As the concept of green manufacturing continues to deepen in the motor industry, accurately grasping energy consumption characteristics and achieving efficient prediction has become an important issue for reducing carbon emissions and increasing efficiency. Aiming at the multivariate coupling and strong time series fluctuations in the motor manufacturing process, this study constructs a multi-algorithm fusion energy consumption modeling framework, combines random forest regression (RF) for key feature screening, and long short-term memory network (LSTM) for time series prediction, so as to improve the model’s adaptability to complex processes and environmental disturbances. Experiments show that the RF-LSTM model is superior to the comparison algorithm in terms of mean square error, mean absolute percentage error, determination coefficient, and fluctuation-sensitive absolute error in high-volatility scenarios. The mean square error is only 0.184 kWh2, and the prediction offset is as low as 0.031 kWh. Different industry groups gave the system ease of use and prediction credibility more than 4 points. The method proposed by the institute can provide an effective technical path for energy consumption optimization and intelligent decision-making in the motor manufacturing industry, and has good industrial application potential.