Publications

A rapid and accurate prediction of byproduct gas flow in steel industry can help not only to become aware of the operational situations of gas system, but it also provides the energy scheduling workers with sound decision-making mechanisms. In this study, a least square support vector machine (LS-SVM) model based on online hyperparameters optimization is proposed, where the variance of effective noise of the sample is estimated, while a conjugate gradient algorithm is developed to optimize the width of Gaussian kernels and the regularization factor. To assess the quality of the proposed method, we experiment with a test function affected by additive noise and an industrial gas flow data from Shanghai Baosteel Company Ltd. A series of comparative experiments are reported as well. The results demonstrate that the proposed method shows the shortest computing time while ensuring the prediction accuracy. These two features make the approach applicable to real-time prediction of gas flow in steel industry.

Granular computing (GrC) has emerged as a unified conceptual and processing framework. Information granules are fundamental constructs that permeate concepts and models of GrC. This paper is concerned with a design of a collection of meaningful, easily interpretable ellipsoidal information granules with the use of the principle of justifiable granularity by taking into consideration reconstruction abilities of the designed information granules. The principle of justifiable granularity supports designing of information granules based on numeric or granular evidence, and aims to achieve a compromise between justifiability and specificity of the information granules to be constructed. A two-stage development strategy behind the construction of justifiable information granules is considered. First, a collection of numeric prototypes is determined with the use of fuzzy clustering. Second, the lengths of the semi-axes of ellipsoidal information granules to be formed around such prototypes are optimized. Two optimization criteria are introduced and studied. Experimental studies involving synthetic data set and data sets coming from the machine learning repository are reported.

Abstract The paper deals with a problem of modeling of fuzzy systems in random environments. A model is proposed that is capable of handling two distinct forms of imprecision, viz. randomness and fuzziness. The model is required to cope with both of them while modeling a variety of problems in management, medical diagnosis, and unsupervised pattern recognition. The models proposed in the paper are constructed and evaluated in a formal framework established by fuzzy relation equations. Randomness is introduced as additional constraints imposed on the structure of the fuzzy relation equation (hence: structured fuzzy models). The forecasting (prediction) problem is studied in detail.

In this study, we revisit the well-known notion of fuzzy state machines and discuss their development through learning. The systematic development of fuzzy state machines has not been pursued as intensively as it could have been expected from the breadth of the possible usage of them as various modelling platforms. We concentrate on the generalization of the well known architectures exploited in Boolean system synthesis, namely Moore and Mealy machines and show how these can be implemented in terms of generic functional modules such as fuzzy JK flip-flops and fuzzy logic neurons (AND and OR neurons) organized in the form of logic processors. It is shown that the design of the fuzzy state machines can be accomplished through their learning. The detailed learning algorithm is presented and illustrated with a series of numeric examples. The study reveals an interesting option of constructing digital systems through learning: the original problem is solved in the setting of fuzzy state machines and afterwards "binarised" into the two-valued format realized via the standard digital hardware.

When dealing with the curse of dimensionality (small sample size with many dimensions), feature subset selection is an important preprocessing strategy. This issue is particularly germane to the discrimination of class-labeled high-dimensional biomedical spectra as is often acquired from magnetic resonance and infrared spectrometers. A technique is presented that stochastically selects feature subsets with varying cardinality for discrimination by probabilistic neural networks. The results are benchmarked against two classifiers using the entire feature set both with and without feature averaging. The new technique had significantly fewer misclassifications than either of the benchmarks.

In recent years, due to the growing demand for computational resources, particularly in cloud computing systems, the data centers’ energy consumption is continually increasing, which directly causes price rise and reductions of resources’ productivity. Although many energy‐aware approaches attempt to minimize the consumption of energy, they cannot minimize the violation of service‐level agreements at the same time. In this paper, we propose a method using a granular neural network, which is used to model data processing. This method identifies the physical hosts’ workloads before the overflow and can improve energy consumption while also reducing violation of service‐level agreements. Unlike the other techniques that use a single criterion, namely, worked on the basis of the history of using the processor, we simultaneously use all the productivity rates criteria, that is, processor productivity rates, main memory, and bandwidth. Extensive real‐world simulations using the CloudSim simulator show the high efficiency of the proposed algorithm.

A problem of handling fuzzy quantities in a process of knowledge acquisition and deriving an inference mechanism by means of fuzzy relation equations is studied in extensive way. It is clearly pointed out that both of them are closely related and correspond to various types of fuzzy relation equations that are considered. Their relevance to the form of knowledge collected is also indicated. A problem of dimension reduction of a knowledge base is considered as well. Two modes of the use of the knowledge base (goal‐, and data‐driven) are also studied.

A class of clustering problems that is studied here is concerned with the development of a structure of a global nature given a collection of structures (clusters) constructed locally for data that are represented by several collections (blocks) of features. These blocks of features come with a well-defined semantics. For instance, in spatiotemporal data, a certain block of features concerns a spatial component of the data (say, x-y or x-y-z coordinates), while another one deals with the features that describe time series associated with the corresponding locations. The results of clustering that are being produced locally are reconciled by minimizing a distance between the proximity matrices that are formed at the higher conceptual level and induced by the individual partition matrices. The optimization problem is formulated and presented along with its iterative scheme.