Abstract: Accurate and computationally efficient forecasting is essential for the undisturbed operation of power systems. Traditional methods cannot detect the nonlinear behaviour of modern electrical grids, affected by various factors such as seasonal variations, peak-hour demand fluctuations the increasing penetration of renewable energy sources, etc. To overcome these limitations, an adaptive hybrid method is proposed, combining the Multi Model Partitioning Filter (MMPF) with Nonlinear Autoregressive Exogenous (NARX) models and a Genetic Algorithm for Resource Allocation (GARA). The MMPF structure remains unchanged, implementing NARX submodels to detect the non-linear data characteristics. Additionally, the GARA optimizes the MMPF overall weights through an evolutionary, procedure. The new method is evaluated using real data from the Hellenic grid and compared against two established methods, MMPF–SVM and MMPF–GA. The results indicate that the proposed method achieves better forecasting accuracy and reduced computational burden, making it suitable for applications including autonomous or mission-critical energy networks.

Keywords: Electric load forecasting, Multi Model Partitioning Filter, Nonlinear Autoregressive Exogenous models, adaptive hybrid modelling, Genetic Algorithm for Resource Allocation, Support Vector Machines

Pages: 511-518