LEE- Kontrol ve Otomasyon Mühendisliği-Doktora
Bu koleksiyon için kalıcı URI
Gözat
Son Başvurular
1 - 5 / 19
-
ÖgeAnalysis and design of robust disturbance observers(Graduate School, 2023-09-04)Robustness has been one of the most defining features of control systems since the classical control period. In the early days, the robustness of the control system was expressed using concepts like phase margin and gain margin, adapted from telecommunications engineering, and this terminology was faithfully used during the period when the significant achievements of modern control theory were demonstrated. However, by the end of the 70s, two separate developments marked the beginning of the golden age of robust control theory. The first of the developments that heralded this new era is Kharitonov's theorem, which established a new field of research for examining the stability of systems with parametric uncertainty. The other is John Doyle's demonstration that even in a single-input, single-output system, the LQG regulator does not have any guaranteed robustness margin, unlike the LQ regulator. While the first formed the basis of the research field known as the parametric approach, the other was one of the precursors of the $H_{\infty}$ theory. Since then, robust control has been seen as an independent sub-branch of control theory. Both approaches reached their peak with both theoretical and practical applications throughout the 1980s and 1990s. On the other hand, it has been shown that more robust closed-loop systems can be developed by changing the structure of the controller. One of the prominent methods is the approach known in the literature as the disturbance-observer (DOB). This approach, which enables the prediction and cancellation of disturbances and uncertainties that impact the system at its input, has been widely implemented, particularly in practical applications. On the other hand, the theoretical limits of the method, its analysis under uncertainty, and its design with newly developed robust control methods have lagged behind practical applications. Although theoretical studies have been carried out especially with the $H_{\infty}$ approach since the 2000s, DOB design and analysis under parametric uncertainties have not attracted the attention of researchers sufficiently. The main purpose of this thesis is to develop new approaches for both the analysis and design of disturbance observers under parametric uncertainties. In the analysis of systems with parametric uncertainty, how the uncertainties are modeled is the factor that directly affects the analysis method. In Kharitonov's paradigm, the parametric uncertainty bounding set is usually expressed as a box, which corresponds to the $l_{\infty}$ representation of the parameter box. However, the $l_{2}$ analog of the same representation is also possible. In fact, this representation is more suitable for the situation where the mathematical model is obtained by linear or nonlinear regression methods under system identification approach. Based on this, in the first part of the thesis, the answer to the question of "How much uncertainty can be tolerated with the DOB structure?", has been sought. Although approaches in the frequency domain produce effective results for DOB analysis, new challenges arise when the problem is expressed in the state space. Two approaches have come to the fore for examining parametric uncertainties in the state space. The first of these is to move the problem to the frequency domain where there are theorems and mathematical tools mature enough to examine parametric uncertainties. However, when this method is utilized, even the simplest interval system matrices show themselves as a affine-linear or more complex polynomial when expressed as a polynomial. Therefore, design in state space was seen as a "hard nut to crack" problem, in Yedevalli's words, and pushed control theorists to different research directions. The other method is to consider the problem directly in the state space. Although similar difficulties exist in this approach, when designing directly in the state space, the use of proven state space methods is also possible. Although new solutions are proposed, especially under the concept of quadratic stability, the nature of the problem condemns control theorists to use conservative approaches. In addition, a suitable Lyapunov function has not yet been proposed in the case where the design regions used to limit the parametric uncertainties are disjoint. The second contribution put forward within the scope of the thesis is the guardian-map approach, which offers less conservative disturbance observer design. Thanks to the method, robustness criteria can be assigned