Neuro-Fuzzy systems are the systems that neural networks (NN) are incorporated in fuzzy systems, which can use knowledge automatically by learning algorithms of NNs. They can be viewed as a mixture of local experts. Adaptive Neuro-Fuzzy inference system (ANFIS) is one of the examples of Neuro Fuzzy systems in which a fuzzy system is implemented in the framework of adaptive networks. ANFIS constructs an input-output mapping based both on human knowledge (in the form of fuzzy rules) and on generated input-output data pairs. Effective control for distillation systems, which are one of the important unit operations for chemical industries, can be easily designed with the known composition values. Online measurements of the compositions can be done using direct composition analyzers. However, online composition measurement is not feasible, since, these analyzers, like gas chromatographs, involve large measurement delays. As an alternative, compositions can be estimated from temperature measurements. Thus, an online estimator that utilizes temperature measurements can be used to infer the produced compositions. In this study, ANFIS estimators are designed to infer the top and bottom product compositions in a continuous distillation column and to infer the reflux drum compositions in a batch distillation column from the measurable tray temperatures. Designed estimator performances are further compared with the other types of estimators such as NN and Extended Kalman Filter (EKF). In this study, ANFIS performance is also investigated in the adaptive Neuro-Fuzzy control of a pH system. ANFIS is used in specialized learning algorithm as a controller. Simple ANFIS structure is designed and implemented in adaptive closed loop control scheme. The performance of ANFIS controller is also compared with that of NN for the case under study.

Over the past of decade, the bullwhip effect has increasingly become a popular topic for researchers and practitioners in the area of supply chain management since it negatively influences cost, inventory, reliability and other important business processes in supply chain agents. Although there are many remedies for the bullwhip effect summarised in existing literature, it still occurs in several industries. This is partly because it is difficult to apply the results from existing research which analyse the bullwhip effect mainly in a simple supply chain. In addition, several tools and methodologies developed are used for analysing the bullwhip effect in a simple supply chain with several constraints. Therefore, this research aims to develop a unique simulation approach based on system dynamics modelling and Adaptive Network Based Fuzzy Inference System (ANFIS) for quantifying and reducing the bullwhip effect in a multi-product, multi-stage supply chain. System dynamics modelling which is a powerful simulation approach for studying and managing complex feedback system was selected as a main tool in this research. In addition, ANFIS was implemented in system dynamics modelling in order to increase the reliability of a system dynamics model for modelling soft variables. The proposed model covers variables influencing the bullwhip effect which are the structure of supply chain network, supply chain contributions and supply chain performances. As a result, a two layer simulation with three generic models was developed. The flexibility of this proposed model is the ability to model various types of ordering policies which are basic inventory policies, Material requirement planning (MRP) system and Just in time (JIT) approach. Three actual manufacturing supply chains were used as case studies to validate and demonstrate the flexibility of the model developed in this research. This model satisfactorily quantifies the bullwhip effect and the bullwhip effect levels identified in these case studies are significantly decreased by using the proposed simulation model. The successful results indicate that the model can be a useful alternative tool for supply chain managers to quantify and reduce the bullwhip effect in multi-product, multi-stage supply chains.

