用统计神经网络进行结构损伤存在检测

用新奇检测的方法来检测结构损伤的存在 ,因其不需要建立结构的计算模型 ,能适合于大型复杂结构的损伤检测。然而 ,目前采用的新奇检测方法—自联想记忆神经网络方法 ,当所用的测试数据具有不同噪声水平或为非正态分布时 ,可能会得出错误的结果。为此本文提出了一种新的方法—统计神经网络方法用于结构的损伤存在检测 ,并用“可能性”来描述结构损伤的存在。通过一个两层框架的数值模拟和一个简支梁的实验数据分析 ,证明统计神经网络可以用来检测结构的损伤存在 。

结构损伤存在检测的两种神经网络的比较研究

目前采用的新奇检测方法———自联想记忆神经网络方法,当所用的测试数据具有不同噪声水平或为非正态分布时,会得出错误的结果.为此,提出了一种新的方法———统计神经网络方法,用于结构的损伤存在检测,并用"可能性"来描述结构损伤的存在.通过一个两层框架的数值模拟和一个简支梁的实验数据进行对比性研究表明,统计神经网络可以用来检测结构的损伤存在,具有比自联想记忆神经网络更好的检测效果.

以六种疾病为例研究基于统计注意力的神经网络模型在证名诊断中的应用

目的研究基于统计注意力的神经网络(SANN)模型在中医证名诊断中的适用性与先进性,探讨其生成的特征贡献度是否符合中医原理。方法选择记载于古今医案云平台及中医药杏林园数据库的高血脂、更年期综合征、冠心病、慢性胃炎、慢性肾炎、尿路感染、脂肪肝病案共1 110例。通过人工神经网络(ANN)、随机森林(RF)、支持向量机(SVC)、K-近邻(KNN)、SANN分别建立诊断模型,对比5种模型评价指标。评价指标包括Macro-F1、Macro-Precision、Macro-Accuracy、Macro-Recall。结果 SANN在6种疾病中的Macro-F1平均值为0.78、Macro-Precision平均值为0.79、Macro-Accuracy平均值为0.79、Macro-Recall平均值为0.8,均优于其他4种基准模型,其参数可解释性与导出的特征对类支持度符合中医原理。结论 SANN在中医证名诊断智能化、中医数据的特征筛选、疾病量表研制等任务中具有适用性与先进性,为相关工作提供了创新性的方法参考。

混凝土拱坝裂缝损伤检测振动法的试验研究

通过一个假想混凝土拱坝的模型试验对混凝土拱坝裂缝损伤检测的振动法进行了研究.试验模拟了一半坝厚深度、3/4坝厚深度和贯穿全部坝厚3种深度的裂缝损伤,通过振动模态测试研究了裂缝损伤对拱坝固有频率的影响,当拱坝出现一半坝厚深度裂缝时,其固有频率就出现了明显的下降,频率下降的规律与数值模拟结果基本一致;基于实测频率数据,采用统计神经网络(SNN)可以检测出一半坝厚深度和3/4坝厚深度裂缝损伤的存在;针对两种损伤指标,利用实测模态振型数据指示裂缝损伤位置的效果进行了比较,结果表明,基于模态应变能变化的损伤指标能够指示出未贯穿混凝土拱坝坝厚裂缝损伤的位置.

Application of neural network merging model in dam deformation analysis

In order to improve the prediction accuracy and test the generalization ability of the dam deformation analysis model, the back-propagation（BP） neural network model for dam deformation analysis is studied, and the merging model is built based on the neural network BP algorithm and the traditional statistical model. The three models mentioned above are calculated and analyzed according to the long-term deformation observation data in Chencun Dam. The analytical results show that the average prediction accuracies of the statistical model and the BP neural network model are ~ 0.477 and ＋- 0.390 mm, respectively, while the prediction accuracy of the merging model is ~0. 318 mm, which is improved by 33% and 18% compared to the other two models, respectively. And the merging model has a better generalization ability and broad applicability.

