Hannu Koivisto

Neural networks in process fault diagnosis

Fault detection and diagnosis is an important problem in process automation. Both model-based methods and expert systems have been suggested to solve the problem, along with the pattern recognition approach. A number of possible neural network architectures for fault diagnosis are studied. The multilayer perceptron network with a hyperbolic tangent as the nonlinear element seems best suited for the task. As a test case, a realistic heat exchanger-continuous stirred tank reactor system is studied. The system has 14 noisy measurements and 10 faults. The proposed neural network was able to learn the faults in under 3000 training cycles and then to detect and classify the faults correctly. Principal component analysis is used to illustrate the fault diagnosis problem in question. >

A Dynamically Constrained Multiobjective Genetic Fuzzy System for Regression Problems

In this paper, a multiobjective genetic fuzzy system (GFS) to learn the granularities of fuzzy partitions, tuning the membership functions (MFs), and learning the fuzzy rules is presented. It uses dynamic constraints, which enable three-parameter MF tuning to improve the accuracy while guaranteeing the transparency of fuzzy partitions. The fuzzy models (FMs) are initialized by a method that combines the benefits of Wang-Mendel (WM) and decision-tree algorithms. Thus, the initial FMs have less rules, rule conditions, and input variables than if WM initialization were to be used. Moreover, the fuzzy partitions of initial FMs are always transparent. Our approach is tested against recent multiobjective and monoobjective GFSs on six benchmark problems. It is concluded that the accuracy and interpretability of our FMs are always comparable or better than those in the comparative studies. Furthermore, on some benchmark problems, our approach clearly outperforms some comparative approaches. Suitability of our approach for higher dimensional problems is shown by studying three benchmark problems that have up to 21 input variables.

Fuzzy classifier identification using decision tree and multiobjective evolutionary algorithms

This paper presents a hybrid method for identification of Pareto-optimal fuzzy classifiers (FCs). In contrast to many existing methods, the initial population for multiobjective evolutionary algorithms (MOEAs) is neither created randomly nor a priori knowledge is required. Instead, it is created by the proposed two-step initialization method. First, a decision tree (DT) created by C4.5 algorithm is transformed into an FC. Therefore, relevant variables are selected and initial partition of input space is performed. Then, the rest of the population is created by randomly replacing some parameters of the initial FC, such that, the initial population is widely spread. That improves the convergence of MOEAs into the correct Pareto front. The initial population is optimized by NSGA-II algorithm and a set of Pareto-optimal FCs representing the trade-off between accuracy and interpretability is obtained. The method does not require any a priori knowledge of the number of fuzzy sets, distribution of fuzzy sets or the number of relevant variables. They are all determined by it. Performance of the obtained FCs is validated by six benchmark data sets from the literature. The obtained results are compared to a recently published paper [H. Ishibuchi, Y. Nojima, Analysis of interpretability-accuracy tradeoff of fuzzy systems by multiobjective fuzzy genetics-based machine learning, International Journal of Approximate Reasoning 44 (1) (2007) 4-31] and the benefits of our method are clearly shown.

Developing a bioaerosol detector using hybrid genetic fuzzy systems

The aim of this work is to develop a model, which works as a reasoning mechanism in a bioaerosol detector. Ability to distinguish between safe and harmful aerosols is one of its main requirements. Instead of commonly used misclassification rate as a metric of accuracy, true positive (TP) and false positive (FP) rates are used because of the uneven misclassification costs and class distributions of the collected data. Interpretability of the model builds up the confidence for the developed model and enables its adjustment in cases when bioaerosol detector is further developed. Thus, it is another crucial requirement for the model. Clearly, the objectives are contradicting and therefore multiobjective evolutionary algorithms (MOEAs) are applied to find tradeoff models. Fuzzy classifiers (FCs) are selected as a model type because their linguistic rules are intuitive to human beings. FCs are identified by hybrid genetic fuzzy system (GFS) which initializes the population adequately using decision trees (DTs) and simplification operations. During MOEA optimization transparency of fuzzy partition is used as a metric of interpretability and TP and FP rates as metrics of accuracy. Heuristic rule and rule condition removal is applied to offspring population in order to keep the rule base consistent. The identified FCs are highly comprehensible yet accurate and their linguistic rules provide valuable insights for further development of bioaerosol detector.

