有教师的线性基本函数前向三层神经网络结构研究

优化选择隐节点数是人们在应用基于误差反传算法的有教师的线性基本函数前向三层神经网络过程中首先遇到的一个十分重要而又困难的问题。本文从国内外大量应用裕列中陪结归纳出了一个初定这种网络隐节点数的经验公式，提出了一种判断所选隐节点数是否多余的具体方法，并从理论上做了详细的推导。     

Approximate similarity retrieval with M-trees

Motivated by the urgent need to improve the efficiency of similarity queries, approximate similarity retrieval is investigated in the environment of a metric tree index called the M-tree. Three different approximation techniques are proposed, which show how to forsake query precision for improved performance. Measures are defined that can quantify the improvements in performance efficiency and the quality of approximations. The proposed approximation techniques are then tested on various synthetic and real-life files. The evidence obtained from the experiments confirms our hypothesis that a high-quality approximated similarity search can be performed at a much lower cost than that needed to obtain the exact results. The proposed approximation techniques are scalable and appear to be independent of the metric used. Extensions of these techniques to the environments of other similarity search indexes are also discussed. Received July 7, 1998 / Accepted October 13, 1998     

Aluminium Foam Sandwich (AFS) Ready for Market Introduction

A novel production process for aluminium foam sandwich panels (AFS) is described. As an example for a serial application of AFS a support for a mobile telescope arm on a small lorry is presented and discussed.     

Least-Squares, Moment-Based, and Hybrid Polynomializations of Drag Forces

In order to carry out nonlinear dynamic analyses of fixed offshore structures in the frequency domain, polynomial approximations of the distributed drag term, u∣u∣, and inundation drag term, ηu∣u∣, of the Morison forces are studied. The methods of least-squares and moment-based approximations are considered for cases with and without current. Numerical results and analytical expressions of the polynomial coefficients are presented for the cubic approximations of u∣u∣ and quartic approximations of ηu∣u∣. The curve shapes, first four central moments, and probability density functions of the different approximations are evaluated and compared with the corresponding exact solutions. For the nonmonotonic inundation drag term with the current effect included, a hybrid polynomialization, based on the least-squares approximation for the odd-order polynomial coefficients and the moment-based approximation for the even-order coefficients, is proposed.     

Modeling Concrete Masonry Walls Subjected to Explosive Loads

Concrete masonry unit walls subjected to blast pressure were analyzed with the finite element method, with the goal of developing a computationally efficient and accurate model. Wall behavior can be grouped into three modes of failure, which correspond to three ranges of blast pressures. Computational results were compared to high-speed video images and debris velocities obtained from experimental data. A parametric analysis was conducted to determine the sensitivity of computed results to critical modeling values. It was found that the model has the ability to replicate experimental results with good agreement. However, it was also found that, without knowledge of actual material properties of the specific wall to be modeled, computational results are not reliable predictors of wall behavior.     

One-Dimensional Numerical Model for Nonuniform Sediment Transport under Unsteady Flows in Channel Networks

In this study, the proposed one-dimensional model simulates the nonequilibrium transport of nonuniform total load under unsteady flow conditions in dendritic channel networks with hydraulic structures. The equations of sediment transport, bed changes, and bed-material sorting are solved in a coupling procedure with a direct solution technique, while still decoupled from the flow model. This coupled model for sediment calculation is more stable and less likely to produce negative values for bed-material gradation than the traditional fully decoupled model. The sediment transport capacity is calculated by one of four formulas, which have taken into consideration the hiding and exposure mechanism of nonuniform sediment transport. The fluvial erosion at bank toes and the mass failure of banks are simulated to complement the modeling of bed morphological changes in channels. The tests in several cases show that the present model is capable of predicting sediment transport, bed changes, and bed-material sorting in various situations, with reasonable accuracy and reliability.     

Hydrologic Impacts due to Changes in Conveyance and Conversion from Flood to Sprinkler Irrigation Practices

Improvements in irrigation efficiency are well documented when changing from flood to sprinkler irrigation methods; however, other impacts to the watershed associated with this change are not well known. The resulting impacts to a river basin hydrology when irrigation and conveyance methods are changed are the focus of this study. In an attempt to improve water application and conveyance efficiencies in the Salt River Basin of western Wyoming, irrigation practices were changed from flood irrigation to sprinkler irrigation beginning in the late 1960s, with completion by the mid-1970s. Based upon a water balance, flow in the Salt River increased an average of 65.62 MCM/year. Return flow timing was also impacted by the conversion to sprinkler irrigation. Flows increased 34% in May and 50% in June, while decreasing 15 and 14% in August and September. These changes may have coincided with decreases in groundwater storage. However, analysis of changes in groundwater levels with time was inconclusive. Surface water total dissolved solids (TDS) appears unaffected by the conversion in irrigation practices, while limited groundwater quality data indicate that TDS values are lower in sprinkler irrigated areas.     

Impaired Higher Order Implicit Sequence Learning on the Verbal Version of the Serial Reaction Time Task in Patients With Parkinson's Disease.

Although neuroimaging studies have strongly implicated basal ganglia involvement in implicit sequence learning, serial reaction time (SRT) studies with Parkinson's disease (PD) patients have yielded mixed results. The present research sought to examine the ability of people with PD to implicitly learn sequences with different sequential structures and to objectively assess explicit knowledge. A version of the SRT task that reduces motor demands was used to compare 19 patients with PD but not dementia and 37 matched controls. PD patients showed less implicit sequence-specific learning for both sequences and reduced response time improvement over sequential trials for the more complex sequence. A closer examination revealed that the deficit involved higher order sequential associations as well as the learning of pairwise information. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Seismic Deformation of Bar Mat Mechanically Stabilized Earth Walls. II: A Multiblock Model

A relatively simple rigid plastic multiblock computational model has been developed to predict the permanent seismic displacement of mechanically stabilized earth (MSE) walls. The model formulation was based on many observations made from a series of centrifuge tests carried out on many different configurations of MSE walls. The proposed model is similar to the sliding block method of Newmark. The approach accounted for the variation in acceleration within the backfill and the nonuniform nature of the permanent wall face deformation. The predictive capability of the proposed model has been verified using centrifuge test results obtained for four MSE walls each subjected to three earthquake excitations with strength varying between 0.48 and 0.9g. The analytical model captures many aspects of the characteristic deformation behavior of MSE walls observed in the centrifuge tests. In each of the eleven wall displacement cases studied, the backfill friction angle that yielded a good match between the computed and measured maximum wall displacement was consistent with the corresponding laboratory measured values.