DESIGN OF NONLINEAR OBSERVER FOR NONLINEAR SYSTEM BASED ON RBF NEURAL NETWORKS

A new type of nonlinear observer for nonlinear systems is presented. Instead of approximating thc cntire nonlinear system with the neural network （NN）, only the un-modeled part left over after the lincarization is approximated. Compared with the conventional linear observer, the observer provides more accurate estimation of the state. The state estimation error is proved to asymptotically approach zero with the Lyapunov method. The simulation result shows that the proposed observer scheme is effective and has a potential application ability in the fault detection and identification （FDI）, and the state estimation.

Artificial neural network based inverse design method for circular sliding slopes

Current design method for circular sliding slopes is not so reasonable that it often results in slope (sliding.) As a result, artificial neural network (ANN) is used to establish an artificial neural network based inverse design method for circular sliding slopes. A sample set containing 21 successful circular sliding slopes excavated in the past is used to train the network. A test sample of 3 successful circular sliding slopes excavated in the past is used to test the trained network. The test results show that the ANN based inverse design method is valid and can be applied to the design of circular sliding slopes.

A neurofuzzy system based on rough set theory and genetic algorithm

This paper presents a hybrid soft computing modeling approach for a neurofuzzy system based on rough set theory and the genetic algorithms (NFRSGA). The fundamental problem of a neurofuzzy system is that when the input dimension increases, the fuzzy rule base increases exponentially. This leads to a huge infrastructure network which results in slow convergence. To solve this problem, rough set theory is used to obtain the reductive rules, which are used as fuzzy rules of the fuzzy system. The number of rules decrease, and each rule does not need all the conditional attribute values. This results in a reduced, or not fully connected, neural network. The structure of the neural network is relatively small and thus the weights to be trained decrease. The genetic algorithm is used to search the optimal discretization of the continuous attributes. The NFRSGA approach has been applied in the practical application of building a soft sensor model for estimating the freezing point of the light diesel fuel in a Fluid Catalytic Cracking Unit (FCCU), and satisfying results are obtained.

A nonlinear PCA algorithm based on RBF neural networks

Traditional PCA is a linear method, but most engineering problems are nonlinear. Using the linear PCA in nonlinear problems may bring distorted and misleading results. Therefore, an approach of nonlinear principal component analysis (NLPCA) using radial basis function (RBF) neural network is developed in this paper. The orthogonal least squares (OLS) algorithm is used to train the RBF neural network. This method improves the training speed and prevents it from being trapped in local optimization. Results of two experiments show that this NLPCA method can effectively capture nonlinear correlation of nonlinear complex data, and improve the precision of the classification and the prediction.

Multi-agent reinforcement learning using modular neural network Q-learning algorithms

Reinforcement learning is an excellent approach which is used in artificial intelligence,automatic control, etc. However, ordinary reinforcement learning algorithm, such as Q-learning with lookup table cannot cope with extremely complex and dynamic environment due to the huge state space. To reduce the state space, modular neural network Q-learning algorithm is proposed, which combines Q-learning algorithm with neural network and module method. Forward feedback neural network, Elman neural network and radius-basis neural network are separately employed to construct such algorithm. It is revealed that Elman neural network Q-learning algorithm has the best performance under the condition that the same neural network training method, i.e. gradient descent error back-propagation algorithm is applied.

Application of artificial neural networks and multivariate statistics to estimate UCS using textural characteristics

Before any rock engineering project,mechanical parameters of rocks such as uniaxial compressive strength and young modulus of intact rock get measured using laboratory or in-situ tests,but in some situations preparing the required specimens is impossible.By this time,several models have been established to evaluate UCS and E from rock substantial properties.Artificial neural networks are powerful tools which are employed to establish predictive models and results have shown the priority of this technique compared to classic statistical techniques.In this paper,ANN and multivariate statistical models considering rock textural characteristics have been established to estimate UCS of rock and to validate the responses of the established models,they were compared with laboratory results.For this purpose a data set for 44 samples of sandstone was prepared and for each sample some textural characteristics such as void,mineral content and grain size as well as UCS were determined.To select the best predictors as inputs of the UCS models,this data set was subjected to statistical analyses comprising basic descriptive statistics,bivariate correlation,curve fitting and principal component analyses.Results of such analyses have shown that void,ferroan calcitic cement,argillaceous cement and mica percentage have the most effect on USC.Two predictive models for UCS were developed using these variables by ANN and linear multivariate regression.Results have shown that by using simple textural characteristics such as mineral content,cement type and void,strength of studied sandstone can be estimated with acceptable accuracy.ANN and multivariate statistical UCS models,revealed responses with 0.87 and 0.76 regressions,respectively which proves higher potential of ANN model for predicting UCS compared to classic statistical models.