Submitted by ADRIANO DE SOUZA MARQUES (adrianosmarques@gmail.com) on 2018-07-24T20:31:46Z No. of bitstreams: 1 TESE-FINAL- OK - ADRIANO DE SOUZA MARQUES.pdf: 4099150 bytes, checksum: 7d637f84bcd1d9458ad52467f0e5bfe8 (MD5) / Approved for entry into archive by Ana Lucia de Grava Kempinas (algkempinas@fca.unesp.br) on 2018-07-25T11:31:18Z (GMT) No. of bitstreams: 1 marques_as_dr_botfca.pdf: 4099150 bytes, checksum: 7d637f84bcd1d9458ad52467f0e5bfe8 (MD5) / Made available in DSpace on 2018-07-25T11:31:18Z (GMT). No. of bitstreams: 1 marques_as_dr_botfca.pdf: 4099150 bytes, checksum: 7d637f84bcd1d9458ad52467f0e5bfe8 (MD5) Previous issue date: 2018-05-30 / Neste trabalho é realizado um estudo para estimar a transmissividade da radiação global (Ktβh) e a fração difusa (Kdβh) incidentes em uma superfície com inclinação de 22,85° na base horária utilizando técnicas de aprendizado de máquina (TAM), a partir de dados obtidos no período de 1998 a 2001 em Botucatu/SP/Brasil. As estimativas foram realizadas usando uma série de combinações de variáveis astronômicas e geográficas por meio de três técnicas de redes neurais artificiais (RNA) do tipo Perceptron Multicamadas (MLP), Função de Base Radial (RBF) e Regressão Generalizada (GRNN) e do Sistema Adaptativo de Inferência Neuro Fuzzy (ANFIS). Como referência foram elaborados modelos estatísticos (ME) de regressão linear e polinomial. No Capítulo 1 as estimativas de (Ktβh) foram realizadas por combinações de variáveis medidas e calculadas a partir da irradiação global na superfície horizontal (HgH) e nas estimativas de (Kdβh) utilizou-se combinações de variáveis medidas e calculadas a partir de (HgH) e da irradiação global na superfície inclinada (Hgβ). No Capítulo 2 as estimativas de (Kdβh) foram realizadas por combinações de variáveis medidas e calculadas a partir das irradiações difusa (HdH) e global (HgH) obtidas na superfície horizontal. Os indicadores estatísticos r (correlação), RMSE(%) (precisão) e MBE(%) (exatidão) foram utilizados para avaliar os resultados das estimativas. No capítulo 1 os melhores resultados nas estimativas de (Ktβh) a partir das combinações realizadas com (HgH) foram: MLP - RMSE=3,73%; RBF - RMSE=3,99%; GRNN - RMSE=5,27%; ANFIS - RMSE=3,78% e ME - RMSE=6,65%. Nesse caso os indicadores de precisão mostram uma redução de aproximadamente 44% com o uso da técnica (MLP) em comparação ao modelo estatístico (ME). Nas estimativas de (Kdβh) a partir das combinações de (HgH) os melhores resultados foram: MLP - RMSE=21,69%; RBF - RMSE=25,43%; GRNN - RMSE=29,39%; ANFIS - RMSE=23,08% e - ME - RMSE=35,35%. Da mesma forma os indicadores de precisão mostram uma redução de aproximadamente 39% com o uso da técnica (MLP) em comparação ao modelo estatístico (ME). E nas estimativas de (Kdβh) a partir das combinações realizadas com (Hgβ) os melhores resultados foram: MLP - RMSE=20,32%; RBF - RMSE=21,95%; GRNN - RMSE=29,11%; ANFIS - RMSE=21,75% e ME - RMSE=36,48%. Os indicadores de precisão mostram uma redução de aproximadamente 44% com o uso da técnica (MLP) em comparação ao modelo estatístico (ME). No capítulo 2 as melhores estimativas de (Kdβh) a partir das combinações realizadas com (HdH) foram: MLP - RMSE=4,03%; RBF - RMSE=5,84%; GRNN - RMSE=10,85%; ANFIS - RMSE=4,15% e ME - RMSE=12,42%. Os indicadores de precisão mostram uma redução de aproximadamente 67% com o uso da técnica (MLP) em comparação ao modelo estatístico (ME). Nas estimativas de (Kdβh) a partir de (HgH) os melhores resultados foram: MLP - RMSE=21,69%; RBF - RMSE=25,43%; GRNN - RMSE=29,39%; ANFIS - RMSE=23,08% e ME - RMSE=35,35%. Os indicadores de precisão mostram uma redução de aproximadamente 39% com o uso da técnica (MLP) em comparação ao modelo estatístico (ME). Os resultados mostram que a técnica de rede neural artificial MLP apresentou os melhores índices em todas as estimativas de (Ktβh) e (Kdβh) com reduções significativas quando comparadas aos resultados obtidos com as estimativas obtidas com os modelos estatísticos. Pela análise dos resultados é possível observar que o uso das técnicas de aprendizado de máquina (TAM) nas combinações de variáveis propostas e com os dados obtidos de Botucatu/SP, se apresentam como alternativa aos modelos estatísticos (ME) para estimar as variáveis de (Ktβh) e (Kdβh). / In this work, a study was carried out to estimate the transmissivity of the global radiation (Ktβh) and the diffuse fraction (Kdβh) incident on a surface with slope of 22.85 ° in the hourly basis using machine learning techniques (MLT), from data obtained from 1998 to 2001 in Botucatu / SP / Brazil. The estimates were made using a series of combinations of astronomical and geographic variables by means of three artificial neural network (ANN) techniques such as MultLayer Perceptron (MLP), Radial Basis Functions Networks (RBF) and Generalized Regression Neural Network (GRNN) Adaptive Neuro Fuzzy Inference System (ANFIS). Statistical models (SM) of linear and polynomial regression were elaborated as reference. In Chapter 1 estimates of (Ktβh) were performed by combinations of variables measured and calculated from global horizontal surface irradiation (HgH) and estimates of (Kdβh) combinations of variables measured and calculated from (HgH) and global radiation on the sloped surface (Hgβ). In Chapter 2 estimates of (Kdβh) were performed by combinations of variables measured and calculated from the diffuse (HdH) and global (HgH) irradiances obtained on the horizontal surface. The statistical indicators r (correlation), RMSE (%) (precision) and MBE (%) (accuracy) were used to evaluate the results of the estimates. In Chapter 1 the best results in the estimates of (Ktβh) from the combinations performed with (HgH) were: MLP - RMSE = 3.73%; RBF - RMSE = 3.99%; GRNN - RMSE = 5.27%; ANFIS-RMSE = 3.78% and SM - RMSE = 6.65%. In this case the precision indicators show a reduction of approximately 44% with the use of the technique (MLP) in comparison to the statistical model (SM). In the estimates of (Kdβh) from the combinations of (HgH) the best results were: MLP - RMSE = 21.69%; RBF - RMSE = 25.43%; GRNN - RMSE = 29.39%; ANFIS - RMSE = 23.08% and SM - RMSE = 35.35%. Likewise, the precision indicators show a reduction of approximately 39% with the use of the technique (MLP) in comparison to the statistical model (SM). And in the estimates of (Kdβh) from the combinations performed with (Hgβ) the best results were: MLP - RMSE = 20.32%; RBF - RMSE = 21.95%; GRNN - RMSE = 29.11%; ANFIS - RMSE = 21.75% and SM - RMSE = 36.48%. The precision indicators show a reduction of approximately 44% with the use of the technique (MLP) in comparison to the statistical model (SM). In Chapter 2 the best estimates of (Kdβh) from the combinations performed with (HdH) were: MLP - RMSE = 4.03%; RBF - RMSE = 5.84%; GRNN - RMSE = 10.85%; ANFIS - RMSE = 4.15% and SM - RMSE = 12.42%. The precision indicators show a reduction of approximately 67% with the use of the technique (MLP) in comparison to the statistical model (SM). In the estimates of (Kdβh) from (HgH) the best results were: MLP - RMSE = 21.69%; RBF - RMSE = 25.43%; GRNN - RMSE = 29.39%; ANFIS - RMSE = 23.08% and SM - RMSE = 35.35%. The precision indicators show a reduction of approximately 39% with the use of the technique (MLP) in comparison to the statistical model (SM). The results show that the artificial neural network MLP technique presented the best indexes in all estimates of (Ktβh) and (Kdβh) with significant reductions when compared to the results obtained with the estimates obtained with the statistical models. By the analysis of the results it is possible to observe that the use of the machine learning techniques (MLT) in the combinations of proposed variables and the data obtained from Botucatu / SP, are presented as an alternative to the statistical models (SM) to estimate the variables of (Ktβh) and (Kdβh).

Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2017-11-01T21:47:48Z No. of bitstreams: 1 JoseKleitonEwertonDaCostaMartins_DISSERT.pdf: 3450112 bytes, checksum: 43beab3e6259be22eeed518fd67eaefc (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2017-11-08T21:43:16Z (GMT) No. of bitstreams: 1 JoseKleitonEwertonDaCostaMartins_DISSERT.pdf: 3450112 bytes, checksum: 43beab3e6259be22eeed518fd67eaefc (MD5) / Made available in DSpace on 2017-11-08T21:43:16Z (GMT). No. of bitstreams: 1 JoseKleitonEwertonDaCostaMartins_DISSERT.pdf: 3450112 bytes, checksum: 43beab3e6259be22eeed518fd67eaefc (MD5) Previous issue date: 2017-06-23 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior (CAPES) / O trabalho faz uma an?lise de diferentes t?cnicas de controle na estrutura do ANFIS modificado, m?todo recente que se originou a partir de uma altera??o na estrutura do ANFIS, para realizar identifica??o e controle de plantas com ampla faixa de opera??o e n?o linearidade acentuada. O ANFIS modificado ? dividido em dois grandes est?gios, o primeiro sendo a identifica??o e o segundo o controle. Para realizar a identifica??o pode-se utilizar quaisquer t?cnicas. Nesse trabalho foram exploradas as t?cnicas de identifica??o de sistemas lineares mais conhecidas na literatura e o m?todo dos m?nimos quadrados. Assim como no est?gio da identifica??o, o est?gio de controle tamb?m permite utilizar quaisquer t?cnicas de projeto. Nesse trabalho foram exploradas as t?cnicas de sintonia de controladores PID mais conhecidas na literatura, na qual os controladores projetados foram incorporados na estrutura do ANFIS modificado para a obten??o de um controlador global n?o linear. Foi escolhido um sistema de tanques com multisse??es como estudo de caso e assim foi realizada a sua identifica??o atrav?s do ANFIS modificado, mostrando as qualidades do m?todo. Em seguida foi realizada uma compara??o de desempenho do ANFIS modificado utilizando os diferentes m?todos de sintonia e ao final chegando a uma metodologia sistem?tica para utiliza??o do ANFIS modificado como controlador global. / This work makes an analysis of different control techniques in the modified ANFIS structure, this method is recent and originated from a change in the ANFIS structure for perform identification and control of plants with wide operating range and accentuated non-linearity. The modified ANFIS is divided into two major stages, the first is the identification and the second is the control. In order to perform the identification, it is possible to use any techniques. In this work was explored the linear system identification more known in the literature and the least square estimation. As in the identification stage, the control stage can also use any techniques. This work the tuning of PID controllers will be explored, in which the designed controllers will be incorporated into the modified ANFIS structure to obtain a non-linear controller. A system of tanks with multisections was chosen as a case study and its identification through the modified ANFIS was performed, showing the qualities of the method. Then a performance comparison of the modified ANFIS will be performed using the different tuning methods and show a systematic methodology for use the modified ANFIS as global controller.

