Considering the complexity and uncertainty of energy consumption in a discrete manufacturing system, a general data-driven energy prediction method is urgently required. Therefore, a knowledge model oriented to energy consumption is built using a knowledge modeling method to guide the combination of the energy consumption data, the process plan and the production schedule. Then, character input variables and numeric input variables are both considered in the process of energy consumption prediction. The attribute importance is calculated by the degree of power fluctuation, and the consistency or correlation of character input variables are guaranteed by using the hierarchical case retrieval method. Finally, energy consumption is predicted by retrieving similar working-steps. The experimental simulation shows the feasibility and effectiveness of the proposed method.