Abstract: A tractor is one of the most important power equipment in sustainable agriculture. Complex and variable field conditions can often lead to frequent fluctuations in traction resistance during plowing. It is often required to maintain the highly efficient and energy-saving operation. The conventional tractors with the internal combustion engine are restricted to carbon emissions and thermal efficiency. Alternatively, the pure electric solutions are also limited by current battery technology, in terms of the energy density and endurance. An adaptive equivalent consumption minimization strategy (A-ECMS) has been widely used in hybrid power systems due to its excellent fuel economy and optimization potential. The energy management strategy can directly dominate the overall performance of the vehicle. The high efficiency, energy savings, and reliable operation can be expected for the hybrid electric tractors. However, the effectiveness of this algorithm can highly depend on the dynamic calibration accuracy of the equivalent factor, which can suffer from the limited adaptability under varying working conditions. In this study, a distributed hybrid electric tractor (DHET) architecture was proposed with a decoupled power output, in order to fully meet the demands of the high-power continuous operation. The energy efficiency and operational adaptability were significantly improved to decouple the multiple power sources, in order to fully meet the higher requirements on the efficiency of the energy allocation under dynamic working conditions. Therefore, an energy management strategy was proposed to integrate the clustering optimization and A-ECMS. A hierarchical architecture was adopted to enhance the adaptability and fuel economy of the control strategy after the multi-stage optimization. Specifically, a K-means clustering algorithm was introduced to analyze the standard cycles in the conventional working condition division. The silhouette coefficient (SC) was also used as an evaluation metric in order to determine the optimal number of clusters. Furthermore, the dynamic programming (DP) was then employed to solve for the optimal equivalent factor corresponding to each cluster. A mapping model of the equivalent factor was constructed to compare the single equivalent factor in online control, according to a continuous weight re-kernel function. The equivalent factor was matched dynamically, according to the real-time features of the working condition and offline cluster centers, thereby enhancing the adaptive adjustment of the A-ECMS algorithm. Hardware-in-the-loop (HIL) experiments demonstrated that the engine and motor operation were maintained within the high-efficiency ranges under most working conditions. Under the tested conditions,the proportion of operating points within the high-efficiency range for the engineincreased by 4.40%, while the proportion in the low-efficiency range decreased by 10.21%. For the motor, the proportion ofoperating points within the high-efficiency range increased by 4.28%, and the proportion in the low-efficiency range decreasedby 1.20%. The equivalent fuel consumption of this strategy was reduced by 6.54% under the tested conditions, compared with the adaptive equivalent consumption minimization using a PI controller. Therefore, the fuel economy of the hybrid electric tractor was significantly improved for the promising prospects in the engineering applications.