A Comparison of Integrated Filtering and Prediction Methods for Smart Grids

The intelligent use of green and renewable energies requires reliable preferably anticipated information regarding their availability the behavior meteorological variables in a scenario natural intermittency. Examples this are smart grids, which can incorporate, among others, charging system for electric vehicles modern predictive management techniques. However, some issues associated with such procedures data captured by sensors transducers noise signals low repeatability under same reading conditions. To tackle problems, numerous filtering fitting techniques various prediction methods have been developed, but an appropriate selection be cumbersome. Also, techniques, as RANdom SAmple Consensus (RANSAC) appear not to used scenarios authors’ knowledge. In regard, paper aims present comparison terms average error, determination coefficient, cross validation that expected schemes Multiple Linear Regression, Vector Support Machines Multilayer Perceptron Regression Neural Network (MLPRNN), filtering/scaling Maximum Minimum, L2 Norm, Standard Scale, RANSAC. Cross allows flag problems like overfitting or bias, is another novelty grid scenarios, Although many combinations were analyzed, RANSAC, Norm MLPRNN prediction, generate best results. RANSAC algorithm predicting through MLPRNN, R2 error reduced 0.8843, MSE 0.8960, accuracy increased 0.44 (±0.2).