for each nominal eigenvalue separately and the disturbance observer is designed to meet this criterion. In this way, the inherent trade-off between robustness of the disturbance observer and the disturbance observer bandwidth is decided according to whether the closed-loop system satisfies the previously determined eigenvalue spread criterion. Advantages of considering the problem in state space include the possibility to use LMI tools and the incorporation of useful methods such as eigenstructure assignment into the solution of the problem. Many control problems can be expressed in LMI form, and these LMIs can be formulated as appropriate convex optimization problems. The LMI framework is particularly useful for expressing parametric uncertainties and constraining eigenvalue spread. However, when the dominant methods in the literature are examined, the design regions defined by the LMI approach are not defined separately for each eigenvalue, but a combined LMI design region is defined for all eigenvalues. This situation complicates the eigenvalue assignment problem and does not allow defining different robustness criteria between the eigenvalues in the non-dominant region, which is less important for the design, and the dominant region eigenvalues, which determine the behavior of the system. In addition, when the eigenstructure assignment methods are considered, the methods for minimizing the sensitivity of the system dominate the literature, instead of expressing the parametric uncertainties directly. Although robust eigenstructure assignment methods based on $H_{\infty}$-based approaches have been proposed, eigenstructure assignment methods have not been sufficiently studied in direct parametric uncertainty system design. In the eigenstructure assignment methods, since the eigenvalues are assigned strictly at the beginning of the design, the vector space to which the eigenvectors can be assigned in the rest of the design is also limited. In order to overcome this, although methods such as regional assignment, partial eigenvalue assignment and loose eigenstructure assignment are suggested in the literature, suppressing the effect of parametric uncertainties has not been the primary design criterion in these approaches. In order to fill these gaps in the literature, a new design method has been proposed, and in this approach, the robustness of the system to parametric uncertainties has been made the primary criterion of the design, and a novel disturbance observer design method has been proposed by using eigenstructure assignment and LMI approaches together for this purpose. The approach does not require any heuristic algorithms or global optimization methods, as well as allowing the solution of the robust root clustering problem for disjoint design regions. As a result, the method inevitably suffers from conservatism. However, the design reduces the problem of finding robust eigenvectors to finding the appropriate one among a finite number of eigenvectors. As a conclusion, within the scope of this thesis, a method is proposed to examine the robustness of the disturbance observer under parametric uncertainties, and two new design methods are proposed to limit the eigenvalue spread in the state space within the disjoint design regions determined for each nominal eigenvalue. By using the obtained results, a disturbance observer in the state space is designed for systems with parametric uncertainty and the results are shared.
-
ÖgeDesign and deployment of deep learning based fuzzy logicsystems(Graduate School, 2023-08-29)In the past decade, significant progress has been made in the field of Deep Learning (DL), driven by innovative learning methods, novel layer structures, and the use of graphics cards for enhanced processing power. This progress has led to the training of neural network models with numerous hidden layers and neurons, resulting in breakthroughs in various domains such as semantic segmentation, object detection, and classification. Deep Neural Networks (DNNs) have proven to be highly effective in machine learning and artificial intelligence applications. DNNs offer advantages over traditional machine learning techniques, including the ability to learn features at multiple layers, which allows them to capture complex features of input data. Through forward pass and backpropagation, DNNs extract meaningful features and outperform other methods in many tasks. As a result, DNNs have gained popularity and are widely used in commercial and industrial applications, contributing to advancements in machine learning. Fuzzy Logic Systems (FLSs) have been employed to various fields and applications over the last years. FLSs use linguistic Fuzzy Sets (FSs) and fuzzy rules, enabling the modeling of human-like reasoning and decision-making processes. This has led to advancements in the development of intelligent control systems capable of effectively handling nonlinear and uncertain dynamics. Besides, FLSs have been applied in image processing, leveraging the FSs to represent uncertain data. FLSs provide robust image analysis, pattern recognition, and image understanding, contributing to advancements in computer vision and image processing applications. Overall, FLSs have been extensively utilized in modeling various systems and phenomena. Their ability to handle uncertainty provides a flexible and interpretable modeling approach, capturing complex relationships and uncertainties in real-world systems. Conventional FLSs, known as Type-1 FLSs (T1-FLSs), have limitations in representing uncertainty. To address this, Type-2 Fuzzy Sets (T2-FSs) have been introduced as an alternative, offering a more flexible representation. T2-FSs can better handle nonlinear and uncertain systems, and T2-Fuzzy Logic Systems (T2-FLSs) have the potential to handle complex problems. However, learning T2-FLSs presents challenges due to their design complexity and the need to learn the parameters associated with fuzzy sets. Different approaches have been proposed, including adapting pre-trained T1-FLSs to T2-FLSs and employing evolutionary algorithms or Neural Network (NN) approaches to optimize the parameters of Interval T2-FLSs (IT2-FLSs). These approaches aim to simplify the design complexity and improve the performance of T2-FLSs. Despite advancements, integrating neural networks and evolutionary algorithms with T2-FLSs faces challenges when applied to extensive datasets. The curse of dimensionality and the increasing number of parameters in T2-FLSs brings some difficulties that is not possible to solve with the current approaches. Recent research has focused on combining FLSs and deep neural networks to overcome these challenges, leading to the development of hybrid models that leverage the strengths of both generalization capabilities of the DNNs and the power of the mini-batch sampled optimization algorithms. In this thesis study, a novel approach is proposed to learn the parameters of T2-FLSs using deep learning-based parameter learning methods. The proposed approach aims to handle extensive datasets and construct models with both a good prediction accuracy and the ability to handle the uncertainties. In the scope of this thesis, specifically, three studies are conducted: the first study (i) is titled with "Learning with Type-2 Fuzzy Activation Functions to Enhance the Performance of Deep Neural Networks", in the second study (ii), we propose a framework which is titled with "More Than Accuracy: A Composite Learning Framework for Interval Type-2 Fuzzy Logic Systems" and in the (iii) last study, we propose reliable uncertainty quantification for GT2-FLSs named as "Towards Reliable Uncertainty Quantification and High Precision with General Type-2 Fuzzy Systems". In the first study (i), we introduce a new method called IT2 Fuzzy Activation Layer (IT2-FAL) that aims to enhance the learning performance of DNNs. The IT2-FAL consists of Single Input IT2 (SIT2) Fuzzy Rectifying Units (FRUs) which used as activation units within the DNN structure to improve learning capabilities. We construct a closed-form representation of the SIT2-FRU structure, and an analysis is conducted to understand how the parameters of this structure influence the generation of input-output mappings. The research findings demonstrate that these mappings can be regarded either as hyperparameters to be set or as parameters to be learned. We provide a learning algorithm to these hyperparameters using DL based frameworks. To evaluate the effectiveness of the proposed IT2-FAL, a comparison is made against existing activation units like ReLU, PReLU, and ELU. The novel SIT2-FRU not only addresses the vanishing gradient problem but also exhibits a fast convergence rate. It achieves this by pushing the mean activation close to zero through the processing of inputs defined in the negative quadrant. This property of SIT2-FRU enables DNNs to exhibit improved learning behavior. The experiments conducted using the selected benchmark datasets show the efficiency and superiority of the IT2-FAL approach. By incorporating the IT2-FAL and its activation units (SIT2-FRU components), DNNs can enhance their learning capabilities and benefit from a more robust and flexible network structure. The proposed approach has the potential to improve the performance of DNNs as the experimental results revealed and it also gives opportunity to enhance the learning capabilities of