基于改进BP网络算法的隧洞围岩分类

围岩分类对指导地下工程的设计和施工具有非常重要的意义.引入人工神经网络的方法, 进行隧洞围岩分类, 在传统BP算法的基础上, 通过改进学习算法、优化传递函数和网络结构进行神经网络方法优化.采用附加动量法和学习速率自适应调整的策略改进学习算法, 使得当误差大于上临界值时, 则降低学习率, 当误差小于下临界值时, 则适当提高学习率, 这样可加快网络的训练速度, 确保网络的稳定性; 通过引入调整学习率参数, 使得传递过程更加敏感, 加快了传递函数的收敛速度, 提高了训练函数的计算精度; 通过给定隐含层节点模型的取值范围, 对网络结构进行优化, 提高了泛化精度.将改进的BP网络模型应用于广东省东深供水改造工程的隧洞围岩分类中, 分类结果与根据《水工隧洞设计规范(SL279-2002) 》的分类结果完全一致, 表明该方法具有良好的工程实用性.      

ASTER数据的自组织神经网络分类研究

传统的遥感数据分类方法大多基于统计学的参数估计，假设数据分布服从高斯正态分布。神经网络方法无需参数估计和统计假设，因而，近来越来越多地应用于遥感数据分类之中。介绍了基于聚类分析的自组织特征映射分类方法。ASTER卫星数据是新型遥感数据，包括 3个15 m分辨率波段和 3个30 m分辨率的短波红外波段。选择北京地区的ASTER数据作为方法实验数据，首先对数据进行了小波融合，然后进行了土地覆盖类型的自组织特征映射神经网络分类研究，把研究结果同最大似然判别法得到的分类结果进行了比较，分类精度比最大似然判别法总体提高了9％。     

基于自组织特征映射网络-随机森林模型的滑坡易发性评价

为深入探讨评价单元和非滑坡样本选取对滑坡易发性预测的影响,构建了一种基于自组织特征映射网络-随机森林模型的滑坡易发性评价模型。该模型针对栅格单元和斜坡单元在滑坡易发性评价中的不足,结合栅格单元和斜坡单元的相互关系,提出了滑坡易发性指数的优化计算方法。在此基础上,基于随机森林Tree Bagger分类器构建滑坡易发性评价模型,通过对比分析自组织特征映射网络和随机方法选取非滑坡样本对评价结果的影响,探讨自组织特征映射网络、随机森林和自组织特征映射网络-随机森林三种评价模型的有效性;将评价模型应用于大余县滑坡易发性评价。结果显示,随机森林模型和自组织特征映射网络-随机森林模型的预测精度较高,分别达到91.19%和94.94%,成功率曲线的AUC值分别为0.822和0.849,表明自组织特征映射网络-随机森林模型具有更高的预测率和成功率, 自组织特征映射网络聚类的预测精度虽然有限,但作为非滑坡样本的选择方法,能够有效提高随机森林模型的评价精度。     

基于SOFM神经网络的边坡稳定性评价

针对边坡工程稳定性分析中参数的不确定性,在分析自组织特征映射神经网络（SOFM）基本学习算法的基础上,从提高算法收敛速度和性能出发,将自组织特征映射神经网络基本学习算法加以改进,据此建立了评价边坡稳定状态的SOFM神经网络模型。然后用收集到的边坡稳定工程实例作为样本,对该模型进行训练和检验,并与BP神经网络判别结果对比。结果表明,SOFM神经网络性能良好、预测精度高,是边坡稳定性评价的一种有效方法。     

数字地貌形态分类优化方法

地貌形态特征分类对生态环境、水文研究及地质构造分析等地学研究具有重要意义,已成为现代地貌学的一个研究热点。利用高分辨率DEM数据,以平均坡度、截面曲率、最大曲率和最小曲率等地貌形态因子为参数,构建地貌形态分类模型;基于自组织映射神经网络提出数字地貌形态分类数学优化方法,对长春地区进行地貌分类。该方法实现自动聚类分析,获得了10种地貌类型阈值,达到地貌形态特征分类要求。这种方法突破了形态参数分类方法6种地貌类型的限制,分类种类达到10种;通过SOM聚类分析,自动获取阈值,从而减少了人为主观因素的干扰。地貌形态分类取得较好效果。     

