神经网络敏感性分析及其在遥感影像分类中的应用

针对基于连接权的神经网络敏感性分析方法中求取敏感性系数的不稳定性,提出一种优化连接权的神经网络敏感性分析方法。首先采用遗传算法根据误差最小化原则对神经网络进行优化,在优化的神经网络模型上进行基于连接权的敏感性分析。以1个数值模拟实例和华盛顿广场地区的遥感图像地物分类为例,验证所提方法的有效性。实验结果表明,所提方法求取输入变量的敏感性系数是稳定有效的,能有效筛选出遥感图像中对分类贡献较大的特征波段,达到降维的同时提高分类精度。     

控制水稻分蘖角度对群体生态特性的影响

控制分蘖角度对群体温度、群体相对湿度、群体CO2浓度和光合有效辐射均会产生一定影响。选用2006年水稻有关研究数据,分析水稻分蘖角度对群体生态特征的影响。结果表明：各生育时期群体温度在07:00－19:00处理大于CK。分蘖高峰期到孕穗期CK白天群体相对湿度大于处理,齐穗期到灌浆期处理群体相对湿度大于CK。群体CO2浓度在拔节期和孕穗期均为处理大于CK,其他生育期差异不显著。光合有效辐射垂直分布是处理前期和后期上层截获的光能均小于CK,群体内消光系数小。控制分蘖角度形成了有较高温度、较低湿度、高CO2浓度和适宜光分布的群体,可为获得高产奠定基础。     

Neural network modelling of planform geometry of headland-bay beaches

The shoreline of beaches in the lee of coastal salients or man-made structures, usually known as headland-bay beaches, has a distinctive curvature; wave fronts curve as a result of wave diffraction at the headland and in turn cause the shoreline to bend. The ensuing curved planform is of great interest both as a peculiar landform and in the context of engineering projects in which it is necessary to predict how a coastal structure will affect the sandy shoreline in its lee. A number of empirical models have been put forward, each based on a specific equation. A novel approach, based on the application of artificial neural networks, is presented in this work. Unlike the conventional method, no particular equation of the planform is embedded in the model. Instead, it is the model itself that learns about the problem from a series of examples of headland-bay beaches (the training set) and thereafter applies this self-acquired knowledge to other cases (the test set) for validation. Twenty-three headland-bay beaches from around the world were selected, of which sixteen and seven make up the training and test sets, respectively. As there is no well-developed theory for deciding upon the most convenient neural network architecture to deal with a particular data set, an experimental study was conducted in which ten different architectures with one and two hidden neuron layers and five training algorithms – 50 different options combining network architecture and training algorithm – were compared. Each of these options was implemented, trained and tested in order to find the best-performing approach for modelling the planform of headland-bay beaches. Finally, the selected neural network model was compared with a state-of-the-art planform model and was shown to outperform it.     

Classification of seismic signals at Villarrica volcano (Chile) using neural networks and genetic algorithms

Each volcano has its own unique seismic activity. The aim of this work is to construct a system able to classify seismic signals for the Villarrica volcano, one of the most active volcanoes in South America. Since seismic signals are the result of particular processes inside the volcano's structure, they can be used to forecast volcanic activity. This paper describes the different kinds of seismic signals recorded at the Villarrica volcano and their significance. Three kind of signals were considered as most representative of this volcano's activity: the long-period, the tremor, and the energetic tremor signals. A classifier is implemented to read the seismic registers at 30-second intervals, extract the most relevant features of each interval, and classify them into one of the three kinds of signals considered as most representative of this particular volcano. To do so, 1033 different kinds of 30-s signals were extracted and classified by a human expert. A feature extraction process was applied to obtain the main characteristics of each of them. This process was developed using criteria which have been shown by others to effectively classify seismic signals, based on the experience of a human expert. The classifier was implemented with a Multi-Layer Perceptron (MLP) artificial neural network whose architecture and training process were optimized by means of a genetic algorithm. This technique searched for the most adequate MLP configuration to improve the classification performance, optimizing the number of hidden neurons, the transfer functions of the neurons, and the training algorithm. The optimization process also performed a feature selection to reduce the number of signal features, optimizing the number of network inputs. The results show that the optimized classifier reaches more than 93% exactitude. identifying the signals of each kind. The amplitude of the signals is the most important feature for its classification, followed by its frequency content. The described methodology can be used to classify more seismic signals to improve the study of the activity of this volcano or to extend the study to other active volcanoes of the region.     

