Tiago Matias

Automatic extraction of the fuzzy control system by a hierarchical genetic algorithm

The paper proposes a new method to automatically extract all fuzzy parameters of a Fuzzy Logic Controller (FLC) in order to control nonlinear industrial processes. The main objective of this paper is the extraction of a FLC from data extracted from a given process while it is being manually controlled. The learning of the FLC is performed by a hierarchical genetic algorithm (HGA), from a set of process-controlled input/output data. The algorithm is composed by a five level structure, being the first level responsible for the selection of an adequate set of input variables. The second level considers the encoding of the membership functions. The individual rules are defined on the third level. The set of rules are obtained on the fourth level, and finally, the fifth level selects the elements of the previous levels, as well as, the t-norm operator, inference engine and defuzzifier methods which constitute the FLC. To optimize the proposed method, the HGAs initial populations are obtained by an initialization algorithm. This algorithm has the main goal of providing a good initial solution for membership functions and rule based populations, enhancing the GAs tuning. Moreover, the HGA is applied to control the dissolved oxygen in an activated sludge reactor within a wastewater treatment plant. The results are presented, showing that the proposed method extracted all the parameters of the fuzzy controller, successfully controlling a nonlinear plant.

Learning of a single-hidden layer feedforward neural network using an optimized extreme learning machine

This paper proposes a learning framework for single-hidden layer feedforward neural networks (SLFN) called optimized extreme learning machine (O-ELM). In O-ELM, the structure and the parameters of the SLFN are determined using an optimization method. The output weights, like in the batch ELM, are obtained by a least squares algorithm, but using Tikhonovs regularization in order to improve the SLFN performance in the presence of noisy data. The optimization method is used to the set of input variables, the hidden-layer configuration and bias, the input weights and Tikhonovs regularization factor. The proposed framework has been tested with three optimization methods (genetic algorithms, simulated annealing, and differential evolution) over 16 benchmark problems available in public repositories.

Fault detection and replacement of a temperature sensor in a cement rotary kiln

This paper proposes a method for fault detection and replacement of the sensor responsible by the measurement of the burning zone temperature in a rotary cement kiln. The control of the burning zone temperature is crucial for the control of kiln temperature and therefore for the control of the cement quality, pollutant emissions, and consumed energy. However the flying dust within the kiln can block the pyrometer sensor, causing faults in the temperature sensor. Exploring the analytical redundancy that usually exist in industrial processes, the proposed methodology uses a neural network trained using an online sequential extreme learning machine to online construct a model to estimate the burning zone temperature. Using the error between the measured and estimated temperatures, faults in the measurement can be detected and thus the replacement of the measured temperatures by the estimated output is made.

Genetically optimized extreme learning machine

This paper proposes a learning algorithm for single-hidden layer feedforward neural networks (SLFN) called genetically optimized extreme learning machine (GO-ELM). In the GO-ELM, the structure and the parameters of the SLFN are optimized by a genetic algorithm (GA). The output weights, like in the batch ELM, are obtained by a least squares algorithm, but using Tikhonovs regularization in order to improve the SLFN performance in the presence of noisy data. The GA is used to tune the set of input variables, the hidden-layer configuration and bias, the input weights and the Tikhonovs regularization factor. The proposed method was applied and compared with four other methods over five benchmark problems available in a public repository. Besides it was applied in the estimation of the temperature at the burning zone of a real cement kiln plant.

Co-evolutionary genetic Multilayer Perceptron for feature selection and model design

This paper proposes a method for Soft Sensors design using a Multilayer Perceptron model based on co-evolutionary genetic algorithms, called CEV-MLP. This method jointly and automatically selects the best input variables and the best configuration of the network for the prediction setting. The CEV-MLP is constituted by three levels, the first level selects the best input variables and respective delays set, the second level is composed by the parameters of hidden layers to be optimized (number of neurons in the hidden layers and transfer function), and the third level is the combination of first and second level. The method was successfully applied, and compared with two state-of-the-art methods, in three real datasets. In all the experiments, the proposed method shows the best approximation accuracy, while all the design of the prediction setting is performed automatically.

