Extreme learning machine (ELM) with random input weights and hidden biases may lead to unstable performance and low prediction accuracy. This paper proposes a new short-term load forecasting method based on artificial bee colony (ABC) algorithm and ELM (ABC-ELM). Firstly, historical load, meteorological factor and day of week are selected as input variables to build the ELM model. Secondly, optimal input weights and hidden biases of ELM are selected by ABC algorithm. Finally, the new model with optimized parameters is constructed. The real load date from a large city in China is applied to estimate the performance of proposed method. Experiment results show that the new method has higher stability and accuracy than ELM and BP neural networks.
