Abstract: In order to accurately predict short-term electricity price in electricity market, the maximum information coefficient (MIC) correlation analysis and the improved multi-hierarchy gated long short-term memory (MHG-LSTM) was combined to propose a new short-term electricity price forecasting method. The method firstly analyzed the maximum information coefficient correlation between the historical sequences and the predicting electricity price sequence, and then selected candidate sequences and wavelet transform price and load sequences to synthesize input sequences, which effectively increased the information density related to the predicted electricity price in the input; secondly, the method improved the traditional LSTM and built a multi-hierachy gated LSTM model that had two levels of forget gates and input gates instead of the traditional first-level gating mechanism, which improved the ability of neural network to select and extract the characteristics of the electricity price sequence. In this paper, the proposed method was simulated on the PJM market price dataset. The prediction error of this method on experimental dataset was 4.506%. Compared with existing prediction methods, the proposed method greatly improves the prediction accuracy and has great adaptability, which can be applied to the short-term electricity price forecast in electricity market, providing a strong decision-making basis for market participants and regulators.