Paris A. Mastorocostas

A recurrent fuzzy-neural model for dynamic system identification

This paper presents a fuzzy modeling approach for identification of dynamic systems. In particular, a new fuzzy model, the Dynamic Fuzzy Neural Network (DFNN), consisting of recurrent TSK rules, is developed. The premise and defuzzification parts are static while the consequent parts of the fuzzy rules are recurrent neural networks with internal feedback and time delay synapses. The network is trained by means of a novel learning algorithm, named Dynamic-Fuzzy Neural Constrained Optimization Method (D-FUNCOM), based on the concept of constrained optimization. The proposed algorithm is general since it can be applied to locally as well as fully recurrent networks, regardless of their structures. An adaptation mechanism of the maximum parameter change is presented as well. The proposed dynamic model, equipped with the learning algorithm, is applied to several temporal problems, including modeling of a NARMA process and the noise cancellation problem. Performance comparisons are conducted with a series of static and dynamic systems and some existing recurrent fuzzy models. Simulation results show that DFNN compares favorably with its competing rivals and thus it can be considered for efficient system identification.

Fuzzy modeling for short term load forecasting using the orthogonal least squares method

A fuzzy modeling method is developed in this paper for short term load forecasting. According to this method, identification of the premise part and consequent part is separately accomplished via the orthogonal least squares (OLS) technique. Particularly, the OLS is first employed to partition the input space and determine the number of fuzzy rules and the premise parameters. In the sequel, a second orthogonal estimator determines the input terms which should be included in the consequent part of each fuzzy rule and calculate its parameters. Input selection is automatically performed, given an input candidate set of arbitrary size, formulated by an expert. A satisfactory prediction performance is attained as shown in the test results, showing the effectiveness of the suggested method.

An orthogonal least squares-based fuzzy filter for real-time analysis of lung sounds

Pathological discontinuous adventitious sounds (DAS) are strongly related with the pulmonary dysfunction. Its clinical use for the interpretation of respiratory malfunction depends on their efficient and objective separation from vesicular sounds (VS). In this paper, an automated approach to the isolation of DAS from VS, based on their nonstationarity, is presented. The proposed scheme uses two fuzzy inference systems (FISs), operating in parallel, to perform the task of adaptive separation, resulting in the orthogonal least squares-based fuzzy filter (OLS-FF). By applying the OLS-FF to fine/coarse crackles and squawks, selected from three lung sound databases, the coherent structure of DAS is revealed and they are efficiently separated from VS. The important time domain DAS features, related to diagnostic information, are preserved and their true location and structural morphology are automatically identified. When compared to previous works, the OLS-FF performs quite similarly, but with significantly lower computational load, resulting in a faster real-time clinical screening of DAS.

A constrained orthogonal least-squares method for generating TSK fuzzy models: Application to short-term load forecasting

In this paper, an orthogonal least-squares (OLS) based modeling method is developed, named the constrained OLS (C-OLS), for generating simple and efficient TSK fuzzy models. The method is a two-stage model building technique, where both premise and consequent identification are simultaneously performed. The fuzzy system is considered as a linear regression model by decomposing the TSK model into a collection of generic rules. The C-OLS algorithm is employed at stage-1 to identify the structure of the model. Given a model building data set, the algorithm selects a subset of most significant regressors which should be included in the model. Based on the similarity measure, a classification tool is developed, which organizes the selected terms into groups with similar premise parts, forming TSK rules. Additionally, input variable selection for the consequent part is performed. The resulting model is reduced in complexity by discarding the unnecessary terms, and is optimized at stage-2 using a richer training data set. This method is used to generate fuzzy models for a real-world problem, the load forecasting of the Greek power system. Extensive simulation results are given and discussed, demonstrating the effectiveness of the suggested method.

A stable learning algorithm for block-diagonal recurrent neural networks: application to the analysis of lung sounds

A novel learning algorithm, the Recurrent Neural Network Constrained Optimization Method (RENNCOM) is suggested in this paper, for training block-diagonal recurrent neural networks. The training task is formulated as a constrained optimization problem, whose objective is twofold: 1) minimization of an error measure, leading to successful approximation of the input/output mapping and 2) optimization of an additional functional, the payoff function, which aims at ensuring network stability throughout the learning process. Having assured the network and training stability conditions, the payoff function is switched to an alternative form with the scope to accelerate learning. Simulation results on a benchmark identification problem demonstrate that, compared to other learning schemes with stabilizing attributes, the RENNCOM algorithm has enhanced qualities, including, improved speed of convergence, accuracy and robustness. The proposed algorithm is also applied to the problem of the analysis of lung sounds. Particularly, a filter based on block-diagonal recurrent neural networks is developed, trained with the RENNCOM method. Extensive experimental results are given and performance comparisons with a series of other models are conducted, underlining the effectiveness of the proposed filter.

