Interactive evolutionary multiple criteria decision making methods for power plants auxiliary services design
- Autores: Ana Belén Ruiz Mora
- Directores de la Tesis: José Manuel Cabello González (dir. tes.), Mariano Luque Gallego (dir. tes.)
- Lectura: En la Universidad de Málaga ( España ) en 2012
- Idioma: español
- Tribunal Calificador de la Tesis: Francisco Ruiz de la Rúa (presid.), Mónica Hernández Huelin (secret.), Carlos Antonio Platero Gaona (voc.), Carlos Romero López (voc.), Xavier Gandibleux (voc.)
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
The auxiliary systems of a thermal power plant are usually formed by a series of engines working in medium and low voltage. Commonly, they are needed for the operation of the plant, mainly for fuel, water or air supply, or for waste disposal. While they are not the main components of the plant, the energy consumption in the auxiliary systems is often significant, and it can be reduced by implementing a series of improvement policies, like changing the wires by copper and wider ones, installing capacitors in different parts of the network to compensate for some reactive power, changing current motors by high efficiency ones and/or installing variable speed drives. All these changes have a direct impact on the electricity consumption and losses at the auxiliary systems, and the benefit is twofold. Given that the coal consumption is reduced, on the one hand, the operation economic cost of the auxiliary systems is reduced, and on the other hand, the CO2 emissions decrease for the same final electricity production.
This thesis is concerned with the optimization of the energy consumption of the auxiliary systems of power plants which can be achieved by means of some efficiency improvement policies or strategies. However, the implementation of such policies implies an initial investment which may be very high and cannot be overlooked. Then, the cost of implementing such strategies will be also optimized simultaneously with the optimization of the energy saving. But the investment required should be economically analyzed so as to find the most profitable solution among the wide range of improvement options. Then, the Internal Rate of Return (IRR), which studies at which rate the initial investment will be recovered by the benefits in a fix period of time, will be taken into account and maximized. This multiobjective optimization problem grew out of a R&D contract with the Endesa Generation S. A. company, one of the largest electrical companies in the world.
The simultaneous optimization of these three criteria will not be an easy task. Given that the whole auxiliary system is interconnected, each particular improvement decision on any element will influence the energy consumption of the rest of the elements in the network. This fact implies that some of our objectives are black-box functions. Besides, the improvement decisions are represented by both binary and continues variables, what also complicates the problem resolution. Moreover, as the proposed mathematical model has been designed to be applied to the auxiliaries of any power plant, the number of decision variables depends on the number of elements and on the configuration of the auxiliary services. As a result, if the auxiliaries are very large, we will have to manage a large number of decision variables.
Taking into account the previous features, traditional multiobjective optimization techniques are not suitable for the resolution of our problem. In recent years, evolutionary algorithms have shown to be able to handle different types of real problems. In this thesis, they are used to generate an approximation of the whole set of efficient solutions of our problem, which enables us to have a better insight of the conflict degree among the objective functions.
The final aim of this thesis is to solve the multiobjective problem associated to a case study together with a real decision maker (DM) from the company Endesa Generation. Given that evolutionary algorithms do not consider the preferences of the DM in the resolution process and the set of solutions generated may be too large, a new preference-based evolutionary algorithm is presented in this thesis and, subsequently, applied to our multiobjective problem in order to find the solution that best satisfies the preferences of the DM.
To do so, the DM will have to indicate desirable objective function values which will constitute the so-called reference point. Given that the DM may not be initially sure about which reference point (s)he wants to use, we will develop a selection procedure in order to help him/her find a reference point which properly expresses his/her preferences, and where (s)he will be able to analyze which solutions could be achieved for several reference points. For this purpose, we will use a new achievement scalarizing function which enables us to find the most suitable solution for a reference point taking into account its achievability by using two different vectors of weights.
Being aware of the importance of an easy and understandable interaction with the DM, we present an interactive procedure and a software application for the resolution of the problem, where the DM can both describe the auxiliary services (s)he wants to analyze and solve the multiobjective problem associated to the given auxiliaries, finding his/her most preferred solution.
Finally, the multiobjective problem associated to a case study is solved using the previous interactive procedure. Here, an engineer of Endesa Generation played the role of the DM, who was able to find an interesting enough solution after a few interactions. The resolution of the case study will show the performance and the usefulness of the process, which will enable us to have a better insight of our multiobjective problem, highlighting the real conflict degree among the objectives. The real interaction with the DM will highlight the importance of considering preferences into an evolutionary algorithm when solving real multiobjective problems. Besides, the fact of showing a not too large set of solutions representing the region of interest of the Pareto optimal set will enable the DM to be more conscious about the different trade-offs in that region, being in a better position to select his/her most preferred solution.
