Aiming at the lack of load power prediction function in electric drive armored vehicles leading to the lag of control action, a real-time energy management strategy with higher load power prediction accuracy was proposed. Based on analyzing the whole vehicle′s structure, each power source′s mathematical model was established using theoretical analysis and data fitting methods. Combining the two forecasting methods of auto regressive integrated moving average model and adaptive Markov chain, a combination forecasting method of non-stationary trend load power was designed. Under the framework of nonlinear model predictive control, a multi-objective optimization function was constructed, and the sequential quadratic programming method was utilized to solve the optimal control command in real-time in the finite time domain. The multi-power source was optimized and coordinated. Relying on the hardware-in-the-loop simulation platform, multi-road driving experiments were carried out, and energy management control effects with or without power prediction method were compared. The results show that the improved real-time energy management strategy has good predictability for future load power. It can significantly optimize the coordinated control process of multiple power sources, improve vehicle fuel economy, stabilize bus voltage and battery state of charge. Moreover, it has specific reference significance in engineering application scenarios under traditional model predictive control.