• photovoltaic energy
• pattern classification
• genetic algorithms
• fuzzy rule-based classifiers
• forecasting
• building’energy consumption
• artificial neural networks
• time series

This thesis presents a few novel applications of Computational Intelligence techniques in the field of energy-related problems. More in detail, we refer to the assessment of the energy produced by a solar photovoltaic installation and to the evaluation of building’s energy consumptions. In fact, recently, thanks also to the growing evolution of technologies, the energy sector has drawn the attention of the research community in proposing useful tools to deal with issues of energy efficiency in buildings and with solar energy production management. Thus, we will address two kinds of problem.
The first problem is related to the efficient management of solar photovoltaic energy installations, e.g., for efficiently monitoring the performance as well as for finding faults, or for planning the energy distribution in the electrical grid. This problem was faced with two different approaches: a forecasting approach and a fuzzy classification approach for energy production estimation, starting from some knowledge about environmental variables. The forecasting system developed is able to reproduce the instantaneous curve of daily energy produced by the solar panels of the installation, with a forecasting horizon of one day. It combines neural networks and time series analysis models. The fuzzy classification system, rather, extracts some linguistic knowledge about the amount of energy produced by the installation, exploiting an optimal fuzzy rule base and genetic algorithms. The developed model is the result of a novel hierarchical methodology for building fuzzy systems, which may be applied in several areas.
The second problem is related to energy efficiency in buildings, for cost reduction and load scheduling purposes, and was tackled by proposing a forecasting system of energy consumption in office buildings. The proposed system exploits a neural network to estimate the energy consumption due to lighting on a time interval of a few hours, starting from considerations on available natural daylight.