DNNs. The second study (ii) introduces a novel composite learning approach that utilizes type-reduced sets of Interval Type-2 Fuzzy Logic Systems (IT2-FLSs) to capture uncertainty and establish Prediction Intervals (PIs). Unlike mainstream training approaches that primarily focus on accuracy, the objective of this new approach is to not only achieve high prediction accuracy but also effectively address and capture uncertainty by exploiting the type-reduced sets of IT2-FLSs. In order to achieve such a goal, we identify three main challenges in this context: (1) the capability to handle uncertainty, (2) the construction of a composite loss function, and (3) the development of a learning algorithm that addresses the training complexity while considering the definitions of IT2-FLSs. In (1), to address these challenges, the proposed approach exploits the type-reduced set of IT2-FLSs by combining quantile regression and DL parameter learning methods with IT2-FLSs. The ability of IT2-FLSs to process uncertainty depends on the methods employed for calculating the center-of-sets, while their representation capability is determined by the structure of their antecedent and consequent membership functions. In the scope of thesis, we introduce various parametric IT2-FLSs and defines the learnable parameters for all IT2-FLSs, along with their constraints that need to be satisfied during the training process. In (2), the construction of the loss function is defined which involves construction of a multi-objective loss that is subsequently converted into a constrained composite loss. This composite loss comprises the log-cosh loss component, which aims to optimize accuracy, and a tilted loss component that focuses on the representation of uncertainty. Notably, the tilted loss explicitly utilizes the type-reduced set. In (3), a DL approach is presented for training IT2-FLSs using unconstrained optimizers. The study also introduces parameterization techniques to convert the constrained optimization problem of IT2-FLSs into an unconstrained one without violating the definitions of fuzzy sets. In order to evaluate the effectiveness of the proposed approach, comprehensive comparative results are provided. In the thesis, we provide a hyperparameter sensitivity analysis and inter/intra-model comparisons conducted on various benchmark datasets. These evaluations shed light on the performance and robustness of the proposed novel approach in handling uncertainty and achieving high prediction accuracy for regression problems. In the third study (iii), we present a new learning approach for 𝛼-plane based General Type-2 Fuzzy Logic Systems (GT2-FLSs) to improve pointwise prediction accuracy and generate reliable Prediction Intervals (PIs). The approach focuses on exploiting the shape and size of the Secondary Membership Functions (SMFs) through a novel composite loss function. The novel composite loss function consists of two main components: an uncertainty quantification-focused loss and an accuracy-focused term. Within the uncertainty-focused loss, only the type-reduced set of IT2-FLS associated with the 𝛼0=0 plane, known as the FOU, is explicitly utilized. This allows the SMF size parameters of the GT2-FLS to quantify uncertainty and learn PIs. For the accuracy-focused part, two alternative loss terms are provided. In one approach, the aggregated output of the GT2-FLSs is used directly, while in the other approach, only the output associated with the 𝛼𝐾=1 level is utilized. In both cases, the SMF shape parameters of the GT2-FLS are enforced to enable pointwise prediction with high precision. Thus, different roles are assigned to the IT2-FLS associated with 𝛼-planes within the proposed loss function. Since the output of the 𝛼0=0 plane does not contribute to the output calculation of the GT2-FLS, a partially independent learning of the GT2-FSs becomes possible, allowing for capturing uncertainty while maintaining high accuracy. We present a DL based parameter learning approach for GT2-FLSs to facilitate efficient learning to be able to handle the complex parameter learning problem of the GT2-FLSs and also in the presence of high-dimensional and complex data. This is achieved by defining an unconstrained learning problem. We also proposed novel parameterization tricks such that the definitions of GT2-FSs are not violated. We also provide statistical comparative analyses using benchmark datasets in order to demonstrate the superiority of the proposed learning approach. The results of these analyses show the potential of learning GT2-FLS with the proposed DL based approach as a promising solution for reliable uncertainty quantification with high precision in real-world applications.
