Day Ahead Load Forecasting Model Using Gaussian Naïve Bayes with Ensemble Empirical Mode Decomposition

Abstract

The importance of load forecasting has provided valuable information for power grid analysis since the early 2000's. It has been established that no specific load forecasting model can be generalized for all demand types. This study aims to fill the gaps among the plethora of existing mathematical forecasting methods, specifically using the Naive Bayes Theorem. Naive Bayes, by itself, has an issue when dealing with large amounts of input which is the reason it has not been used in load forecasting. The integration of Naive Bayes along with the Ensemble method and Empirical Mode Decomposition provided our Hybridized Naive Bayes Algorithm with adequate improvement in its accuracy given the large amount of input data. The results were justified using key performance indicators MAE, MAPE and MSE. We obtained an average of 34.35 for MSE, 60.72MW for MAE and 4.41% for its MAPE. Although the hybridized Naive Bayes presented in this study is not ready for industrial use, it is very promising due to its mathematical prediction model and even more improvement is highly feasible.