Modelling the mechanical behaviour of unsaturated soils using a genetic algorithm-based neural network

Modelling the mechanical behaviour of unsaturated soils has been the subject of many research works in the past few decades. A number of constitutive models have been developed to describe the complex behaviour of unsaturated soils. Despite the significant advances in the constitutive theories for unsaturated soils, none of the existing models can completely describe the various aspects of the real behaviour of unsaturated soils. In this paper, a new unified approach is presented, based on the integration of a neural network and a genetic algorithm, for the modelling of unsaturated soils. In the proposed approach, a genetic algorithm was used to optimise the weights of the neural network. A three-layer sequential architecture was chosen for the neural network. The network had eight input neurons, five neurons in the hidden layer and three neurons in the output layer. The eight input neurons represented the initial gravimetric water content, initial dry density, degree of saturation, net mean stress with respect to pore-air pressure, axial strain, deviatoric stress, soil suction and volumetric strain, and the three neurons in the output layer represented the deviatoric stress, suction and volumetric strain at the end of each increment. The network was trained and tested using a database that included results from a comprehensive set of triaxial tests on unsaturated soils from the literature. The predictions of the proposed model were compared with the experimental results. The comparison of the results indicates that the proposed approach was accurate and robust in representing the mechanical behaviour of unsaturated soils.     

Flow estimations for the Sohu Stream using artificial neural networks

In this study, daily rainfall–runoff relationships for Sohu Stream were modelled using an artificial neural network (ANN) method by including the feed-forward back-propagation method. The ANN part was divided into two stages. During the first stage, current flows were estimated by using previously measured flow data. The best network architecture was found to utilise two neurons in the input layer (the delayed flows from the first and second days), two hidden layers, and one output layer (the current flow). The coefficient of determination (R ²) in this architecture was 81.4%. During the second stage, the current flows were estimated by using a combination of previously measured values for precipitation, temperature, and flows. The best architecture consisted of an input layer of 2 days of delayed precipitation, 3 days of delayed flows, and temperature of the current. The R ² in this architecture was calculated to be 85.5%. The results of the second stage best reflected the real-world situation because they accounted for more input variables. In all models, the variables with the highest R ² ranked as the previous flow (81.4%), previous precipitation (21.7%), and temperature.     

深度学习在煤层气测井解释中的应用研究

将常规储层测井解释方法应用于煤层气储层测井解释,其效果存在一定的折扣。为了改善传统方法在煤层气测井解释中出现的问题,将深度学习的思想引入测井解释,提出受限玻尔兹曼机的数量、受限玻尔兹曼机隐含层神经元数量、分类阈值的确定方法,利用深度信念网络进行煤层识别及煤层气含气量的预测。实验结果表明：首先,在交会图法效果不好的情况下,通过深度信念网络进行煤层识别,继而对识别结果进行适当校正,煤层识别成功率可达到90%以上;其次,经过多种方法的对比,利用深度信念网络进行煤层气含气量预测的效果,要好于BP神经网络、多元回归统计以及Langmuir方程三种方法。深度学习改进了传统的BP神经网络,具备更强的复杂函数泛化能力,适用于煤层气测井解释,并具有进一步的推广价值。     

Prediction of agriculture derived groundwater nitrate distribution in North China Plain with GIS-based BPNN

