神经网络分析方法在瓦斯预测中的应用

论述了瓦斯预测技术的研究现状及其在现代矿业中面临的新问题，介绍了神经网络技术在处理复杂地质条件方面的优越性，探讨了瓦斯预测技术与人工神经网络等高新技术相结合的可能性与必要性，并举例论证了它们在瓦斯预测过程中的适用性。实践证明：瓦斯预测技术与人工神经网络相结合所建立的预测模型，不仅能够综合考虑各种影 响因素，较好地处理地质条件中的各种非线性关系，而且预测精度高，结论可靠，为瓦斯预测技术的进一步发展提供了新的思路。     

Comparison of static-feedforward and dynamic-feedback neural networks for rainfall–runoff modeling

A systematic comparison of two basic types of neural network, static and dynamic, is presented in this study. Two back-propagation (BP) learning optimization algorithms, the standard BP and conjugate gradient (CG) method, are used for the static network, and the real-time recurrent learning (RTRL) algorithm is used for the dynamic-feedback network. Twenty-three storm-events, about 1632 rainfall and runoff data sets, of the Lan-Yang River in Taiwan are used to demonstrate the efficiency and practicability of the neural networks for one hour ahead streamflow forecasting. In a comparison of searching algorithms for a static network, the results show that the CG method is superior to the standard BP method in terms of the efficiency and effectiveness of the constructed network's performance. For a comparison of the static neural network using the CG algorithm with the dynamic neural network using RTRL, the results show that (1) the static-feedforward neural network could produce satisfactory results only when there is a sufficient and adequate training data set, (2) the dynamic neural network generally could produce better and more stable flow forecasting than the static network, and (3) the RTRL algorithm helps to continually update the dynamic network for learning—this feature is especially important for the extraordinary time-varying characteristics of rainfall–runoff processes.     

Data generation for shear modulus and damping ratio in reinforced sands using adaptive neuro-fuzzy inference system

Neuro-fuzzy inference systems have been used in many areas in civil engineering applications. This study was conducted to estimate low strain dynamic properties of composite media from easily measurable physical properties using the adaptive neuro-fuzzy inference system (ANFIS). The inference system was employed to predict the shear modulus and the damping coefficient of the sand samples as an alternative to lengthy laboratory testing. ANFIS was trained using low strain dynamic test results of samples of sand reinforced with particulate rubber inclusions from a resonant column device. The training was performed with an improved hybrid method, which was found to deliver better results than classical back-propagation method such as multi-layer perceptron (MLP) and multiple regression analysis method (MRM). Using the new approach, the optimal precise value of a parameter could be estimated within the constraints of the experimental design. The ANFIS model has appeared very effective in modeling complex soil properties such as shear modulus and damping coefficient, and performs better than MLP and MRM.     

地震预报专家系统ESEP 3.0

近期研制成功的地震预报专家系统ESEP 3．0将模糊系统、神经网络与专家系统技术相结合，引入了驾驭式的推理机制，除具有第一代专家系统的符号推理与解释功能、以及第二代专家系统的学习功能外，还具有较强的人机交互能力，是一个全新的专家系统。ESEP 3．0由知识编辑、机器学习、驾驭式模糊推理机和解释等4个子系统组成，本文介绍了系统的组成和概况。     

遗传神经网络测井解释技术在宝北区块的应用初探

将遗传算法与人工神经网络(BP网络)的原理结合,对BP网络进行了改进,提高了收敛速度,并防止陷入局部极小。应用改进后的BP算法,建立了针对测井解释的神经网络模型,并应用此模型定量计算了宝北区块的多口井的渗透率值,其解释结果及精度均令人满意。     

Displacement analysis of tunnel support in soft rock around a shallow highway tunnel at Golovec

Within the last 10 years Slovenia has been constructing its highway network. The Golovec tunnel, as a part of Slovenia's capital ring is thus one of the most important connections of Ljubljana to the east and to the north. It is a double tube three-lane tunnel in soft rock with small to medium overburden. Its construction, following NATM, caused huge problems to all parties involved. The tunnel support was well monitored during its construction, which gave the authors a good opportunity to analyse the results.The Golovec tunnel is constructed through one of few hills surrounding Ljubljana, of Carboniferous age, consisting of clastic rock: siltstone, claystone and sandstone. Golovec hill belongs to the first of two overthrusting zones from this area, so the rock is strongly faulted.Tunnel monitoring consisted of daily 3-D tunnel tube displacement measurements in 97 measuring sections, and of two measuring sections within the tunnel with more complex measuring equipment, to monitor stress changes and rock deformations around both tunnel tubes. Monitoring of the surface 3-D movements gave us the opportunity to study the influence of the tunnel construction on the surface above it. The tunnel, its geology, construction procedure and monitoring results are described in the first part of the paper.The second part consists of the interpretation of monitored results, with an emphasis on results concerning development and evolution of the excavation-damaged zone in the rock around the tunnel. Back-calculations, performed as a basis for the interpretation procedure, are also presented in this part. Calculations of the propagation of the tunnel destressed zone and stress-field around the tunnel, up to the surface, were performed by means of numerical model with the finite difference method. The evolution of tunnel displacements and their prediction was based on the use of Back Propagation Neural Networks, whose principles are presented in one chapter of this paper. Results showed that the most important, for the final settlement at the surface above the tunnel, was the time of installation and rigidity of the primary support. On the basis of the calculated final displacements, this support could easily be strengthened in a short time, when necessary.     

