In this study, a novel electric load forecaster based on adaptive Fuzzy Neural Networks (FNN) and using Genetic Algorithm (GA) mixed with Gradient Descent (GD) is proposed to make it to posses the human learning ability. The proposed SDSA-FNN is firstly compared with various methods applied on function approximations. Moreover, it is applied on electric load forecasting application and verified on electric load data recorded on Macao power system. The simulation results reveal that the proposed methodology not only keeps the traditional objective function.