Catálogo de publicaciones - libros

The main objective of this paper is two folds. First is to assess some well-known linear and nonlinear techniques comparatively in modeling and forecasting financial time series with trend and seasonal patterns. Then to investigate the effect of pre-processing procedures, such as seasonal adjustment methods, to the improvement of the modeling capability of a nonlinear structure implemented as ANNs in comparison to the classical Box-Jenkins seasonal autoregressive integrated moving average (ARIMA) model, which is widely used as a linear statistical time series forecasting method. Furthermore, the effectiveness of seasonal adjustment procedures, i.e. direct or indirect adjustments, on the forecasting performance is evaluated. The Autocorrelation Function (ACF) plots are used to determine the correlation between lags due to seasonality, and to determine the number of input nodes that is also confirmed by trial-and-errors. The linear and nonlinear models mentioned above are applied to aggregate retail sales data, which carries strong trend and seasonal patterns. Although, the results without any pre-processing were in an acceptable interval, the overall forecasting performance of ANN was not better than that of the classical method. After employing the right seasonal adjustment procedure, ANN has outperformed its linear counterpart in out-of-sample forecasting. Consequently, it is confirmed that the modeling capability of ANN is improved significantly by using a pre-processing procedure. The results obtained from both ARIMA and ANNs based forecasting methodologies are analyzed and compared with Mann-Whitney statistical test.

Import of intelligent features to time series analysis including the possibility of operating with linguistic information, reasoning and replying on intelligent queries is the prospective direction of development of such systems. The paper proposes novel methods of perception based time series data mining using perceptual patterns, fuzzy rules and linguistic descriptions. The methods of perception based forecasting using perceptual trends and moving approximation (MAP) transform are discussed. The first method uses perception based function for modeling qualitative forecasting given by expert judgments. The second method uses MAP transform and measure of local trend associations for description of perceptual pattern corresponding to the region of forecasting. Finally, the method of generation of association rules for multivariate time series based on MAP and fuzzy trends is discussed. Multivariate time series are considered as description of system dynamics. In this case association rules can be considered as relationships between system elements additional to spatial, causal etc. relations existing in the system. The proposed methods are illustrated on examples of artificial and real time series.

This paper is concerned with redundancy detection and elimination in databases via the solution of a key equivalence problem. The approach is related to the hardening of soft databases method due to Cohen , [4]. Here, the problem is described in graph theoretic terms. An appropriate optimization model is drawn and solved indirectly. This approach is shown to be effective. Computational results on test databases are included.

Using the geodesic distance metric instead of the Euclidean distance metric, ISOMAP can visualize the convex but intrinsically flat manifolds such as the swiss roll data set nicely. But it’s well known that ISOMAP performs well only when the data belong to a single well-sampled manifold, and fails when the data lie on disjoint manifolds or imperfect manifolds. Generally speaking, as the data points are farer from each other on the manifold, the approximation of the shortest path to the geodesic distance is worse, especially for imperfect manifolds, that is, long distances are approximated generally worse than short distances, which makes the classical MDS algorithm used in ISOMAP unsuitable and thus often leads to the overlapping or ”overclustering” of the data. To solve this problem, we improve the original ISOMAP algorithm by replacing the classical MDS algorithm with the Sammon’s mapping algorithm, which can limit the effects of generally worse-approximated long distances to a certain extent, and thus better visualization results are obtained. As a result, besides imperfect manifolds, intrinsically curved manifolds such as the fishbowl data set can also be visualized nicely. In addition, based on the characteristics of the Euclidean distance metric, a faster Dijkstra-like shortest path algorithm is used in our method. Finally, experimental results verify our method very well.

Conversational Case-Based Reasoning (CCBR) provides a mixed-initiative dialog for guiding users to refine their problem descriptions incrementally through a question-answering sequence. Most CCBR approaches assume that there is at most one discrete value on each feature. While a generalized case (GC), which has been proposed and used in traditional CBR processes, has multiple values on some features. Motivated by the conversational software component retrieval application, we focus on the problem of extending CCBR to support GCs in this paper. This problem is tackled from two aspects: similarity measuring and discriminative question ranking.

In this paper we proposed an approach to maintain large case library, which based on the idea that a large case library can be transformed to a compact one by using a set of case-specific weights. A linear programming technique is being used to obtain case-specific weights. By learning such local weights knowledge, many of redundant or similar cases can be removed from the original case library or stored in a secondary case library. This approach is useful for case library with a large number of redundant or similar cases and the retrieval efficiency is a real concern of the user. This method of maintaining case library from scratch, as proposed in this paper, consists of two main steps. First, a linear programming technique for learning case-specific weights is used to evaluate the importance of different features for each case. Second, a case selection strategy based on the concepts of case coverage and reachability is carried out to select representative cases. Furthermore, a case retrieval strategy of the compact case library we built is discussed. The effectiveness of the approach is demonstrated experimentally by using two sets of testing data, and the results are promising.

Recently proposed manifold learning algorithms, e.g. Isometric feature mapping (Isomap), Locally Linear Embedding (LLE), and Laplacian Eigenmaps, are based on minimizing the construction error for data description and visualization, but not optimal from classification viewpoint. A discriminant isometric feature mapping for face recognition is presented in this paper. In our method, the geodesic distances between data points are estimated by Floyd’s algorithm, and Kernel Fisher Discriminant is then utilized to achieve the discriminative nonlinear embedding. Prior to the estimation of geodesic distances, the neighborhood graph is constructed by incorporating class information. Experimental results on two face databases demonstrate that the proposed algorithm achieves lower error rate for face recognition.

This paper presents a Grey-Markov forecasting model for forecasting the electric power requirement in China. This method takes into account the general trend series and random fluctuations about this trend. It has the merits of both simplicity of application and high forecasting precision. This paper is based on historical data of the electric power requirement in China, and forecasts and analyzes the electric power requirement in China by the Grey–Markov forecasting model. The forecasting precisions of Grey-Markov forecasting model from 2002 to 2004 are 99.42%, 98.05% and 97.56%, and those of GM(1,1) grey forecasting model are 98.53%, 94.02% and 88.48%. It shows that the Grey-Markov forecasting models have higher precision than GM(1,1) grey forecasting model. The results provides scientific basis for the planned development of the electric power supply in China.

We present a new approach for process fault detection based on models generated by machine learning techniques. Our work is based on a residual generation scheme, where the output of a model for process normal behavior is compared against actual process values. The residuals indicate the presence of a fault. The model consists of a general statistical inference engine operating on discrete spaces. This model represents the maximum entropy joint probability mass function (pmf) consistent with arbitrary lower order probabilities. The joint pmf is a rich model that, once learned, allows one to address inference tasks, which can be used for prediction applications. In our case the model allows the one step-ahead prediction of process variable, given its past values. The relevant past values for the forecast model are selected by learning a causal structure with an algorithm to learn a discrete bayesian network. The parameters of the statistical engine are found by an approximate method proposed by Yan and Miller. We show the performance of the prediction models and their application in power systems fault detection.

Mutation operators play an important role in evolutionary programming. Several different mutation operators have been developed in the past decades. However, each mutation operator is only efficient in some type of problems, but fails in another one. In order to overcome the disadvantage, a possible solution is to use a mixed mutation strategy, which mixes various mutation operators. In this paper, an example of such strategies is introduced which employs five different mutation strategies: Gaussian, Cauchy, Levy, single-point and chaos mutations. It also combines with the technique of species conservation to prevent the evolutionary programming from being trapped in local optima. This mixed strategy has been tested on 21 benchmark functions. The simulation results show that the mixed mutation strategy is superior to any pure mutation strategy.