GENIE INFORMATIQUE

In general the undergraduate curriculum in statistics offers one semester basic course on linear programming (LP). The conventional course on LP includes simplex algorithm, duality theory, sensitivity analysis and some applications like the feedmix and transport problems. This outline seems to be distant from the mainstream curriculum in statistics. In order to partially overcome this shortcoming, we propose to introduce a technique based on linear programming called Data Envelopment Analysis (DEA) on the LP courses for students of statistics. DEA has gained momentum as a powerful complementary method for statistical tools in the research agenda as well as in practical activities of economic efficiency evaluation.

The electric load data quality is fundamental in the short-term load forecasting. In general, data quality is poor, since the load records often have aberrant data, discontinuities and gaps resulting from failures and problems in the measuring system. The presence of bad data in historical load data contributes to poor specification of a forecasting model and therefore affects the accuracy of load forecasts. Thus, the construction of a forecasting model should be preceded by the treatment of the load records. This paper presents a methodology based in statistical and artificial intelligence techniques to automatically cleanse historical load records.
Resumo. A qualidade dos dados de carga é fundamental na previsão de curto prazo. Em geral, esta qualidade é ruim, uma vez que os registros de carga frequentemente apresentam dados aberrantes, descontinuidades e lacunas de dados resultantes de falhas e problemas no sistema de medição. A presença de erros no histórico de dados contribui para uma especificação errônea do modelo de previsão e consequentemente para sua perda de precisão. Portanto, a construção de um modelo de previsão deve ser precedida pelo tratamento dos dados históricos da carga. Este trabalho apresenta uma metodologia baseada em métodos estatísticos e de inteligência artificial para tratamento de dados de carga.