Identification of interpretable and accurate fuzzy classifiers and function estimators with hybrid methods

This paper studies the identification of fuzzy classifiers and function estimators focusing on improving their interpretability while maintaining their accuracy. Advances of various methods, such as, input variable selection, appropriate initialization algorithms, evolutionary algorithms and simplification techniques are hybridized to form a framework capable of identifying interpretable and accurate fuzzy models (FMs). FMs are initialized by two algorithms. Modified Gath-Geva (MGG) is used for function estimation and C4.5 for classification problems. The initialized FMs go through a three-step GA-based optimization, in which the adequate structure and parameters of FMs are searched. The proposed fitness function makes the favoring of simple FMs possible. Furthermore, the rule base is made more comprehensible by reducing the number of conditions in the rules. The validity of FMs is verified through studying several well-known benchmark problems. The results indicate, that by means of the proposed framework, interpretable, yet accurate FMs are obtained.

Circular correlation based identification of switching power converter with uncertainty analysis using fuzzy density approach

Abstract Switching power converters are extensively used in powering various electronics loads and processes. The proper functioning of those processes may be vital for the every day life of the society. Therefore, the reliable operation of the switched-mode converters is of prime importance and the functioning has to be verified reliably both during the design phase and in the production. It has been shown lately that the main deficiencies in the verification process are related to the dynamics of the converter which can be characterized with a certain set of transfer functions. This paper investigates the frequency response measurement methods by means of which the transfer functions can be identified fast and accurately being economically feasible to apply also in the production phase. Multi-period maximum length pseudo random binary sequence (m.l.b.s.) is used as the excitation signal and the transfer functions are identified from the measurement data with circular cross-correlation method. The measurement uncertainty is computed by means of fuzzy density approach yielding a certain confidence band around the measured nominal response. The proposed methods are verified both by simulations and experimental data from high-frequency switched-mode converters.

Optimization of Neural Network Topologies using Genetic Algorithm

ABSTRACTNeural networks (NN) are widely applied in modeling. The modeling process in which neural networks are applied involves the same problems as identification in general. In addition, the parameterization selected for the NN model is a key issue i.e., what number of hidden nodes is appropriate and what is their connectivity.This paper describes an evolutionary method for selecting the topology of a feedforward neural network by means of a genetic algorithm. The approach is a hybrid method utilizing a genetic algorithm for structure selection and a second-order training algorithm for parameter estimation. The method is tested with two modeling problems, one a well-known benchmark of an infra-red laser and the other modeling of a laboratory-sized pilot process imitating a paper machine head box.

Neural predictive control-a case study

Adaptive neural network predictive controllers are briefly reviewed. Some of the first results of an online application to a laboratory scaled nonlinear heating process are presented. Both the one-step-ahead and the long-range (multistep) predictive controller are considered. Both neural network controllers exhibit good, robust performance.<<ETX>>

The GARCH-FuzzyDensity method for density forecasting

Abstract: This paper introduces a new class of GARCH method, termed the GARCH-Fuzzy-Density method for density forecasting the realizations of a process in which the higher-order moments may be time-varying. The method is based on a probabilistic Takagi-Sugeno fuzzy system and GARCH model. Traditional GARCH models usually assume that the shape of the conditional distribution of a process is conditional only on the first two moments. However, it is well documented that in empirical applications the conditional distribution beyond the first two moments may be conditional on higher-order moments such as skewness. Therefore, traditional GARCH models are insufficient for capturing all aspects of such processes. To resolve the problem mentioned above, the GARCH-FuzzyDensity model was developed. The method is capable of modeling conditional distributions in which the higher-order moments may be time-varying. Therefore, the GARCH-FuzzyDensity model provides more accurate density forecasts for the higher-order moment varying processes than the traditional GARCH models.

Quality management in GPRS networks with fuzzy case-based reasoning

Mobile networks pose challenges to quality management because of the limited capacity of the air interface and the mobility of the users. GPRS is the prevailing system for mobile connectivity at the moment. This paper approaches quality management in GPRS networks with a two-phase system, where a detector block first culls quality disturbances and a fuzzy case-based reasoning engine then proposes a solution to the problem. The main advantage of the concept is model maintenance: the experienced network operator can take part in the decision-making and his or her knowledge thus accumulates in the case base. We also present simulated examples of GPRS network data classified with the detector and inserted into the case base.