In recent years, nitrate contamination of groundwater has become a growing concern for people in rural areas in North China Plain (NCP) where groundwater is used as drinking water. The objective of this study was to simulate agriculture derived groundwater nitrate pollution patterns with artificial neural network (ANN), which has been proved to be an effective tool for prediction in many branches of hydrology when data are not sufficient to understand the physical process of the systems but relative accurate predictions is needed. In our study, a back propagation neural network (BPNN) was developed to simulate spatial distribution of NO₃-N concentrations in groundwater with land use information and site-specific hydrogeological properties in Huantai County, a typical agriculture dominated region of NCP. Geographic information system (GIS) tools were used in preparing and processing input–output vectors data for the BPNN. The circular buffer zones centered on the sampling wells were designated so as to consider the nitrate contamination of groundwater due to neighboring field. The result showed that the GIS-based BPNN simulated groundwater NO₃-N concentration efficiently and captured the general trend of groundwater nitrate pollution patterns. The optimal result was obtained with a learning rate of 0.02, a 4-7-1 architecture and a buffer zone radius of 400 m. Nitrogen budget combined with GIS-based BPNN can serve as a cost-effective tool for prediction and management of groundwater nitrate pollution in an agriculture dominated regions in North China Plain.     

Three-dimensional multilayer cylindrical tunnel model for calculating train-induced dynamic stress in saturated soils

This study proposes an improved tunnel model for evaluating train-induced dynamic stress in saturated soils, which can consider multiple moving loads, grouting layer and pore-water pressure. Using Shanghai Metro’s actual parameters for train speed, tunnel, grouting layer and soils, the analysis of the spatial distribution of dynamic stress for soils and stress state of various locations under moving train loads shows that neglecting effects such as pore-water pressure can lead to underestimating dynamic normal stress and overestimating dynamic shear stress in the soils below tunnel. This model can be further extended to investigate principal stress axes rotations and tunnel settlement.     

全球变化研究中的数据融合

文中给出了一种研究全球变化中数据融合的新方法———粗集神经网络结构。用改进的方法训练网络权值 ,分析和仿真结果表明 ,此种模型具有较好的泛化、学习、映射能力。说明新模型能够解决传感器输出为二值或一个范围的多传感器数据融合问题。     

用遗传神经网络分析泥石流活动性

泥石流是我国山区的主要地质灾害之一。影响泥石流活动性的因素十分复杂，并且具有随机性和模糊性。遗传神经网络结合了神经网络和遗传算法的优点，可以模拟学习和进化之间的交互作用，很适合用于分析泥石流活动性。文章简要讨论了遗传神经网络的原理，建立了泥石流活动性分析的遗传神经网络模型，并将该模型用于川藏公路沿线30条泥石流沟的活动性分析。网络的拓扑结构为(9，6，4，3)，即输入节点(评价指标)、第l隐含层、第2隐含层和输出接点(分析结果)分别为9、6、4、3。首先以其中25条泥石流沟作为样本对网络进行训练，训练时网络的连接权采用遗传算法进行自适应演化，待模型稳定后将其余5条泥石流沟的数据输入模型，计算它们的活动性，计算结果与实际观测基本相符，证明模型是可行的，各个参数的选取也是合适的。     

基于自组织神经网络的砂土液化评价

在分析自组织特征映射(SOFM)神经网络基本学习算法的基础上,从提高算法收敛速度和性能出发,提出了一种改进算法:根据实际应用并结合专家经验确定初始连接权值;采用高斯函数作为拓扑邻域函数;将算法分为粗调整和细调整两个阶段,分别采用不同的学习率和邻域函数,然后采用改进后的SOFM算法对砂土液化进行评价。实例研究表明,应用SOFM神经网络评价砂土液化高效可行,为砂土液化评价提供了新方法。     

边坡稳定性预测的模糊神经网络模型

根据边坡稳定问题具有的模糊性,提出了一种判定边坡稳定性的模糊神经网络模型。该系统仅从期望输入输出数据集即可达到获取知识、确定模糊初始规则基的目的。再利用神经网络学习能力便不难修改规则库中的模糊规则以及隶属函数和网络权值等参数,这样大大减少了规则匹配过程,加快了推理速度,从而极大程度地提高了系统的自适应能力。最后用收集到的边坡数据样本训练和测试模糊神经网络模型,结果表明该模糊神经网络预测边坡稳定性是可行的、有效的。     