Self-adaptive Takagi-Sugeno model identification methodology for industrial control processes

A novel adaptive evolving Takagi-Sugeno (T-S) model identification method is investigated and integrated in a control architecture to control of nonlinear processes is investigated. The proposed system identification approach consists of two main steps: antecedent T-S fuzzy model parameters identification and consequent parameters identification. First, a new unsupervised fuzzy clustering algorithm (NUFCA) is introduced to combine the K-nearest neighbor and fuzzy C-means methods into a fuzzy modeling method for partitioning of the input-output data and identifying the antecedent parameters of the fuzzy system. Then, a recursive procedure using a particle swarm optimization (PSO) algorithm is exploited to construct an online fuzzy model identification and adaptive control methodology. For better demonstration of the robustness and efficiency of the proposed methodology, it is applied to the identification of a model for the estimation of the flour concentration in the effluent of a real-world wastewater treatment plant (WWTP), and identification and control, using a generalized predictive controller (GPC), of a real experimental setup composed of two coupled DC motors. The results show that the developed evolving T-S fuzzy model methodology can identify nonlinear systems satisfactorily and can be successfully used for a prediction model of the process for the GPC.

Predicting gas emissions in a cement kiln plant using hard and soft modeling strategies

In this work, two alternative methodologies for modeling and predicting gas emissions of NO, NO2 and SO2 are presented. The first method involves hard modeling strategies with Parsimonious Multivariate Least Squares (PMLS) assuming simple polynomial functions as base model. The second is a soft modeling approach using Extreme Learning Machine (ELM). In this work we found that both methods have similar capabilities for data description, providing an in depth analysis of the system under study. Results also reveal further insights in predicting gas emissions and enlighten on which of the factors can be useful for prediction, and consequently for system characterization and emission abatement.

Adaptive identification and predictive control using an improved on-line sequential extreme learning machine

This paper proposes a method for adaptive identification and predictive control using an online sequential extreme learning machine based on the recursive partial least-squares method (OS-ELM-RPLS). OL-ELM-RPLS is an improvement to the online sequential extreme learning machine based on recursive least-squares (OS-ELM-RLS) introduced in [1]. Like in the batch extreme learning machine (ELM), in OS-ELM-RLS the input weights of a single-hidden layer feedforward neural network (SLFN) are randomly generated, however the output weights are obtained by a recursive least-squares (RLS) solution. However, due to multicollinearities in the columns of the hidden-layer output matrix caused by the presence of redundant input variables or by a large number of hidden-layer neurons, the problem of estimation the output weights can become ill-conditioned. In order to circumvent or mitigate such ill-conditioning problem, it is proposed to replace the RLS method by the recursive partial least-squares (RPLS) method. The identification methodology is proposed for two application problems: (1) construction of a inferential model, and (2) the learning of a model for the Generalized predictive control (GPC) algorithm. The integration of the proposed adaptive identification method with the GPC results in an adaptive predictive control methodology. To validate and demonstrate the performance and effectiveness of the proposed methodologies, they are applied on modeling of two public regression data sets and on control of the flow through a simulated valve.

Evolutionary learning of a fuzzy controller for industrial processes

The paper proposes a new framework to learn a Fuzzy Logic Controller (FLC), from data extracted from a process while it is being manually controlled, in order to control nonlinear industrial processes. The learning of the FLC is performed by a hierarchical genetic algorithm (HGA). First, the fuzzy c-means (FCM) clustering algorithm is applied to initialize the HGA population, in order to reduce the computational cost and increase the performance of the HGA. The HGA is composed by five hierarchical levels and it is an automatic tool since it does not require any prior knowledge concerning the structure (e.g. the number of rules) and the database (e.g. antecedent and consequent fuzzy sets) of the FLC, and concerning the selection of the adequate input variables and their respective time delays. After the extraction of the FLC by the proposed method, in order to obtain a better control results, if necessary, the learned FLC can be improved manually by using the information transmitted by a human operator, and/or the learned FLC could be easily applied to initialize the required fuzzy knowledge-base of adaptive controllers. In order to improve the results of the learned FLC, a direct adaptive fuzzy controller is applied. Moreover, the proposed method is applied on control of the dissolved oxygen in an activated sludge reactor within a simulated wastewater treatment plant. The results are presented, showing that the proposed method successfully extracted the parameters of the FLC.

A comparison of adaptive PID methodologies controlling a DC motor with a varying load

This work addresses the problem of controlling unknown and time varying plants for industrial applications. To deal with such problem several Self-Tuning Controllers with a Proportional Integral and Derivative (PID) structure have been chosen. The selected controllers are based on different methodologies, and some use implicit identification techniques (Single Neuron and Support Vector Machine) while the others use explicit identification (Dahlin, Pole placement, Deadbeat and Ziegler-Nichols) based in the Least Squares Method. The controllers were tested on a real DC motor with a varying load. The results have shown that all the tested methods were able to properly control an unknown plant with varying dynamics.