Intelligent determination of peak ground motion parameters based on intensity assessed by artificial neural network

Artificial Neural Network (ANN) model of computation based on mathematical model of neural processes is applied to establish an intelligent computing network from seismic intensity to peak ground parameter instead of the conventional statistical relationship in this paper. For a give seismic intensity rating, the network formed with actual strong ground motion records directly produces the corresponding peak ground parameters and the effects of earthquake magnitude and epicentral distance are included. The computed results of the network trained with a number of strong motion records in the West America show that such networks have obtained good conversion relationship from seismic intensity to peak ground parameters. The Chinese version of this paper appeared in the Chinese edition ofActa Seismologica Sinica,15, 208–216, 1993.     

Prediction model for occurrence of impact wave force

This study investigates the applicability of neural networks to predict whether impact wave force will act on the upright section of a composite breakwater. We employ a three-layered neural network whose units of input layer are h/L, H/h, d/h and B_M/h (h: the total water depth; L: the wavelength; H: the wave height; d: the water depth above the mound; B_M: the horizontal distance from the shoulder of mound to the caisson). Teach signals are 0.99 and 0.01 according to the cases of occurrence and absence of impact wave force, respectively. The neural network whose parameters are determined through self-learning can accurately predict whether impact wave force occurs.     

Application of spectral decomposition and neural networks to characterise deep turbidite systems in the outer fold and thrust belt of the Niger Delta

We have applied a wavelet‐based spectral decomposition scheme and a multi‐layered feed‐forward neural network to interpret turbidite depositional systems from three‐dimensional reflection seismic data and well logs for a prospective hydrocarbon zone in the outer fold and thrust belt of the Niger Delta. The goal was to overcome difficulties in interpreting depositional systems from deep sections of the Field, occasioned by loss of seismic resolution with depth and the sparse distribution of wells. The decomposition scheme allowed us to delineate multiple depositional systems not apparent on the conventional seismic amplitude display. These systems include linear channel systems with terminal splay lobes, a sinuous channel system and its abandoned meander loops, and sediment wave features in overbank areas. Delineated channel morphologies and transport directions varied both laterally and vertically and were possibly dependent upon the disposition of the pre‐thrusting paleo‐seafloor. Terminal splay lobes are fragmented and coincident with the locations of topographic lows, which are possibly related to the initial configurations of the oceanic basement below. Predicted porosity and resistivity distributions have morphologies that correlate well with the mapping provided by the spectral decomposition scheme. The property distributions indicate that reservoir prone systems in the Field and possibly within the outer fold and thrust belt are composed primarily of channel systems, both linear and sinuous, and their associated splay lobes. The channel systems appear vertically stacked, and this situation possibly increases the potential success rate for exploration wells in the region. Beyond channel limits, redistributive bottom currents varying rapidly in speed and direction apparently encouraged the dispersal of sand‐rich sediments to form sediment waves. Despite the limited well control, the methodology significantly aided our interpretation. It proved effective at revealing the distribution of reservoir prone facies within the Field and provided insight into the dominant factors that controlled deposition within the Field.     

A hybrid machine learning ensemble approach based on a Radial Basis Function neural network and Rotation Forest for landslide susceptibility modeling: A case study in the Himalayan area,India

In this paper, a hybrid machine learning ensemble approach namely the Rotation Forest based Radial Basis Function (RFRBF) neural network is proposed for spatial prediction of landslides in part of the Himalayan area (India). The proposed approach is an integration of the Radial Basis Function (RBF) neural network classifier and Rotation Forest ensemble, which are state-of-the art machine learning algorithms for classification problems. For this purpose, a spatial database of the study area was established that consists of 930 landslide locations and fifteen influencing parameters (slope angle, road density, curvature, land use, distance to road, plan curvature, lineament density, distance to lineaments, rainfall, distance to river, profile curvature, elevation, slope aspect, river density, and soil type). Using the database, training and validation datasets were generated for constructing and validating the model. Performance of the model was assessed using the Receiver Operating Characteristic (ROC) curve, area under the ROC curve (AUC), statistical analysis methods, and the Chi square test. In addition, Logistic Regression (LR), Multi-layer Perceptron Neural Networks (MLP Neural Nets), Naïve Bayes (NB), and the hybrid model of Rotation Forest and Decision Trees (RFDT) were selected for comparison. The results show that the proposed RFRBF model has the highest prediction capability in comparison to the other models (LR, MLP Neural Nets, NB, and RFDT); therefore, the proposed RFRBF model is promising and should be used as an alternative technique for landslide susceptibility modeling.