Statistics for the provision of reactive power in electrical systems

The forecasting of reactive and active power is an important tool in the monitoring of an Electrical Energy System. The present work has as main purpose the introduction of a new short-term reactive power hourly forecast technique by substations, based on the linearity between reactive and active power through linear regression. In order to improve the forecasting, distributed lags of powers are included in the simple model with a correction for serial autocorrelation (Iterative Method of Cochrane–Orcutt). Moreover as reactive power data have heterocedasticity behavior, the estimation method of the coefficients through least squares is not appropriate. For that reason, it is used a robust solution known as Iteratively Reweighted Least Squares Estimation (IRLS). The short-term reactive power forecast is divided into two periods ‘in sample’ and ‘out of sample’. In order to increase the forecast results, it is necessary to reduce the sample dimension using a methodology to cluster of data. These clusters are classified via an unsupervised learning neural network Kohonen Self-Organized Map.