Effective neuro-evolutionary schemes for solar radiation estimation problems

Resumen

This Ph.D. thesis’ goal is focused on the optimization of renewable energy resources development, specifically solar PV energy, using different hybrid computational Machine Learning techniques. Energy is the engine of our society, allowing us performing almost every action taken by human beings in our daily routine, and providing a constant evolution and development in all our fields. Currently, fossil fuels entail the higher percentage of energy sources in our planet. They have several advantages, such as easy and constant production, but, at the same time, they present substantial disadvantages, like the extreme pollution associated with these resources, and their contribution to global warming and climate change. This is the reason why the largest and most powerful economies are working for a energy change towards renewable sources for a sustainable development. In the introduction of this thesis, a large number of studies are presented, which foresee a penetration by over a 50% of this kind of energies in the next decades. Strong investments are being made in this field, looking for technology development and, besides, introducing these energies into society as a matter to be taken into account, for it is related to economic and social status. However, the development of energy systems mainly based on renewable energy will surely be slow, since these energies depend on variables which are out of our control, mainly atmospheric and climatic variables, which are intrinsically intermittent. This matter must be taken into account, due to the amount of energy demanded by society at the present, as well as the tremendous increase that is predicted for this demand in the future, due to new technologies and new ways of daily routines, like electric vehicles or IoT, etc. To obtain a solution for this problem, on one hand, it is fundamental to achieve the capability of predicting the quantity of energy obtained in each moment, avoiding increases or decreases of energy, being this matter the core of this Ph.D. Thesis, where the optimal feature selection for predicting the quantity of global solar radiation in a given point is studied. On the other hand, all the information to do the prediction process will be obtained from a numerical weather mesoscale model called WRF (Weather Research and Forecasting), a static model based on different physic equations which involve different variables like humidity, pressure or percentage of cloud fraction in any point and different heights in the planet. Additionally, dynamic information, like global solar radiation can be obtained from a radiometric measuring point in Toledo, Spain, allowing us to get a database of the solar global radiation in the past few years. The result of mixing both of these data will be added as inputs in our hybrid systems. In this work, a deep analysis in the state of art for machine learning models is performed, so as to solve the problems previously considered. Different contributions have been proposed: 1. One of the pillars of this work is focused on the optimal feature selection in the exploitation of solar PV radiation in a given point. For this purpose, Extreme-Learning Machine (ELM) will be used as regressor element in the system, where the output of the ELM will be calculated from the WRF outcome features added as inputs in the system. 2. The second contribution of this thesis is related to parameters selection problems. More specifically, the use of EAs such as Grouping Genetic Algorithm (GGA) or Coral Reef Optimization (CRO) hybridized with others ML are used as classifiers and regressors. Regarding this, the GGA or CRO look for several subsets of basic parameters to solve the problem, and the regressor employed provides the prediction in terms of the selected by the Genetic Algorithm (GA), reducing the computational cost maintaining a good accuracy. Finally, the several of the mentioned algorithms are applied in the same problem already defined, in order to get the global solar radiation prediction in different points, dealing to improve previous results in other works and obtaining new applications and techniques, as new paths of research in the future.