Data mining models for short-term solar radiation prediction and forecast-based assessment of photovoltaic facilities
- Autores: Pedro Francisco Jiménez Pérez
- Directores de la Tesis: Llanos Mora López (dir. tes.)
- Lectura: En la Universidad de Málaga ( España ) en 2016
- Idioma: español
- Tribunal Calificador de la Tesis: Rafael Morales Bueno (presid.), Gustavo Eduardo Nofuentes Garrido (secret.), Santiago Silvestre Berges (voc.)
- Materias:
- Enlaces
- Tesis en acceso abierto en: RIUMA
- Resumen
Solar radiation prediction is useful to integrate photovoltaic power plants into the electrical system. Integrating energy generation in urban environments is interesting because that is where the most energy is consumed and avoids wasting energy in transport infrastructure. Renewable energies are often the easiest to integrate into these environments because they require less infrastructure and cause fewer problems related to noise, dirt, pollution, etc.
The overall objective of this thesis is to develop data mining models to forecast solar global radiation 24 hours ahead and to use these predictions to evaluate the performance of photovoltaic systems. The specific objectives are: 1. Propose an index that allows us to remove the seasonal and daily trends observed in global hourly radiation data.
2. Analyze the different sources of meteorological variables that can be used to predict solar radiation and use API's to access external sources of meteorological data.
3. Develop data mining models that allow including the different relationships observed between the radiation values of the next day depending on the values of the current day radiation and other meteorological parameters.
4. Development of a web system that include the proposed models for short-term radiation forescasting and integrate the developed models in the evaluation models of photovoltaic systems.
Chapter 3 introduces the methods and models used in this work (Cumulative Probability Distribution Function, Artificial Neural Networks and Support Vector Machines). Also classification methods are presented (Decision Trees and Support Vector Machines for Classification). Performance metrics are presented to measure the accuracy of the proposed models. The data sets and data sources used in this work to test the proposed models are presented, including data from the meteorological station installed at University of Malaga, data from OpenWeatherMap website and data from AEMET (Agencia Estatal de Meteorología).
Chapter 4 is dedicated to the solar radiation fundamentals, including astronomical concepts related to Earth-Sun position, characterization of solar radiation hourly series, clearnes index, used to remove seasonal trends, persistence model, used to compare with proposed models and the forecast skill, based on persistence model and used as reference model as well.
Chapter 5 introduces a model to model and characterize hourly solar global radiation using statistical methods like CPDF, K-means, and also using the clearness index. This models aims to predict the hourly solar radiation using the daily clearness index as input.
Chapter 6 details the proposed model to forecast the hourly global solar radiation using data mining methods and daily profiles of clearness index. K-means is again used to cluster daily solar radiation profiles, then a new variable is defined from the clearness index daily profiles. Support Vector Machines, Decision Trees and Artificial Neural Networks are used to predict the desired hourly solar radiation values.
Chapter 7 presents a methodology to assess solar power plants performance based on forecasted solar radiation. A OPC-based system is presented, which is able to obtain data from a large variety of equipment, then an algorithm to assess the performance of the plants is presented.
