Mineral Potential Mapping Using Bayesian Learning for Multilayer Perceptrons

Multilayer perceptrons (MLPs) can be used to discover a function which can be used to map from a set of input variables onto a value representing the conditional probability of mineralization. The standard approach to training MLPs is gradient descent, in which the error between the network output and the target output is reduced in each iteration of the training algorithm. In order to prevent overfitting, a split-sample validation procedure is used, in which the data is partitioned into two sets: a training set, which is used for weight optimization, and a validation set, which is used to optimize various parameters that can be used to prevent overfitting. One of the problems with this approach is that the resulting maps can display significant variability which stems from (i) the (randomly initialized) starting weights and (ii) the particular training/validation set partition (also determined randomly). This problem is especially pertinent on mineral potential mapping tasks, in which the number of deposit cells is a very small proportion of the total number of cells in the study area. In contrast to gradient descent methods, Bayesian learning techniques do not find a single weight vector; rather, they infer the posterior distribution of the weights given the data. Predictions are then made by integrating over this distribution. An important advantage of the Bayesian approach is that the optimization of parameters such as the weight decay regularization coefficient can be performed using training data alone, thus avoiding the noise introduced through split-sample validation. This paper reports results of applying Bayesian learning techniques to the production of maps representing gold mineralization potential over the Castlemaine region of Victoria, Australia. Maps produced using the Bayesian approach display significantly less variability than those produced using gradient descent training. They are also more reliable at predicting the presence of unknown deposits.     

Geochemical patterns of schists from the Bushmanland Group: An artificial neural networks approach

The Mesoproterozoic Bushmanland Group is situated in the central region of the 1000 to 1200 Ma Namaqualand Metamorphic Complex (NMC). The NMC comprises a belt of highly deformed medium- to high-grade metamorphic rocks to the west of the Archean Kaapvaal Craton of southern Africa. The Bushmanland Group, one of the many supracrustal sequences that make up the NMC, is a metavolcano-sedimentary succession that hosts economically significant concentrations of sillimanite and base-metal sulfide deposits. The present investigation was carried out to study the geochemistry of a large set of representative samples of psammo-pelitic schists from the Bushmanland Group, which includes data from three different schist units: Namies Schist Formation, Shaft Schist Formation and Ore Equivalent Schist. The objective was three-fold: to test the lateral correlatability of these schist units as determined by field relationships, to identify the geochemical signature of the schists and to test the validity of an Artificial Neural Network approach as an exploration tool. Two multidimensional datasets, respectively comprising 10 major and 18 trace elements, were constructed using selected published schist analyses. Both schist datasets were analyzed using self-organizing neural maps for visualizing and clustering high-dimensional geochemical data. Geochemical differences between the various schists were visualized using colored two-dimensional maps that can be visually and quantitatively interpreted. The results of this study confirm the lateral correlatability of the schist units evaluated in this communication. It was also found that each schist unit or portions of them represent a distinct geochemical signature that is related to true lithological variations. The results show that the Artificial Neural Network approach can be used as a powerful tool for regional mineral exploration in poly-deformed and metamorphosed terrains where identification of stratigraphic units through lateral correlation by means of fieldwork and petrography remains highly speculative.     

GIS技术和神经网络模型在地震地质灾害小区划中的应用研究

在前人对地震地质灾害小区划研究的基础上，以甘肃省景泰县城为实例，综合地震地质、地形地貌、活动断裂和场地条件等资料，利用地理信息系统技术和神经网络模型进行地震地质灾害小区划分类．结果表明该方法使以前定性的各类指标尽可能定量化，实现了计算机自动成图，提高了工作效率，结果更加准确直观．     

Reservoir permeability prediction by neural networks combined with hybrid genetic algorithm and particle swarm optimization

Reservoir characterization involves describing different reservoir properties quantitatively using various techniques in spatial variability. Nevertheless, the entire reservoir cannot be examined directly and there still exist uncertainties associated with the nature of geological data. Such uncertainties can lead to errors in the estimation of the ultimate recoverable oil. To cope with uncertainties, intelligent mathematical techniques to predict the spatial distribution of reservoir properties appear as strong tools. The goal here is to construct a reservoir model with lower uncertainties and realistic assumptions. Permeability is a petrophysical property that relates the amount of fluids in place and their potential for displacement. This fundamental property is a key factor in selecting proper enhanced oil recovery schemes and reservoir management. In this paper, a soft sensor on the basis of a feed‐forward artificial neural network was implemented to forecast permeability of a reservoir. Then, optimization of the neural network‐based soft sensor was performed using a hybrid genetic algorithm and particle swarm optimization method. The proposed genetic method was used for initial weighting of the parameters in the neural network. The developed methodology was examined using real field data. Results from the hybrid method‐based soft sensor were compared with the results obtained from the conventional artificial neural network. A good agreement between the results was observed, which demonstrates the usefulness of the developed hybrid genetic algorithm and particle swarm optimization in prediction of reservoir permeability.     