Distribution network planning algorithm based on Hopfield neural network

This paper presents a new algorithm based on Hopfield neural network to find the optimal solution for an electric distribution network. This algorithm transforms the distribution power network-planning problem into a directed graph-planning problem. The Hopfield neural network is designed to decide the in-degree of each node and is in combined application with an energy function. The new algorithm doesn’t need to code city streets and normalize data, so the program is easier to be realized. A case study applying the method to a district of 29 street proved that an optimal solution for the planning of such a power system could be obtained by only 26 iterations. The energy function and algorithm developed in this work have the following advantages over many existing algorithms for electric distribution network planning: fast convergence and unnecessary to code all possible lines.

Wear Debris Identification Using Feature Extraction and Neural Network

A method and results of identification of wear debris using their morphological features are presented. The color images of wear debris were used as initial data. Each particle was characterized by a set of numerical parameters combined by its shape, color and surface texture features through a computer vision system. Those features were used as input vector of artificial neural network for wear debris identification. A radius basis function (RBF) network based model suitable for wear debris recognition was established, and its algorithm was presented in detail. Compared with traditional recognition methods, the RBF network model is faster in convergence, and higher in accuracy.

A Fuzzy Neural Network Model of Linguistic Dynamic Systems Based on Computing with Words

Linguistic dynamic systems(LDS)are dynamic processes involving computing with words(CW)for modeling and analysis of complex systems.In this paper,a fuzzy neural network(FNN)structure of LDS was proposed.In addition,an improved nonlinear particle swarm optimization was employed for training FNN.The experiment results on logistics formulation demonstrates the feasibility and the efficiency of this FNN model.

Experimental study of fatigue degree quantification for multi-feature fusion identification

A comprehensive quantification method of fatigue degree is proposed concerning subjective and objective quantifications.Using the fatigue degree test software,fatigue degree is objectively quantified by analyzing the reaction and operation abilities of drivers about traffic signals.By comparison experiment with that EEG signal based,multivariate statistical analysis and fusion identification based on BP neural network(BPNN) results show that the experimental procedure is simple and practical,and the proposed method can reveal the correlation between fatigue feature parameters and fatigue degree in theory,and also can achieve accurate and reliable quantification of fatigue degree,especially under the associated action of multiple fatigue feature parameters.

Hybrid Features for an Arabic Word Recognition System

This research proposes and implements an Arabic Sub-Words Recognition System （ASWR）. The system focuses on employing a combination of statistical and structural features to provide complete pattern＇s description and enhances the recognition rate. Support Vector Machines （SVMs） is utilized as a promising pattern recognition tool. In addition to that, the problems of dots and holes are solved in a completely different way from the ones previously employed. The proposed system proceeds in several phases as follows： （1） image acquisition, （2） binarisation, （3） morphological processing, （4） feature extraction, which includes statistical features, i.e., moment invariants, and structural features, i.e., dot number, dot position, and number of holes, features, and （5） classification, using multi-class SVMs and applying a one-against-all technique. The proposed system has been tested using different sets of words and subwords and has achieved a nearly 98.90% recogiaition rate. Comparative results with NNs are also presented.