-
ÖgeBir metro hattının anahtarlamalı sistem olarak modellenmesi(Graduate School, 2023-08-15)Özellikle büyük şehirlerde yaşanan nüfus artışları insanların bir yerden bir yere ulaşma ihtiyaçlarını arttırmaktadır. Ulaşım için bireysel araçların kullanımı trafik sıkışıklığına neden olmaktadır, bu nedenle trafik sıkışıklıklarını azaltıp konforlu ve etkin bir ulaşım sağlamak için bireysel araçlarla ulaşım yerine toplu taşıma tercih edilmelidir. Toplu taşımanın cazip hale getirilmesi için toplumun ihtiyaçlarını karşılayacak şekilde kurgulanması, ihtiyaç duyulan sefer sıklığını konforlu bir yolculuk ile sunması gerekmektedir. Bir ulaşım sisteminde bulunan istasyon ve araçlardaki yolcu sayıları ile sefer aralıkları arasında doğrusal bir ilişki bulunmaktadır. Sefer aralıklarını azaltmak yolcu konforunu arttırsa da, işletme maliyetlerini arttırmaktadır. Bu nedenle iyi ayarlanmış bir sefer aralığı hem işletmeci hem de yolcular açısından önem kazanmaktadır. Sefer aralığının sağlıklı şekilde ayarlanması için iyi kurgulanmış bir modele ihtiyaç duyulmaktadır. Bu model, hem araçların ve istasyonların konumlarını içermeli, hem de yoğunluğu takip edebilmek için yolcu sayılarını dikkate almalıdır. Sefer aralığının güncellenmesi araçların sistemde bir bölgede toplanmasını engellemek için ilk istasyondan başlatılmalıdır, bu da diğer istasyonlardaki sefer aralıklarının belirli bir zaman gecikmesi ile güncellenmesine neden olmaktadır. Anahtarlamalı sistemler, birden çok sistemin birbirleri arasında bir kurala bağlı olarak geçiş yaptığı sistemler olarak tanımlanmaktadırlar. Bir toplu taşıma ağındaki yolcu sayılarının davranışı bir aracın bir durağa yanaşıp yanaşmamasına göre değişkenlik gösterdiği için, bu sistemler anahtarlamalı sistem olarak ele alınabilirler. Bu nedenle, bu tez çalışmasında, bir metro hattındaki yolcu sayıları anahtarlamalı sistem olarak modellenmiştir. Sistemin matematiksel modeli oluşturulduktan sonra, dinamik model MATLAB Simulink® yazılımında gerçeklenmiştir. Modelin doğruluğunu sınamak için, sistem bir ayrık olay benzetim yazılımı olan Rockwell Arena®'da da gerçeklenmiştir. Her iki modelleme yazılımında yapılan benzetimlerde Metro İstanbul'dan M2 hattı için alınan gerçek yolcu verileri kullanılmıştır. Bu benzetimler sonucunda elde edilen yolcu grafikleri dışarı alınıp karşılaştırılmıştır ve dinamik modelin sonuçlarının ayrık olay benzetimi ile oluşturulan model ile aynı olduğu doğrulanmıştır. Buna ek olarak, dinamik model ve modeli oluşturmaya olanak sağlayan MATLAB Simulink® yazılımının, ayrık olay benzetimi ile modellemeyi sağlayan Rockwell Arena® yazılımına göre benzetimi daha hızlı koşturduğu ortaya konmuştur. Dinamik modelin geçerliliği doğrulandıktan sonra, sistemin kararlılık analizi yapılmıştır. Kararlılık analizi için anahtarlamalı sistemlerin kararlılık analizinde kabul gören iki farklı yöntem seçilmiştir. Sistemde bozucunun olmadığı durumlar için 'ortak Lyapunov', bozucunun olmadığı durumlar için de 'girişten duruma kararlılık analizi' yöntemleri ile analizi yapılmıştır. MATLAB Simulink®'te oluşturulan modelde bozuculu durumlar için sınır koşullarında Metro İstanbul'dan alınan M2 hattı verileri ile koşturulan benzetimlerin sonuçlarıyla kararlılık analiz sonuçları doğrulanmıştır. Dinamik modelin davranışının iyileştirilmesi için yolcu transferlerinin daha dinamik olduğu bir model de kurgulanmıştır ve MATLAB Simulink®'te oluşturulan model bu yaklaşıma göre güncellenip daha gerçekçi bir yolcu davranışı elde edilmiştir. Bu tez çalışmasında kurulan model, literatürdeki yolcu sayılarının dinamik olarak modellenmesindeki boşluğu doldurmuş olup, kurulan modelin, bu tez çalışması sonrasında geliştirilmesi planlanan, istasyon ve trenlerdeki aktif yolcu sayıları ve işletme maliyetlerini dikkate alarak dinamik olarak sefer aralığını güncelleyecek bir kontrolör tasarımını doğrulamak için kullanılması hedeflenmektedir.
