Abstract: A hybrid power system has been widely used in Hybrid Electric Tractors (HETs) under farmland clustering and energy shortage in China. However, a great design difficulty can be found in the dynamic and variable load conditions of a working tractor on an unstructured road. Particularly, the powertrain flow can be used to realize the coupling and decoupling between the output power of the driven axle and Power Take-Off (PTO). It is a high demand to develop flexible powertrain for tractors, in order to improve the operation performance of agricultural machinery. Thus, the coupled-split powertrain system has been proposed with the principle of graph theory towards the single engine, dual motors, and clutch. All power ranges demand has been satisfied with the variable combination mode between the clutch and power units. In addition, the energy allocation can be optimized between the engine and motors in the background of non-linear loads. In this study, a Markov Decision Process (MDP) based Energy Management Strategy (EMS) was proposed to allocate the power between the engine and motors along with the dynamic and variable load. Firstly, the spectrum of the working load was collected in the period to distinguish the working scenarios. Specifically, the sample of the working environment included plains, hills, and basins. Secondly, the demand power in plowing was abstracted as the state transition of the MDP in the premise of tractor parameters collected under the plow condition, with which the comprehensive dynamics of tractor loads were mathematically formulated. Thirdly, energy consumption was defined as the cost function in the optimal control process, which was solved by the value iteration function. The working range of motor-2 was determined under the guidance of optimal control. In the actual plowing condition, the torque of motor-2 was optimized and determined along with the demand power and state of charge (SOC), which was converted to a look-up table and download in the Vehicle Control Unit (VCU). Finally, the effectiveness and feasibility of the system were validated with the hardware-in-loop test. Among them, the program was also conducted in the VCU on the actual test bench. Meanwhile, the model of the tractor was established for the co-simulation. The result indicated that the improved EMS reduced fuel consumption by 7.2%, compared with the traditional. The demand power forecast strategy further improved the energy efficiency characteristics of the drive motor in the plain plowing environment. In addition, the novel powertrain configurations of the tractor contained the direct and indirect coupled-split power system, which expanded the path of power flow between power units and wheels. Besides, the direct coupled-split configuration has the potential application to replace the technical difficulties of traditional power shifts and Continuously Variable Transmission (CVT). The new strategy can be expected to serve as high energy efficiency. The powertrain of coupled-split configuration can provide a strong reference to breaking through the difficult situation of power shift and CVT for high-power tractors. The finding can be expected to make great progress in the hybrid power system of the tractor in agricultural machinery.