Short-term load forecasting (STLF) plays an important role in the efficient management of electric systems. Building an optimization model that will enhance forecasting accuracy is not only a challenging task but also a concern for electrical load prediction. Especially due to artificial neural networks (ANNs), the final results are dependent on the initial random weights and thresholds, which influence the forecasting stability. Most analyses are based on accuracy improvements, but the effectiveness of a forecasting model is determined equally by its stability. Considering only one criterion (accuracy or stability) is insufficient. Thus, for the model to achieve these two relatively independent objectives at the same time, high accuracy and strong stability, a modified generalized regression neural network (GRNN) based on a multi-objective firefly algorithm (MOFA), employed to optimize the initial weights and thresholds of the GRNN, is proposed. A new hybrid model composed of multiple seasonal patterns, a data pre-processing technique to reduce interferences from the original data, and MOFA-GRNN for electrical load forecasting is successfully developed in this paper. Case studies utilizing half-hourly electrical load data from three states in Australia are used as illustrative examples to evaluate the effectiveness and efficiency of the developed hybrid model. Experimental results clearly showed that both the accuracy and stability of the developed hybrid model is superior to the models compared. (C) 2016 Elsevier Ltd. All rights reserved.