Catálogo de publicaciones - libros

The business field is one of the important fields where the data mining technology is applied. The study mainly focuses on different attribute object’s quantitative prediction and customer structure’s qualitative prediction. Aiming at the characteristics of time series in business field, such as near-periodicity, non-stationarity and nonlinearity, the wavelet-neural networks-ARMA method is proposed and its application is examined in this paper. The hidden period and the non-stationarity existed in time series are extracted and separated by wavelet transformation. The characteristic of wavelet decomposition series is applied to BP networks and an autoregressive moving average (ARMA) model. The given example elucidates that the forecasting method mentioned in this paper can be employed to the business field successfully and efficiently.

This paper addresses implementation of on-line trained neural network for fast color image segmentation. A pre-selecting technique, based on mean shift algorithm and uniform sampling, is utilized as an initialization tool to largely reduce the training set while preserving the most valuable distribution information. Furthermore, we adopt Particle Swarm Optimization (PSO) to train neural network for a faster convergence and escaping from a local optimum. The results obtained from a wide range of color blood cell images show that under the compatible image segmentation performance on the test set, the training set and running time can be reduced significantly, compared with traditional training methods.

Numerous studies on time series prediction have been undertaken by a lot of researchers. Most of them relate to the construction of structure-invariable system whose parameter values do not change all the time. In fact, the parameter values of many realistic systems are always changing with time. In this case, the embedding theorems are invalid, predicting the behavior of parameter-varying systems is more difficult. This paper presents a new prediction technique, which is multiwavelets neural network. This technique absorbs the advantage of high resolution of wavelets and the advantages of learning and feed-forward of neural networks. The procedure of using the multiwavelets neural network for predicting is described in detail in this paper. Principal components analysis (PCA) as a statistical technique has been used to simplify the time series analysis in our experiments. The effectiveness of this network is demonstrated by applying it to predict Ikeda time series.

The study of the quality of electric power lines is usually known as Power Quality. Power quality problems are increasingly due to a proliferation of equipment that is sensitive and polluting at the same time. The detection and classification of the different disturbances which cause power quality problems is a difficult task which requires a high level of engineering knowledge. Thus, neural networks are usually a good choice for the detection and classification of these disturbances. This paper describes a powerful system for detection of electrical disturbances by means of neural networks.

Recent researches have been focusing on fusion of audio and visual features for reliable speech recognition in noisy environments. In this paper, we propose a neural network based model of robust speech recognition by integrating audio, visual, and contextual information. Bimodal Neural Network (BMNN) is a multi-layer perceptron of 4 layers, which combines audio and visual features of speech to compensate loss of audio information caused by noise.  In order to improve the accuracy of speech recognition in noisy environments, we also propose a post-processing based on contextual information which are sequential patterns of words spoken by a user. Our experimental results show that our model outperforms any single mode models. Particularly, when we use the contextual information, we can obtain over 90% recognition accuracy even in noisy environments, which is a significant improvement compared with the state of art in speech recognition.

Feature extraction is an important issue in electromyography (EMG) pattern classification, where feature sets of high dimensionality are always used. This paper proposes a novel classification method to deal with high-dimensional EMG patterns, using a probabilistic neural network, a reduced-dimensional log-linearized Gaussian mixture network (RD-LLGMN) [1]. Since RD-LLGMN merges feature extraction and pattern classification processes into its structure, lower-dimensional feature set consistent with classification purposes can be extracted, so that, better classification performance is possible. To verify feasibility of the proposed method, phoneme classification experiments were conducted using frequency features of EMG signals measured from mimetic and cervical muscles. Filter banks are used to extract frequency features, and dimensionality of the features grows significantly when we increase resolution of frequency. In these experiments, the proposed method achieved considerably high classification rates, and outperformed traditional methods that are based on principle component analysis (PCA).

The multi-class probability SVM (MPSVM) is designed by training the sigmoid function to map the output of each binary class SVM into a posterior probability, and then combining these learned binary-class PSVMs using one-against-all strategy. The method of basic probability assignment is proposed according to the probabilistic output and performance of the PSVM. The outputs of all the binary-class PSVMs comprising an MPSVM are represented in the frame of Dempster-Shafer theory. A Dempster-Shafer theory based multi-class SVM (DSMSVM) is constructed by using the combination rule of evidences. To deal with the distributed multi-source multi-class problem, the DSMSVM is trained corresponding to each information source, and then the Dempster-Shafer theory is used to combine these learned DSMSVMs. Our proposed method is applied to fault diagnosis of a diesel engine. The experimental results show that the accuracy and robustness of fault diagnosis can be improved by using our proposed approach.

Cold rolling is often carried out between the solid solution treatment and aging to assist in the aging hardening of Cu-Cr-Zr lead frame alloys. This paper presents the use of an artificial neural network(ANN) to model the non-linear relationship between parameters of rolling and aging with respect to hardness properties of Cu-Cr-Zr alloy. Based on the Gauss-Newton algorithm, Levenberg-Marquardt algorithm with high stability is deduced. High precision of the model is demonstrated as well as good generalization performance. The results show that the Levenberg-Marquardt(L-M) backpropagation(BP) algorithm of ANN system is effective for predicting and analyzing the hardness properties of Cu-Cr-Zr lead frame alloy.

A kind of recurrent fuzzy neural network (RFNN) is constructed by using recurrent neural network (RNN) to realize fuzzy inference. In this kind of RFNN, temporal relations are embedded in the network by adding feedback connections on the first layer of the network. And a RFNN based adaptive control (RFNNBAC) is proposed, in which, two RFNN are used to identify and control plant respectively. Simulation experiments are made by applying proposed RFNNBAC on AC servo control problem to confirm its effectiveness.

A novel approach for the implementation of nonlinear model predictive control (MPC) is proposed using neural network and particle swarm optimization (PSO). A three-layered radial basis function neural network is used to generate multi-step predictive outputs of the controlled process. A modified PSO with simulated annealing is used at the optimization process in MPC. The proposed algorithm enhances the convergence and accuracy of the controller optimization. Applications to a discrete time nonlinear process and a thermal power unit load system are studied. The simulation results demonstrate the effectiveness of the proposed algorithm.