Modeling unusual eruptive behavior of Mt. Etna,Italy, by means of event bush

One of the best-studied volcanoes of the world, Mt. Etna in Sicily, repeatedly exhibits eruptive scenarios that depart from the behavior commonly considered typical for this volcano. Episodes of intense explosive activity, pyroclastic flows, dome growth and cone collapse pose a variety of previously underestimated threats to human lives in the summit area of the volcano. However, retrospective analysis of these events shows that they were likely caused by the same very sets of premises and starting conditions as “normal” eruptions, yet combined in an unexpected, probably unique, way. To cope with such unexpected consequences, we involve an approach of artificial intelligence developed specially for needs of the geosciences, the event bush. Scenarios inferred from the event bush fit the observed ones and allow to foresee other low-probability events that may occur at the volcano. Application of the event bush provides a more impartial vision of volcanic phenomena and may serve as an intermediary between expert knowledge and numerical assessment, e.g., by means of Bayesian Belief Networks.     

地球物理资料群体智能反演(英文)

复杂地球物理资料的反演问题往往是一个求解多参数非线性多极值的最优解问题。而鸟和蚂蚁等群体觅食的过程,正好与寻找地球物理反演最优解的过程相似。基于自然界群体协调寻优的思想,本文提出了交叉学科的群体智能地球物理资料反演方法,并给出了其对应的数学模型。用一个有无限多个局部最优解的已知模型对该类方法进行了试验。然后,将它们应用到了不同的复杂地球物理反演问题中:(1)对噪声敏感的线性问题;(2)非线性和线性同步反演问题;(3)非线性问题。反演结果表明,群体智能反演是可行的。与常规遗传算法和模拟退火法相比,该类方法有收敛速度相对快、收敛精度相对高等优点;与拟牛顿法和列文伯格一马夸特法相比,该类方法有能跳出局部最优解等优点。     

Landslide susceptibility mapping using geological data, a DEM from ASTER images and an Artificial Neural Network (ANN)

An efficient and accurate method of generating landslide susceptibility maps is very important to mitigate the loss of properties and lives caused by this type of geological hazard. This study focuses on the development of an accurate and efficient method of data integration, processing and generation of a landslide susceptibility map using an ANN and data from ASTER images. The method contains two major phases. The first phase is the data integration and analysis, and the second is the Artificial Neural Network training and mapping. The data integration and analysis phase involve GIS based statistical analysis relating landslide occurrence to geological and DEM (digital elevation model) derived geomorphological parameters. The parameters include slope, aspect, elevation, geology, density of geological boundaries and distance to the boundaries. This phase determines the geological and geomorphological factors that are significantly correlated with landslide occurrence. The second phase further relates the landslide susceptibility index to the important geological and geomorphological parameters identified in the first phase through ANN training. The trained ANN is then used to generate a landslide susceptibility map. Landslide data from the 2004 Niigata earthquake and a DEM derived from ASTER images were used. The area provided enough landslide data to check the efficiency and accuracy of the developed method. Based on the initial results of the experiment, the developed method is more than 90% accurate in determining the probability of landslide occurrence in a particular area.     

地震和爆破时频域能量分布特征的对比分析

运用小波分析方法计算和分析了CDSN乌鲁木齐台记录的天然地震和人工爆破在时频域的能量分布特征,分析结果表明,在震中距大致相同的范围内,同一台站记录的不同当量的人工爆破具有相似的时频变化规律,而天然地震能量分布却存在较大差别。天然地震与人工爆破的时频域能量分布谱图存在比较明显的差异。     

Daily reservoir inflow forecasting using artificial neural networks with stopped training approach

In this paper, an early stopped training approach (STA) is introduced to train multi-layer feed-forward neural networks (FNN) for real-time reservoir inflow forecasting. The proposed method takes advantage of both Levenberg–Marquardt Backpropagation (LMBP) and cross-validation technique to avoid underfitting or overfitting on FNN training and enhances generalization performance. The methodology is assessed using multivariate hydrological time series from Chute-du-Diable hydrosystem in northern Quebec (Canada). The performance of the model is compared to benchmarks from a statistical model and an operational conceptual model. Since the ultimate goal concerns the real-time forecast accuracy, overall the results show that the proposed method is effective for improving prediction accuracy. Moreover it offers an alternative when dynamic adaptive forecasting is desired.