-
ÖgeEnergy management of P2 hybrid electric vehicle based on event triggered nonlinear model predictive control and deep Q network(Graduate School, 2024-01-22)The energy management problem of a P2 hybrid electric vehicle (HEV) involves determining how to allocate the available energy between the internal combustion engine and the electric motor, as well as how to use the energy stored in the battery. The goal of the energy management system is to minimize the fuel consumption of the vehicle while maintaining its performance and drivability. The energy management problem of a P2 HEV is challenging because it involves multiple objectives and constraints, as well as uncertain and varying driving conditions. The system must balance the power demand of the vehicle with the available power from the engine and battery, while also considering factors such as the state of charge of the battery, the efficiency of the components, and the driving cycle of the vehicle. In this study, the P2 Hybrid Electric Vehicle (HEV) model in Simscape includes components such as the internal combustion engine, electric motor, battery, transmission, and vehicle dynamics. The Kia Niro 2018 vehicle specification provides information about the characteristics of these components, such as their power ratings, efficiencies, and physical dimensions. By combining this information with the Simscape model, the behavior of the P2 HEV is simulated under different driving conditions. The Simscape model is based on physical equations and principles, which means that it provides accurate and reliable predictions of the behavior of the P2 HEV. The model is used to analyze the performance of the vehicle under different scenarios, such as different driving cycles or changes in environmental conditions. By using the Kia Niro 2018 vehicle specification as a reference, the P2 HEV model is validated and adjusted to improve its accuracy. After that, to track the desired velocity profile for a P2 hybrid electric vehicle (HEV) based on the World Harmonized Light Vehicle Test Procedure (WLTP), first model predictive controller (MPC) is implemented. The MPC uses a mathematical model of the vehicle dynamics and powertrain components to predict future behavior over a certain time horizon, taking into account acceleration limits according to ISO 2631-5. The reference signal is determined based on the WLTP standard velocity profile, and an objective function is defined to minimize deviation from the reference signal. In addition to tracking the desired velocity profile, the torque distribution between the engine and motor in a P2 HEV is controlled using a second MPC. The MPC uses a mathematical model of the vehicle's powertrain components to predict future behavior over a certain time horizon, taking into account physical limits of the battery, engine, and motor. The objective of this MPC is to distribute the torque between the engine and motor in an optimal way to achieve the desired performance metrics, such as minimizing power losses. Constraints are established on the system, such as maximum and minimum torque of the engine and electric motor, state of charge of the battery, and current limits of the battery and total torque equality. To decrease the computational cost, an event-triggered mechanism is constructed in a P2 HEV energy management system using a Deep Q Network (DQN) algorithm. This mechanism triggers the model predictive controllers only when needed, reducing the computational burden and improving the energy efficiency of the system. The DQN algorithm is used to learn a policy that determines when to trigger the torque distribution MPC based on the current state of the system. The algorithm uses a neural network to estimate the value function and select actions that minimize the expected long-term cost. The event-triggered mechanism provides a flexible and adaptive approach to energy management in the P2 HEV, allowing for real-time adjustments based on changing driving conditions. The use of DQN allows for efficient and effective decision-making, improving the overall performance and efficiency of the P2 HEV. As a last, in a P2 hybrid electric vehicle (HEV) energy management system with two model predictive controllers (MPCs), the weights of the second MPC's cost function are trained using a deep Q-network (DQN) algorithm. This approach allows for the optimal distribution of torque between the engine and motor, taking into account physical limits of the battery, engine, and motor, as well as other desired performance metrics. By adjusting the weight of the cost function based on the current state of the system, the P2 HEV achieves optimal energy management and improved performance and efficiency. The use of DQN allows for efficient and effective decision-making, reducing the computational burden and improving the overall performance of the system.
