Load data cleaning with data mining techniques

摘要

Data quality is critical in the short-term load forecasting. Often, load data show outliers, discontinuities, and gaps resulting from abnormal operation of the electrical power system or failures and problems in the measurement system. The presence of corrupted data impairs the specification of load prediction models and consequently affects the quality of the predictions obtained. Therefore, the construction of a load prediction model must be preceded by data cleaning. This work presents a methodology based on statistical methods and data mining techniques for load data cleaning. The application of the proposed methodology is illustrated by results from computational experiments conducted with load data from the Brazilian Interconnected Power System (BIPS).