Diagnostic and Input Selection Tool applied on Weather Variables for Studies of Short-Term Load Forecasting

One of the main studies applied on Operation Planning is short-term load forecasting, which results are obtained mainly by hour in few days ahead horizon. The load forecasting model quality is directly related to two factors, the definition of the input variables of the forecasting model and the mathematical modeling of this model. About the first aspect, the most significant exogenous influences on demand in the short-term load forecasting horizon are the weather variables, mainly temperature, humidity and wind speed. Regardless of the weather variable considered, it is necessary to perform a correlation study of the influence of this variable on demand, in each different region and climatic season. Thus, the objective of this work is to present a study of feature selection related to weather variables considering the application of a short-term load forecasting model to four different climate zones. This study, based on a feature selection tool that generates results of relation between the demand and the weather variables, aims to justify the definition of different weather inputs for each region, seeking to increase the quality of the load forecasting results. Also, the feature selection tool is based on different math definitions, that contributes to a better pattern recognition.