On chaos control of nonlinear fractional chaotic systems via a neural collocation optimization scheme and some applications

In this paper, we study chaos control of a class of fractional-order chaotic systems where the dynamic control system depends on the Caputo fractional derivatives. We first propose an infinite horizon optimal control problem related to the given fractional chaotic system. With the help of an approximation, we replace the Caputo derivative to integer order derivative. We then convert the obtained infinite horizon optimal control problem into an equivalent finite horizon one. Based on the Pontryagin minimum principle (PMP) for optimal control problems and by constructing an error function, we define an unconstrained minimization problem. In the optimization problem, we use trial solutions for state, costate and control functions where these trial solutions are constructed by using a two-layered perceptron neural network. A learning procedure of the proposed neural network with convergence properties are also given. Some numerical results are introduced to explain our main results. Three applicable examples on chaos control of Malkus waterwheel, finance fractional chaotic models and fractional-order Geomagnetic Field models are finally considered.     

果蝇优化算法优化灰色神经网络的卫星钟差预报

针对导航卫星短期钟差预报精度不高的问题,文章提出了一种基于果蝇优化算法(FOA)优化灰色神经网络的卫星钟差预报方法.利用FOA较强的全局寻优能力对灰色参数进行迭代动态微调,改善随机初始化所导致网络进化易陷入局部最优的问题,以提高灰色神经网络的预报精度;选取IGS产品中典型的卫星钟差数据,分别采用FOA优化灰色神经网络模型、神经网络模型、灰色系统模型和灰色神经网络模型进行短期钟差预报.仿真结果表明:FOA优化灰色神经网络模型的预报精度优于其他三种模型,性能满足卫星短期高精度钟差预报的要求.     

Comparison of two inverse analysis techniques for learning deep excavation response

Performance observation is a necessary part of the design and construction process in geotechnical engineering. For deep urban excavations, empirical and numerical methods are used to predict potential deformations and their impacts on surrounding structures. Two inverse analysis approaches are described and compared for an excavation project in downtown Chicago. The first approach is a parameter optimization approach based on genetic algorithm (GA). GA is a stochastic global search technique for optimizing an objective function with linear or non-linear constraints. The second approach, self-learning simulations (SelfSim), is an inverse analysis technique that combines finite element method, continuously evolving material models, and field measurements. The optimization based on genetic algorithm approach identifies material properties of an existing soil model, and SelfSim approach extracts the underlying soil behavior unconstrained by a specific assumption on soil constitutive behavior. The two inverse analysis approaches capture well lateral wall deflections and maximum surface settlements. The GA optimization approach tends to overpredict surface settlements at some distance from the excavation as it is constrained by a specific form of the material constitutive model (i.e. hardening soil model); while the surface settlements computed using SelfSim approach match the observed ones due to its ability to learn small strain non-linearity of soil implied in the measured settlements.     

利用深度神经网络和先兆信号的江苏夏季降水客观预测方法

利用1961—2019年江苏省67个站降水量和气候指数数据集等资料,选取大气环流、海温和积雪等先兆信号的不同组合作为预测因子方案,通过对比不同机器学习方法对江苏省夏季降水开展预测试验。结果表明,深度神经网络（Deep Neural Network,DNN）较传统统计方法和其他机器学习方法有一定优势,深度神经网络结合动态权重集合因子方案对江苏省夏季降水的预测技巧最高,其独立样本检验结果稳定,2015—2019年的平均PS评分为76.0,距平符号一致率为0.62,距平相关系数达0.35,尤其对江苏省中南部的预测技巧更高,具有业务应用价值。不同预测因子方案对比分析表明,大气环流因子在江苏省夏季降水预测中做主要贡献,而海温因子和积雪等其他因子也有正贡献,说明使用综合性预测因子以及集合方案有助于提升季节预测准确率。      

Multi-hazard susceptibility mapping based on Convolutional Neural Networks

Multi-hazard susceptibility prediction is an important component of disasters risk management plan. An effective multi-hazard risk mitigation strategy includes assessing individual hazards as well as their interactions. However, with the rapid development of artificial intelligence technology, multi-hazard susceptibility prediction techniques based on machine learning has encountered a huge bottleneck. In order to effectively solve this problem, this study proposes a multi-hazard susceptibility mapping framework using the classical deep learning algorithm of Convolutional Neural Networks (CNN). First, we use historical flash flood, debris flow and landslide locations based on Google Earth images, extensive field surveys, topography, hydrology, and environmental data sets to train and validate the proposed CNN method. Next, the proposed CNN method is assessed in comparison to conventional logistic regression and k-nearest neighbor methods using several objective criteria, i.e., coefficient of determination, overall accuracy, mean absolute error and the root mean square error. Experimental results show that the CNN method outperforms the conventional machine learning algorithms in predicting probability of flash floods, debris flows and landslides. Finally, the susceptibility maps of the three hazards based on CNN are combined to create a multi-hazard susceptibility map. It can be observed from the map that 62.43% of the study area are prone to hazards, while 37.57% of the study area are harmless. In hazard-prone areas, 16.14%, 4.94% and 30.66% of the study area are susceptible to flash floods, debris flows and landslides, respectively. In terms of concurrent hazards, 0.28%, 7.11% and 3.13% of the study area are susceptible to the joint occurrence of flash floods and debris flow, debris flow and landslides, and flash floods and landslides, respectively, whereas, 0.18% of the study area is subject to all the three hazards. The results of this study can benefit engineers, disaster managers and local government officials involved in sustainable land management and disaster risk mitigation.