Abstract

The performance of industrial processes can be greatly improved using control strategies that optimize operational, economical and environmental criteria, instead of only taking into account stability and reference tracking aspects. Thus, a relevant issue to be analyzed and solved in this project is the development of control strategies to optimize the aforementioned aspects for these processes. Model Predictive Control is a mature family of control strategies that consist on using a model of the process to predict its future behavior and find a set of control actions that optimize a performance criterion over a prediction horizon, which may include reference tracking, control effort, economical and environmental terms, among others. The optimization problem is solved in every time instant, and only the first control action is applied to the system, according to a receding horizon strategy. The main aspects for the success of the strategy are the quality of the predictive model of the process, the objective function design, the inclusion of constraints and the optimization method that guarantees that solutions can be found fast enough so that applying the strategy in real time is possible. Based on this, this proposal considers the design of methods based on non-linear modeling and non-linear predictive control strategies using computational intelligence techniques for modeling and optimization of dynamic non-linear hybrid processes. These methodologies will be formulated using the hybrid predictive control theory that allows optimizing different objectives in order to obtain the optimal control actions for dynamic non-linear processes with continuous and quantized/discrete variables (hybrid behavior). First, we will consider the use of hybrid fuzzy modeling to represent the discrete or quantized behavior of the process and all other kinds of non-linearities. Therefore, identification methods for these proposed hybrid fuzzy models will be developed Then, predictive control strategies based on hybrid fuzzy models will be formulated, taking account special emphasis in the development of adequate objective functions, models and constraints. The resulting non-linear dynamic optimization problems are usually NP-hard, and will be faced with the challenge of design efficient solution algorithms. Several optimization algorithms (such as classical and/or evolutionary optimization algorithms) will be analyzed and tested in order to generate the best solutions in terms of accuracy and computational time and then, the best algorithms will be properly implemented for generating efficient dynamic solutions of the proposed non-linear predictive control strategies. In summary, the analysis and development of predictive control strategies based on hybrid fuzzy models under an efficient optimization approach will be studied and developed in this project. Due to the quickly increasing energy demand and depletion of conventional resources, the optimization of the operation of renewable energy plants is of special importance. Based on this, the developed strategies in this project will be applied and tested in renewable energy generation systems such as wind turbines and new combined cycle power plant with solar units. Also, for evaluation purposes of the proposed modeling and control methods, the phenomenological simulators of combined cycle power plants with solar units and wind turbines will be used. The dynamics of combined cycle power plants with solar units require hybrid fuzzy models due to their non-linear continuous behavior and discrete states and inputs involved (on-off valves, start up/shut down modes, etc.). Then, hybrid fuzzy models will be used for a predictive control design in order to optimize operational, economical and environmental criteria. For wind turbines, the performance optimization implies a tracking problem under hard disturbances as wind turbulences and also the optimization of operational and economical criteria. Both renewable energy processes require to be modeled as hybrid fuzzy models due to their non-linear behavior and quantized variables involved (position of blades winds, operation modes, etc.).