarimetric interferometric (PolInSAR) data for land cover classification. A large feature set was derived from each of these four modalities and a two-level fusion method was developed: Logistic regression (LR) as ‘feature-level fusion’ and the neural-network (NN) method for higher level fusion. For comparison, a support vector machine (SVM) was also applied. NN and SVM were applied on various combinations of the feature sets.     

Crest level assessment of coastal structures — General overview

The present paper gives a brief overview of the CLASH project, making reference to other relevant papers in this issue and elsewhere. Emphasis is put on the two main objectives of the project and how these objectives were realised: development of a generic prediction method for wave overtopping and guidance on possible scale/model effects for wave overtopping.     

Prediction models for tidal level including strong meteorologic effects using a neural network

Accurate prediction of tidal level including strong meteorologic effects is very important for human activities in oceanic and coastal areas. The contribution of non-astronomical components to tidal level may be as significant as that of astronomical components under the weather, such as typhoon and storm surge. The traditional harmonic analysis method and other models based on the analysis of astronomical components do not work well in these situations. This paper describes the Back-Propagation Neural Network (BPNN) approach, and proposes a method of iterative multi-step prediction and the concept of periodical analysis. The prediction among stations shows that the BPNN model can predict the tidal level with great precision regardless of different tide types in different regions. Based on the non-stationary characteristic of hourly tidal record including strong meteorologic effects, three Back-Propagation Neural Network models were developed in order to improve the accuracy of prediction and supplement of tidal records: (1) Difference Neural Network model (DNN) for the supplementing of tidal record; (2) Minus-Mean-Value Neural Network model (MMVNN) for the corresponding prediction between tidal gauge stations; (3) Weather-Data-based Neural Networks model (WDNN) for set up and set down.The results show that the above models perform well in the prediction of tidal level or supplement of tidal record including strong meteorologic effects.     

基于循环神经网络的欧亚中高纬夏季极端高温年代际预测模型研究

近几十年来频繁发生的极端高温事件严重威胁着自然生态系统、社会经济发展和人类生命安全。针对生态环境脆弱的欧亚中高纬地区,首先评估了当前主流动力模式(CMIP6 DCPP)对于该地区夏季极端高温的年代际预测水平,并构建了基于循环神经网络(Recurrent Neural Networks,RNN)的年代际预测模型。多模式集合平均(Multi-Model Ensemble,MME)的评估结果显示,得益于大样本和初始化的贡献,当前动力模式对于60°N以南区域(South Eurasia,SEA)展现了预测技巧,准确预测出了其线性增长趋势和1968—2008年间主要的年代际变率,然而模式对于60°N以北区域(North Eurasia,NEA)极端高温的年代际变率几乎没有任何预测技巧,仅预测出比观测低的线性增长趋势。基于86个初始场的动力模式大样本预测结果,RNN将2008—2020年间NEA和SEA极端高温的年代际变率预测技巧显著提高,距平相关系数技巧从MME中的-0.61和-0.03,提升至0.86和0.83,均方差技巧评分从MME中的-1.10和-0.94,提升至0.37和0.52。RNN的实时预测结果表明,在2021—2026年,SEA区域的极端高温将持续增加,2026年很可能发生突破历史极值的极端高温事件,NEA区域在2022年异常偏低,而后将呈现波动上升。     

深度学习农作物分类的弱样本适用性

基于大数据驱动的深度学习挖掘图像数据的规律和层次已成为遥感影像解译的研究热点。海量标签样本是训练深度学习模型的前提条件,但成本昂贵的人工标记样本限制了深度学习技术在遥感领域的应用。本文提出了一种基于弱样本的深度学习模型农作物分类策略：以GF-1影像为数据源,将传统分类器SVM分类结果视为弱样本,训练深度卷积网络模型DCNN (Deep Convolutional Neural Networks),获取辽宁省水稻和玉米的空间分布,分析弱样本的适用性。结果显示：测试集总体精度达到0.90,水稻和玉米F1分数分别为0.81和0.90;在不同地形地貌、复杂种植结构的农业景观下均表现出良好的分类效果;与SVM结果的空间一致性为0.90;当弱样本最大面积误差比例小于0.36时,弱样本仍适用于DCNN作物分类,结果的总体精度保持在0.86以上。综上,该策略一定程度上消除了深度学习模型对大量人工标记样本高度依赖的局限性,为实现大尺度农作物遥感分类提供了一种新途径。