Abstract: With the continuous development of the power industry, the consumption of energy resources such as coal, oil, electricity, and water is also increasing. Only by properly allocating these important strategic resources can we promote the sustainable long-term development of the national economy. In order to achieve energy conservation and emission reduction, this study explores the 660 MW coal-fired power generation units in a coal-fired power plant through different data-driven model methods, analyzes the characteristics and adaptation conditions of each data-driven model, and provides technical means for energy consumption modeling. This study has reference significance for energy conservation and emission reduction of similar coal-fired power generation units. Among the four data-driven models, the LSTM model demonstrated excellent fitting and generalization abilities in modeling the energy consumption of coal-fired power generation units. The predicted RMSE of unit coal consumption on the training and testing sets were 0.05 g/(kWh) and 0.09 g/(kWh), respectively, the MAPE values were 0.02% and 0.03%, respectively, with a correlation of 1; however, SVR has the worst modeling performance, with predicted RMSE of 1.25 g/(kWh) and 1.58 g/(kWh) for unit coal consumption on the training and testing sets, respectively, the MAPE values were 0.96% and 1.28%, respectively, with correlations of 0.93 and 0.89.