Catálogo de publicaciones - libros

Temporal antecedent and implicative constraints are required to ensure the relevancy of events in analyzing temporal data from multiple sources. Fuzzy predicates are used to represent imprecise temporal constraints and durations and a fuzzy partitions provide a hierarchy that allows the analysis of implicative constraints on several levels of granularity. In this paper we have outlined how standard data mining strategies can be adapted to utilize fuzzy representations in the discovery of imprecise temporal relationships between data obtained from multiple sources.

The use of soft computing methodologies in the field of traffic and transport systems is of particular interest to researchers and practitioners due to their ability to handle quantitative and qualitative measures, and to efficiently solve problems which involve complexity, imprecision and uncertainty. This paper provides a survey of soft computing applications. A classification scheme for soft computing applications is defined. The current frameworks and some future directions of soft computing applications to traffic and transport systems are discussed.

Artificial neural networks are one of the recently explored advanced technologies, which show promise in the area of transportation engineering. However, in contrast to the availability of a large number of successful application demonstrations, it is hard to find studies in the literature that provide systematic examinations of the state-of-the-art, the application domains, and the applicability of artificial neural networks to transportation problems. On the other hand, some unseen artificial neural network development has been motivated by transportation engineering objectives. Therefore, this study presents the development of a neural network paradigm for the purpose of daily trip flow forecasting.

In this study, the prediction problem was transformed into the following minimum norm problem: Find a back - propagation neural network such that ‖()−()‖ is minimal. () represents the observed values and () represents the neural network generated values. The averages of half-hourly public transport (PT) trip flows using AKBIL (a ticketing system integrated to all PT modes) in Istanbul Metropolitan Area are used as data. As the objective of the study was the forecast of the half-hourly mean trip flows (trips) by using ANN, the analysis process consisted of two steps, which are training and testing respectively. The feed forward back-propagation (FFBP) method was used to train the ANNs. Determination of the number of hidden layers and the number of nodes in the input layer providing the best training results was the initial process of the training procedure. The mean square error (MSE) and the auto-correlation function (ACF) were used as criteria to evaluate the performance of the training simulations. Three hidden layer nodes were found appropriate for the four-nodded input layer. More than three nodes for the hidden layer could not improve the MSE during the testing stage. Following the ANN forecast, a comparison was made by means of a stochastic model. An auto-regressive model of order four was used. The results were then compared. Among the studied models, a significant improvement in the prediction made by the neural network model is noticed, due to its flexibility to adapt to time-series database. Making forecasts by generating negative values after trials is a disadvantage of the algorithm used to train the neural network. It can be said that various kinds of learning algorithms should be assessed during the training process to overcome this drawback.

Finding groups of individuals with similar activity patterns (a sequence of activities within a given time period, usually 24 hours) has become an important issue in models of activity-based approaches to travel demand analysis. This knowledge is critical to many activity-based models, and it aids our understanding of activity/travel behavior. This paper aims to develop a methodology for the clustering of these patterns. There is a large number of well-known clustering algorithms, such as hierarchical clustering, or k-means clustering (which belongs to the class of partitioning algorithm). However, these algorithms cannot be used to cluster categorical data, so they do not suit the problem of clustering of activity patterns well. Several other heuristics have been developed to overcome this problem. The k-medoids algorithm, described in this paper, is a modification of the k-means algorithm with respect to categorical data. However, similar to the k-means algorithm, the k-medoids algorithm can converge to local optima. This paper approaches the medoids-based formulation of clustering problem using genetic algorithms (GAs), a probabilistic search algorithm that simulates natural evolution. The main objective of this paper is to develop a robust algorithm that suits the problem of clustering of activity patterns and to demonstrate and discuss its properties.

Travel choices are made according to people’s personal preferences and knowledge of the system. Since increase, improvement and updating of knowledge is achieved through information, consequentially information itself is a crucial issue in transportation problems. If information was perfect, users could easily choose the best path from their point of view, but unfortunately complete and precise information about network conditions is rarely available, therefore uncertainty can cause anxiety and stress in decision makers.

In specific technical literature, uncertainty has been usually modelled through random utility models. Randomness is then used to represent uncertainty, and therefore the probability of a choice can be calculated.

Since recent studies linked uncertainty to the concepts of approximate reasoning rather than randomness, in this paper we quantify the influence of information provision on drivers’ behaviour, according to Uncertainty-based Information Theory. A modelling framework based on Evidence Theory, measuring the rather than the of a choice, has been carried out to represent the uncertainties in the perception of travel attributes.

A sequential model has been also used to simulate updating of user knowledge, and finally a numerical application shows that users’ trust level in information plays a relevant role in choice processes. Within this framework, the importance of conditional Uncertainty in updating knowledge of the system results evident.

In practical system identification it is often desirable to simultaneously handle several objectives and constraints. In some cases, these objectives and constraints are often non-commensurable and the objective functions are explicitly/mathematically not available. In this paper, Interactive Evolutionary Computation (IEC) is used to effectively handle these identification problems. IEC is an optimization method that adopts evolutionary computation (EC) among system optimization based on subjective human evaluation. The proposed approach has been implemented in MATLAB (EAsy-IEC Toolbox) and applied to the identification of a pilot batch reactor. The results show that IEC is an efficient and comfortable method to incorporate a priori knowledge of the user into a user-guided optimization and identification problems. The developed EASy-IEC Toolbox can be downloaded from the website of the authors: http://www.fmt.vein.hu/softcomp/EAsy.

In this paper, we propose a replacement strategy for steady-state genetic algorithms that takes into account two features of the element to be included into the population: a measure of the contribution of diversity to the population and the fitness function. In particular, the proposal attempts to replace an element in the population with worse values for these two features. In this way, the diversity of the population is increased and the quality of its solutions is improved, simultaneously, maintaining high levels of useful diversity. Experimental results show that the use of the proposed replacement strategy allows significant performance to be achieved for problems with different difficulties, which regards to other replacement strategies presented in the literature.

Nowadays a lot of libraries and software systems for Evolutionary Computation (EC) are in existance. Unfortunately, an EC created by one of those can rarely be exchanged between them. In recent works different Soft Computing methods are modelled by using XML notation to overcome this problem. But to be able to model ECs in, e.g., XML notation, a concept is needed which allows the hierarchical description of ECs. This paper proposes a hierarchically modelling concept for Evolutionary Computation which allows the description of ECs as EC-hierarchy on an abstract level. A created EC-hierarchy can be mapped on every notation which is capable of representing tree-like structures like XML or hierarchy-oriented programming languages. Such a hierarchical model could be used for exchanging ECs between software systems, libraries as well as researchers.

Several important applications require a time-dependent (on-line) in which either the objective function or the problem parameters or both vary with time. Several studies are available in the literature about the use of genetic algorithms for time dependent fitness landscape in single-objective optimization problems. But when dynamic multi-objective optimization is concerned, very few studies can be found. Taking inspiration from Artificial Life (ALife), a strategy is proposed ensuring the approximation of Pareto-optimal set and front in case of unpredictable parameters changes. It is essentially an ALife-inspired evolutionary algorithm for variable fitness landscape search. We describe the algorithm and test it on some test cases.

This paper deals with the problem of optimum test program development in analog circuit testing process. Pre-production testing and production testing are taken into account by the presented algorithm. Genetic algorithms are used as a new method for optimum test selection. This approach enhances quality and speeds up finding suboptimal solutions by finding more then one good result in each genetic algorithm cycle. Results for a hypothetical example are given to clarify and discuss the method and they seem to be very promising.