Cognitive radio is a intelligent technology that helps in resolving the issue of spectrum scarcity. In a spectrum sharing network, where secondary user can communicate simultaneously along with the primary user in the same frequency band, one of the challenges in cognitive radio is to obtain balance between two conflicting goals that are to minimize the interference to the primary users and to improve the performance of the secondary user. In our thesis we have considered a primary link and a secondary link (cognitive link) in a fading channel. To improve the performance of the secondary user by maintaining the Quality of Service (Qos) to the primary user, we considered varying the transmit power of the cognitive user. Efficient utilization of power in any system helps in improving the performance of that system. For this we proposed ANFIS based opportunistic power control strategy with primary user’s SNR and primary user’s channel gain interference as inputs. By using fuzzy inference system, Qos of primary user is adhered and there is no need of complex feedback channel from primary receiver. The simulation results of the proposed strategy shows better performance than the one without power control. Initially we have considered propagation environment without path loss and then extended our concept to the propagation environment with path loss where we have considered relative distance between the links as one of the input parameters.

No description available.

In electric power distribution systems, the major determinant in electricity supply strategy is the quantity of demand. Customers need to be accurately represented using updated nodal load information as a requirement for efficient control and operation of the distribution network. In Distribution Load Estimation (DLE), two major categories of data are utilized: historical data and direct real-time measured data. In this thesis, a comprehensive survey on the state-of-the-art methods for estimating loads in distribution networks is presented. Then, a novel method for representing historical data in the form of Representative Load Curves (RLCs) for use in realtime DLE is also described. Adaptive Neuro-Fuzzy Inference Systems (ANFIS) is used in this regard to determine RLCs. An RLC is a curve that represents the behavior of the load during a specified time span; typically daily, weekly or monthly based on historical data. Although RLCs provide insight about the variation of load, it is not accurate enough for estimating real-time load. This therefore, should be used along with real-time measurements to estimate the load more accurately. It is notable that more accurate RLCs lead to better real-time load estimation in distribution networks. This thesis addresses the need to obtain accurate RLCs to assist in the decision-making process pertaining to Radial Distribution Networks (RDNs).This thesis proposes a method based on Adaptive Neuro-Fuzzy Inference Systems (ANFIS) architecture to estimate the RLCs for Distribution Networks. The performance of the method is demonstrated and simulated, on a test 11kV Radial Distribution Network using the MATLAB software. The Mean Absolute Percent Error (MAPE) criterion is used to justify the accuracy of the RLCs.

