The rapid growth of wind generation in northeast Brazil has led to multiple benefits to many different stakeholders of energy industry, especially because the wind is a renewable resource – an abundant and ubiquitous power source present in almost every state in the northeast region of Brazil. Despite the several benefits of wind power, forecasting the wind speed becomes a challenging task in practice, as it is highly volatile over time, especially when one has to deal with long-term predictions. Therefore, this paper focuses on applying different Machine Learning strategies such as Random Forest, Neural Networks and Gradient Boosting to perform regression on wind data for long periods of time. Three wind farms in the northeast Brazil have been investigated, whose data sets were constructed from the wind farms data collections and the National Institute of Meteorology (INMET). Statistical analyses of the wind data and the optimization of the trained predictors were conducted, as well as several quantitative assessments of the obtained forecast results.

The total losses volume represents a substantial amount of energy and, consequently, a large cost that is often included in the tariffs structure. Uneven connection of single-phase loads is a major cause for three-phase unbalance and a fundamental cause for active power losses, particularly in Low Voltage (LV) networks. This paper analyzes the impact of load unbalance on LV network losses. In the first phase, several load scenarios per phase are considered to characterize how losses depend on load unbalance. The second phase examines the data collected per phase on a set of real networks, aiming at illustrating real-world cases. The third phase analyzes the effect that public lighting and microgeneration may have in the load unbalance and on the subsequent energy losses. The results of this work clearly demonstrate that it is possible to reduce three-phase unbalance (and losses) through a judicious distribution of loads and microgeneration.