Please use this identifier to cite or link to this item:
https://hdl.handle.net/11000/29191
Mejoras de eficiencia computacional y precisión para sistemas predictivos de demanda eléctrica
Title: Mejoras de eficiencia computacional y precisión para sistemas predictivos de demanda eléctrica |
Authors: Candela Esclapez, Alfredo ![]() |
Tutor: VALERO, SERGIO ![]() López García, Miguel |
Editor: Universidad Miguel Hernández de Elche |
Department: Departamentos de la UMH::Ingeniería de Comunicaciones |
Issue Date: 2022 |
URI: https://hdl.handle.net/11000/29191 |
Abstract: Debido a la inviabilidad del almacenamiento de energía eléctrica a gran escala, la energía eléctrica se genera y consume simultáneamente. En consecuencia, las entidades eléctricas necesitan sistemas de previsión de demanda para planificar operaciones y gestionar suministros. Predicciones de demanda... Ver más Due to the infeasibility of large-scale electrical energy storage, electrical energy is generated and consumed simultaneously. Therefore, electricity entities need demand forecasting systems to plan operations and to manage supplies. Improving the forecasts accuracy allows economic savings of energy generation supplies, as well as reinforcing the reliability of energy supply to electricity consumers. In addition, demand forecasts allow renewable energies to be managed in electricity networks, indirectly reducing greenhouse gas emissions. This thesis focuses on improving, at peninsular scale, the forecasting system of Red Eléctrica de España (REE) developed by the Miguel Hernández University (UMH). An independent approach of mathematical models is presented, offering methodologies applicable to other forecasting systems from different electrical grids. Two improvements are tackled: a deterministic and automatic schedule obtention and a preprocessing of temperature data, which can be used as a tool for demographic studies. Both enhancements also increase the forecasting accuracy. In Europe, due to directives and new technologies, forecasting systems are transitioning from hourly intervals to quarter-hourly intervals, which reduces the calculation time and increases the computational burden. Therefore, a predictive system may not have enough time to compute all future forecasts. Forecasting systems perform calculations throughout the day, repeating the same forecasts while the forecast time approaches. However, there are predictions that are not more accurate than others already calculated, which leads to not executing them and using previous predictions to save computational effort and maintain accuracy. With the intention of avoiding counterproductive calculations, an algorithm is developed, that estimates which forecasts provide better accuracy than previous ones, then it builds a computing schedule. The algorithm adapts to the computational needs and the predictive system. It has been applied to the REE prediction system, obtaining a computing schedule that achieves greater precision and adapts to the computational load. Temperature affects electricity consumption through air conditioning and heating equipment. This thesis proposes an automatic method of processing and selecting variables with a double objective: to improve both the accuracy and the interpretability of the global forecasting system. The experimental methodology has been carried out with the REE predictive system. The new way of working with temperatures is interpretable as it separates the effect of temperature based on location and time, using variables with a specific meaning. Both studies experimentally demonstrate that the proposed techniques fulfill their purpose, improving the accuracy and computational cost of the predictive system. It is also observed that in Spain heat has a greater influence on demand than cold. On hot days, the temperature of the second previous day has a greater influence than that of the previous one, while on cold days the opposite occurs. Based on the construction of the execution schedule, it has been concluded that temperatures have reduced effect on demand during the early morning hours; temperature forecasts for less than four days ahead provide more accuracy than those more than four; and according as the time difference between the moment of prediction and the moment of execution decreases, the accuracy increases. |
Notes: Programa de Doctorado en Tecnologías Industriales y de Telecomunicación |
Keywords/Subjects: energía eléctrica almacenamiento de energía demanda eléctrica |
Knowledge area: CDU: Ciencias aplicadas: Ingeniería. Tecnología |
Type of document: info:eu-repo/semantics/doctoralThesis |
Access rights: info:eu-repo/semantics/openAccess |
Appears in Collections: Tesis doctorales - Ciencias e Ingenierías |