Active research of SMAs has shown that its Seebeck coefficient is sensitive to its martensitic phase transformation and has the potential to determine the SMAs state of transformation. The combination of Shape Memory Alloys, which have a positive Seebeck coefficient, and Constantan which has a negative Seebeck coefficient (-35 mV/K) results in a thermocouple capable of measuring temperature. The work presented in this thesis is based on the development and design of this sensor. This sensor is used to study the hysteretic behaviour of SMAs. Although Shape Memory Alloys (SMAs) exhibit a myriad of nonlinearities, SMAs show two major types of nonlinear hysteresis. During cyclic loading of the SMAs, it is observed that one type of hysteretic behavior depends on the rate of heating the SMAs, whilst the variation of maximum temperature of an SMA in each cycle results in the other hysteretic behavior. This later hysteretic behavior gives rise to major and minor nonlinear loops of SMAs. The present work analyzes the nonlinearities of hysteretic envelopes which gives the different maximum temperatures reached for each hysteretic cycle with respect to stress and strain of the SMA. This work then models this behavior using Adaptive Neuro Fuzzy Inference System (ANFIS) and compares it to experimental results. The nonlinear learning and adaptation of ANFIS architecture makes it suitable to model the temperature path hysteresis of SMAs. / Master of Science

This research presents an investigation into the development of real time system identification using intelligent algorithms. A simulation platform of a flexible beam vibration using finite difference (FD) method is used to demonstrate the real time capabilities of the identification algorithms. A number of approaches and algorithms for on line system identifications are explored and evaluated to demonstrate the merits of the algorithms for real time implementation. These approaches include identification using (a) traditional recursive least square (RLS) filter, (b) Genetic Algorithms (GAs) and (c) adaptive Neuro_Fuzzy (ANFIS) model. The above algorithms are used to estimate a linear discrete second order model for the flexible beam vibration. The model is implemented, tested and validated to evaluate and demonstrate the merits of the algorithms for real time system identification. Finally, a comparative performance of error convergence and real time computational complexity of the algorithms is presented and discussed through a set of experiments.

In this dissertation we address three different sustainability concepts: [1] modeling of biodiesel production via heterogeneous catalysis, [2] life cycle analysis for pyrolysis of switchgrass for using in power plant, and [3] modeling of pyrolysis of biomass. Thus we deal with Specific Aim 1, 2 and 3. In Specific Aim 1, the models for esterification in biodiesel production via heterogeneous catalysis were developed. The models of the reaction over the catalysts were developed in two parts. First, a kinetic study was performed using a deterministic model to develop a suitable kinetic expression; the related parameters were subsequently estimated by numerical techniques. Second, a stochastic model was developed to further confirm the nature of the reaction at the molecular level. The deterministic and stochastic models were in good agreement. In Specific Aim 2, life cycle analysis and life cycle cost for pyrolysis of switchgrass for using in power plant model were developed. The greenhouse gas (GHG) emission for power generation was investigated through life cycle assessment. The process consists of cultivation, harvesting, transportation, storage, pyrolysis, transportation and power generation. Here pyrolysis oil is converted to electric power through co- combustion in conventional fossil fuel power plants. The conventional power plants which are considered in this work are diesel engine power plant, natural gas turbine power plant, coal-fired steam-cycle power plant and oil-fired steam-cycle power plant. Several scenarios are conducted to determine the effect of selected design variables on the production of pyrolysis oil and type of conventional power plants. In Specific Aim 3, pyrolysis of biomass models were developed. Since modeling of pyrolysis of biomass is complex and challenging because of short reaction times, temperatures as high as a thousand degrees Celsius, and biomass of varying or unknown chemical compositions. As such a deterministic model is not capable of representing the pyrolysis reaction system. We propose a new kinetic reaction model, which would account for significant uncertainty. Specifically we have employed fuzzy modeling using the adaptive neuro-fuzzy inference system (ANFIS) in order to describe the pyrolysis of biomass. The resulting model is in better agreement with experimental data than known deterministic models. / Ph. D.