Accurate short-term load prediction for metro stations can help the metro operations department make decisions about purchasing power, further ensuring the stability and cost-effectiveness of the metro station's power supply. Considering the non-linearity and non-stationarity characteristics of metro station load series, this paper proposes a short-term metro station power lighting load prediction method based on the TimesNet network. Firstly, data are collected on factors that may affect power lighting load in metro stations. Secondly, the Pearson correlation coefficient method is employed, and the main correlated characteristics of the power lighting load are selected from the multi-dimensional features. Thirdly, TimesNet is trained to predict the short-term power lighting load. Finally, a simulation experiment is conducted to compare the proposed method with six other typical load prediction methods. The experimental results demonstrate that TimesNet can reduce the MAPE to 3.8% and is generally superior to several typical prediction methods.