On Hybrid Computer-Aided Tools to Interface Design Data

Recognizing the drawbacks of stand-alone computer-aided tools in engineering, several hybrid systems are suggested with varying degree of success. In transforming the design concept to a finished product, in particular, smooth interfacing of the design data is crucial to reduce product cost and time to market. Having a product model that contains the complete product description and computer-aided tools that can understand each other are the primary requirements to achieve the interfacing goal. This article discusses the development methodology of hybrid engineering software systems with particular focus on application of soft computing tools such as genetic algorithms and neural networks. Forms of hybridization options are discussed and the applications are elaborated using two case studies. The forefront aims to develop hybrid systems that combine the strong side of each tool, such as, the learning, pattern recognition and classification power of neural networks with the powerful capacity of genetic algorithms in global search and optimization. While most optimization tasks need a certain form of model, there are many processes in the mechanical engineering field that are difficult to model using conventional modeling techniques. The proposed hybrid system solves such difficult-to-model processes and contributes to the effort of smooth interfacing design data to other downstream processes.

基于人工智能的不孕症中医辨证模型的构建与应用

目的:建立基于人工智能的不孕症中医辨证模型,为不孕症中医智能辨证模型的构建与应用提供方法和依据。方法:检索中国期刊全文数据库、万方期刊数据库、中医智库、古今病案云平台,收集关于不孕症的中医名医病案300例,建立不孕症病案中医信息数据库,采用经过超参数调优的随机森林(RF)、支持向量分类(SVC)、K-近邻(KNN)、人工神经网络(ANN)及统计学注意力神经网络模型(SANN)对数据集进行量化分析,建立不孕症辨证模型,采用五折交叉验证对模型进行评价,评价指标包括Accuracy、Precision、Recall、F1。结果:不孕症中医四诊信息为输入变量共86项,输出变量为不孕症中医证型共6项。5种模型的拟合效果较好,Accuracy、Precision、Recall、F1值均在0.77以上;其中SANN模型的准确率、查准率与查全率最高,Accuracy、Precision、Recall、F1分别为0.90、0.90、0.91、0.90,均高于其他算法模型,其参数的中医解释基本符合中医诊断原理。结论:基于SANN算法模型建立的不孕症中医辨证模型具有良好的诊断能力,将人工智能应用于不孕症中医辨证模型的构建与临床应用方法学可行,且具有较高的准确率。

Distribution Assessment and Source Identification Using Multivariate Statistical Analyses and Artificial Neutral Networks for Trace Elements in Agricultural Soils in Xinzhou of Shanxi Province, China

Multivariate statistical analyses were used to assess the contents and distributions of trace elements in agricultural soils in Xinzhou of Shanxi Province, China, and to identify their sources. Samples with high levels of trace elements were concentrated in eastern Xinzhou, with contents declining from the east to west. Principal component and redundancy analyses revealed strong correlations among Co, Cu, Mn, Ni, Se, V, and Zn contents, suggesting that these elements were derived from similar parent materials. There were also strong correlations between the contents of these elements and soil properties. Contents of Cd and Pb were significantly higher in the agricultural soil samples than in the background soil samples(P < 0.05), and were higher in areas with higher levels of gross domestic product but decreased with distance to the nearest road. Therefore, human activities appear to have a strong influence on the Cd and Pb distribution patterns. A novel artificial neural network(ANN) model, using environmental input data, was used to predict the soil Cd and Pb contents of specified test dates. The performances of the ANN model and a traditional multilinear model were compared. The ANN model could successfully predict Cd and Pb content distributions, projecting that soil Cd and Pb contents will increase by 128% and 25%, respectively, by 2020. The results thus indicated that the economic condition of an area has a greater effect on trace element contents and distributions in the soil than the scale of the economy itself.

FORECASTING EXCHANGE RATES: AN OPTIMAL APPROACH

This paper looks at forecasting daily exchange rates for the United Kingdom, European Union, and China. Here, the authors evaluate the forecasting performance of neural networks （NN）, vector singular spectrum analysis （VSSA）, and recurrent singular spectrum analysis （RSSA） for fore casting exchange rates in these countries. The authors find statistically significant evidence based on the RMSE, that both VSSA and RSSA models outperform NN at forecasting the highly unpredictable exchange rates for China. However, the authors find no evidence to suggest any difference between the forecasting accuracy of the three models for UK and EU exchange rates.