In view of the nonlinear and non-stationary characteristics of electricity price time series, the novel model for electricity price forecasting in this paper is proposed by the combination of Singular Spectrum Analysis (SSA) and Extreme Learning Machine (ELM) optimized by the Improved Cuckoo Search algorithm (ICS). SSA is used to extract the trend components and oscillation components of the original data and reconstruct all the components, and the ELM model is used to predict the reconstruction sequences. Against the parameters that in the ELM prediction model tend to fall into the local optimum, the ICS algorithm is introduced to optimize the model parameters to further improve the predictive value. Finally, all the predicted sequences are summed up to get the final forecast of electricity price. The empirical results demonstrate that the SSA-ICS-ELM model can improve the prediction accuracy and stability of electricity price forecasting considerably in comparison with other methods by analyzing the price data in Australia electricity market. This work is supported by Science and Technology Project of Guangdong Province (No. 2016A010104016) and Science and Technology Project of Guangdong Power Grid Company (No. GDKJQQ20152066).