A hybrid fuzzy modeling method for short-term load forecasting

This paper presents the development of a hybrid fuzzy modeling method for short-term load forecasting. The new approach employs the orthogonal least squares method to create the fuzzy model and a constrained optimization algorithm to perform the parameter learning. The proposed model is tested using data of the Greek power system, while load forecasts with satisfying accuracy are reported.

An orthogonal least-squares method for recurrent fuzzy-neural modeling

Abstract This paper presents an orthogonal least-squares (OLS)-based modeling method, named dynamic OLS (D-OLS), for generating recurrent fuzzy models. A dynamic-neuron-based fuzzy neural network is proposed, comprising generalized Takagi–Sugeno–Kang (TSK) fuzzy rules, whose consequent parts consist of dynamic neurons with local output feedback. From an arbitrarily large set of candidate dynamic neurons, the D-OLS method selects automatically the most important ones. Thus, in the resulting model, the consequent part of each fuzzy rule contains dynamic neurons with different time delays. The proposed dynamic model, equipped with the learning algorithm, is applied to two temporal problems, where the effectiveness of the suggested method as well as the advantages of the resulting dynamic model are demonstrated.

Classification of Remotely Sensed Images Using the GeneSIS Fuzzy Segmentation Algorithm

In this paper, we propose an integrated framework of the recently proposed Genetic Sequential Image Segmentation (GeneSIS) algorithm. GeneSIS segments the image in an iterative manner, whereby at each iteration, a single object is extracted via a genetic algorithm-based object extraction method. This module evaluates the fuzzy content of candidate regions, and through an effective fitness function design provides objects with optimal balance between fuzzy coverage, consistency and smoothness. GeneSIS exhibits a number of interesting properties, such as reduced over-/undersegmentation, adaptive search scale, and region-based search. To enhance the capabilities of GeneSIS, we incorporate here several improvements of our initial proposal. On one hand, two modifications are introduced pertaining to the object extraction algorithm. Specifically, we consider a more flexible representation of the structural elements used for the objects extraction. Furthermore, in view of its importance, the consistency criterion is redefined, thus providing a better handling of the ambiguous areas of the image. On the other hand we incorporate three tools properly devised, according to the fuzzy principles characterizing GeneSIS. First, we develop a marker selection strategy that creates reliable markers, particularly when dealing with ambiguous components of the image. Furthermore, using GeneSIS as the essential part, we consider a generalized experimental setup embracing two different classification schemes for remote sensing images: the spectral-spatial classification and the supervised segmentation methods. Finally, exploiting the inherent property of GeneSIS to produce multiple segmentations, we propose a segmentation fusion scheme. The effectiveness of the proposed methodology is validated after thorough experimentation on four data sets.

Brief paper: A computational intelligence-based forecasting system for telecommunications time series

In this work a computational intelligence-based approach is proposed for forecasting outgoing telephone calls in a University Campus. A modified Takagi-Sugeno-Kang fuzzy neural system is presented, where the consequent parts of the fuzzy rules are neural networks with an internal recurrence, thus introducing the dynamics to the overall system. The proposed model, entitled Locally Recurrent Neurofuzzy Forecasting System (LR-NFFS), is compared to well-established forecasting models, where its particular characteristics are highlighted.

FUNCOM: a constrained learning algorithm for fuzzy neural networks

A novel learning algorithm, the FUNCOM (Fuzzy Neural Constrained Optimization Method) is suggested in this paper, for training fuzzy neural networks. The training task is formulated as a constrained optimization problem, whose objective is twofold: (i) minimization of an error measure, leading to successful approximation of the input/output mapping and (ii) optimization of an additional functional, which aims at formulating suitable internal representations of the fuzzy model. Optimization of the above functionals is carried out under the constraints imposed by the fuzzy system, which appear in the form of state equations. A fuzzy adaptation scheme is also suggested, which continuously modifies the step size during training, with the scope to improve the learning attributes of the algorithm. The FUNCOM qualities are investigated by a series of simulation examples. Comparisons with other learning algorithms are given and discussed, indicating the effectiveness of the proposed algorithm.