To address the issues of high energy consumption and low energy management efficiency in the steel industry, an onsite energy medium optimization system was designed and developed. The system integrates multiple functions, including energy data acquisition, real-time monitoring, intelligent analysis, and optimized control, covering key processes such as converters and LF refinement, with the aim of improving energy utilization efficiency, reducing energy consumption, and supporting green and low-carbon transformation in enterprises. A random forest regression model optimized by a genetic algorithm was constructed, which enhanced the prediction accuracy of oxygen consumption in converters by analysing the nonlinear relationships among 12 process parameters, such as scrap steel quantity and furnace age, resulting in a 1.3% reduction in oxygen consumption. For LF refining, an LSTM-based temperature prediction model was introduced, achieving a prediction error of ±5 ℃ for the molten steel temperature, which enabled the optimization of power supply strategies and reduced electricity consumption by 1.5%.