利用土壤地球化学数据和BP神经网络预测松嫩平原油气资源

基于东北地区多目标区域地球化学调查获得的海量土壤地球化学数据,利用BP神经网络模型,在土壤地球化学性质与油气田空间位置之间建立模型,构造最优的油气资源预测模型. 以土壤54项地球化学指标以及XY坐标值共同作为模型输入层,以样本是否在油气田内(1代表油气田内,0代表油气田外)作为模型输出层,基于随机抽取的油气田内和油气田外各500个土壤样本数据进行模型训练. 结果显示,多次训练后识别准确率保持在90%左右,说明该模型分类效果较好,可用于油气资源预测. 利用该模型获得了松嫩平原11 291个土壤样本的含油气概率,并绘制了油气资源预测图. 研究表明,神经网络对于解决复杂的非线性地质问题可以发挥重要作用.     

Inversion of quasi-3D DC resistivity imaging data using artificial neural networks

The objective of this paper is to investigate the applicability of artificial neural networks in inverting quasi-3D DC resistivity imaging data. An electrical resistivity imaging survey was carried out along seven parallel lines using a dipole-dipole array to confirm the validation of the results of an inversion using an artificial neural network technique. The model used to produce synthetic data to train the artificial neural network was a homogeneous medium of 100Ωm resistivity with an embedded anomalous body of 1000Ωm resistivity. The network was trained using 21 datasets (comprising 12159 data points) and tested on another 11 synthetic datasets (comprising 6369 data points) and on real field data. Another 24 test datasets (comprising 13896 data points) consisting of different resistivities for the background and the anomalous bodies were used in order to test the interpolation and extrapolation of network properties. Different learning paradigms were tried in the training process of the neural network, with the resilient propagation paradigm being the most efficient. The number of nodes, hidden layers, and efficient values for learning rate and momentum coefficient have been studied. Although a significant correlation between results of the neural network and the conventional robust inversion technique was found, the ANN results show more details of the subsurface structure, and the RMS misfits for the results of the neural network are less than seen with conventional methods. The interpreted results show that the trained network was able to invert quasi-3D electrical resistivity imaging data obtained by dipole-dipole configuration both rapidly and accurately.     

基于粗糙神经网络的坡面雨滴溅蚀量研究

提出采用粗糙神经网络预测坡面雨滴溅蚀量。用粗糙集方法中条件属性与决策属性相对依赖的概念约简某雨滴溅蚀量试验中的冗余信息,去掉了坡度、雨强、水深、单宽流量4个试验指标中水深和单宽流量两个指标,建立了以坡度、雨强为输入,溅蚀量为输出的2−5−1的粗糙神经网络模型,简化了神经网络的结构,减少了网络的训练时间。实例计算中信息约简后预测值与试验值线性回归的相关系数大于未约简时的相关系数值,计算速度也有所提高。实例计算表明,粗糙神经网络为坡面雨滴溅蚀量预测研究提供了一种有效可行的算法。     

基于深度学习的智能矿物识别方法研究

矿物识别在许多研究领域都有着重要作用,基于深度学习技术的智能矿物识别为这些领域带来了新的发展方向,不仅能有效节省人工成本,还能减小识别错误。针对石英、角闪石、黑云母、石榴石和橄榄石共5种矿物进行实验,提出了一种准确高效的智能矿物识别方法。实验采用图像分析常用的卷积神经网络建立模型,设计出一套基于残差神经网络的矿物识别方法。本实验独立采集了5种矿物的偏光显微图像数据集,用于模型的训练、验证和测试,并通过合理的数据增强策略来扩充训练数据集。在卷积神经网络的结构设计上,选取了ResNet-18作为框架,最终于模型测试中取得89%的准确率,成功训练出一个较为精准的矿物识别模型,实现了基于深度学习的智能矿物识别方法。