-
ÖgeIntelligent control system design and deployment for fuel cell air supply systems(Graduate School, 2024-06-10)The mobility industry invests in intensive research and development programs to find a sustainable energy source without polluting the environment. A fuel cell system is an electrochemical device that generates electricity and is one of the promising energy sources. However, the commercialization of fuel cell systems is limited due to their lifespan. In this thesis, an intelligent control system has been designed and deployed for fuel cell air supply systems to enhance the net output power and mitigate the degradation due to oxygen starvation, resulting in a longer cell lifespan. To minimize the risk of starvation, more air than needed for chemical reactions is supplied to the fuel cell system. Here, the ratio between supplied and consumed oxygen amounts is defined as the excess ratio. The net output power of fuel cell systems deteriorates when the oxygen excess ratio is high, yet starvation occurs if the oxygen excess ratio is low. Therefore, an accurate control of oxygen excess ratio is crucial to not only maximize the net output power but also reduce the risk of starvation to mitigate cell degradation. To address this challenge, a 2DOF control structure fused with artificial intelligence is proposed in this thesis. The proposed control system involves a data-driven reference generator, a feedforward controller, and a feedback controller. The data-driven reference generator calculates the setpoint value of the oxygen excess ratio. On the other hand, the data-driven feedforward controller calculates the open-loop control signal to anticipate known system dynamics for improving control performance and reducing the control effort by the feedback controller. A PI controller is used as the feedback controller to track the desired setpoint value and calculate the closed-loop control signal. Then, the sum of open-loop and closed-loop control signals is applied to the compressor motor as a voltage input. A fuel cell system was simulated for various current loads and oxygen excess ratio values at the optimal stack temperature to understand the characteristics of fuel cell systems. The results showed that each stack current maximizes the net output power for a specific oxygen excess ratio. The relationship between the stack current and oxygen excess ratio that produces maximum net output power is highly nonlinear, which is challenging to model via traditional lookup-based solutions. Similarly, the compressor voltage needed to reach the optimal oxygen excess ratio, maximizing the net output power and stack current, also has a complex relationship. Therefore, data-driven reference generators and feedforward controllers are considered for learning the complex characteristics of fuel cell systems. The data to learn the data-driven models is acquired through steady-state analysis. Firstly, the data for single input models where the stack current is the only input of data-driven models is acquired by alternating the current and oxygen excess ratio at optimal stack temperature. Moreover, the stack temperature is changed by considering possible temperature fluctuations around the optimal stack temperature, and its effect on net power output is investigated. The results show that the net output power significantly changes with stack temperature. Therefore, it needs to be considered in the design of data-driven models. In this manner, the double-input models are designed with stack current and temperature inputs. The data-driven reference generator and feedforward controller for single and double-input models are learned via fuzzy models and neural networks. Various internal configurations of fuzzy models and neural networks have been studied to investigate their effects on modeling accuracy. The fuzzy models were constructed with various membership functions and learned through various techniques. On the other hand, different activation functions were utilized to build the neural network models. Moreover, the learning data was pre-processed through standardization and normalization to examine their effect on learning performance. Besides, polynomial regression-based reference generators and feedforward controllers were learned for performance comparison. Even though the learning performances of data-driven reference generators and feedforward controllers are pretty satisfactory compared to polynomial regression-based models, their contribution to net output power and degradation must be shown. In this manner, the proposed artificial intelligence fused 2DOF control system was simulated with various fuzzy models and neural networks as the reference generator and feedforward controller. A PI controller was utilized as the closed-loop controller. The PI controller coefficients were tuned through an iterative trial-and-error method in a defined operating point. The same PI controller was employed during the simulations of each design variant to have a fair comparison. In addition, a 1DOF control system was designed to expose the contribution of the 2DOF control structure. To assess the test results, evaluation criteria for the net output power and oxygen excess ratio were defined. In the evaluations, the tracking error and settling time of both targets were considered. In addition, a degradation model that depends on oxygen starvation was created to assess the contribution of data-driven models on cell life. A set of operation points depending on stack current was developed to test the proposed control system. Moreover, stack temperature changes were considered to assess the performance of the proposed control system under disturbances. The results showed that the proposed intelligent control system with fuzzy models and neural networks could efficiently track the desired oxygen excess ratio. Thanks to the data-driven models, the high-performing oxygen excess ratio control structure increases the net output power of fuel cell systems and reduces cell degradation due to oxygen starvation. In brief, the intelligent control system proposed in this thesis is a promising development in fuel cell systems to enable their widespread usage as a clean and sustainable energy source.