Abstract: Driven by growing fossil resource scarcity and increasingly stringent agricultural emission standards, low-carbon electrification of agricultural power machinery has become an inevitable approach to achieving carbon peaking and carbon neutrality targets. Conventional diesel tractors suffer from low energy utilization efficiency and severe pollutant emissions, while pure electric tractors are constrained by limited battery capacity, high procurement costs and inadequate endurance to support long-hour and heavy-duty farm operations. Diesel-electric hybrid tractors integrate the superior cruising capability of internal combustion engines and the high-efficiency, fast-response characteristics of electric motors, emerging as the most feasible transitional power solution for green agricultural mechanization. As the core control module of hybrid tractor powertrains, the energy management strategy (EMS) undertakes real-time power distribution between engines and motors and coordinates the charge and discharge behaviors of energy storage units, which fundamentally determines the overall traction performance, fuel economy, emission level and battery service life of hybrid tractors. Different from hybrid passenger vehicles operating under regular urban cyclic conditions, agricultural tractors work under typical farm loads featured by drastic load fluctuations, random tillage resistance, continuous low-speed and high-torque traction, and frequent alternation of ploughing, rotary tillage, sowing and transportation tasks. Direct transplantation of vehicle-oriented EMS algorithms will degrade control performance and waste energy-saving potential, making systematic sorting and in-depth evaluation of hybrid tractor EMS research crucial for promoting high-efficiency and industrialized hybrid agricultural machinery. This review systematically classifies hybrid tractor powertrain topologies into series, parallel and series-parallel power-split configurations and compares their power coupling logics, energy transfer paths and practical operation constraints. Series structures realize complete decoupling between engines and traction loads to maintain stable and efficient power generation but induce extra energy losses from dual mechanical-electrical conversion. Parallel structures deliver composite mechanical power with fewer conversion links yet lack flexible load decoupling capacity under extreme variable working conditions. Series-parallel architectures integrate dual operation modes and multi-channel power flow to possess the highest control freedom, while confronting prominent difficulties in mode switching coordination. The powertrain topology constitutes the fundamental design constraint of tractor EMS, which defines optimization boundaries, control objectives and optional algorithm frameworks, indicating that customized EMS design is essential rather than blindly adopting universal vehicle control architectures. Existing EMS methods are categorized into rule-based, optimization-based and data-driven intelligent systems according to control principles and solution mechanisms, with comprehensive comparison of their theoretical frameworks, implementation approaches, core strengths and field application limitations. Rule-based EMS includes deterministic and fuzzy logic branches. Deterministic strategies such as thermostat control, power-following control and multi-point control adopt fixed threshold judgment, presenting low computation burden, outstanding real-time performance and high embedding adaptability for stable cyclic farm tasks. Fuzzy control quantifies uncertain parameters to enhance system robustness against random soil resistance, yet its performance fully relies on manually established rule bases without online adaptive correction, restricting energy optimization effects under dynamically varying field loads. Optimization-based EMS transforms power allocation into equivalent fuel consumption minimization problems and covers instantaneous and global optimization types. Typical instantaneous strategies including ECMS and MPC implement rolling real-time optimization to balance transient fuel consumption and battery SOC stability, whereas ECMS is sensitive to time-varying equivalent factors and MPC requires high-precision dynamic models and powerful onboard computing hardware. Global optimization algorithms such as DP, PMP and intelligent optimization methods can acquire theoretically optimal power allocation results under known full-cycle conditions, which are widely applied as offline calibration benchmarks but fail in real-time deployment and unexpected farm disturbance suppression due to massive iterative computation and full load sequence dependence. Data-driven intelligent EMS based on machine learning and deep reinforcement learning represents the cutting-edge research direction, which eliminates dependence on accurate physical models through data self-learning. Various neural networks and advanced reinforcement learning algorithms including DQN, DDPG, TD3 and SAC construct adaptive power allocation strategies by interacting with complex farm environments, achieving coordinated optimization of fuel saving, battery protection and traction stability under highly nonlinear and time-variant tillage conditions. Nevertheless, insufficient high-quality field datasets, huge training computation costs and poor model interpretability severely restrict its industrial promotion and application reliability. Comparative analysis reveals differentiated application scenarios for the three EMS types. Rule-based strategies serve as low-cost baseline schemes for low-end tractors with fixed working conditions; optimization-based methods balance energy-saving performance and real-time capability for medium-power hybrid equipment; data-driven intelligent strategies are the preferred technical route for high-end unmanned hybrid tractors equipped with edge computing devices. Current EMS research encounters four core bottlenecks: inadequate online adaptive calibration for time-varying ECMS factors under battery aging and sensor noise, insufficient rolling optimization efficiency for frequent load mutations, imperfect fusion frameworks of physical models and intelligent algorithms, and absent global collaborative energy scheduling across multiple subsystems. Facing the integrated development trend of agricultural intelligence, digital twin and edge computing, future EMS research will focus on multiple key synergistic development directions. This review summarizes the technical evolution, topology-strategy matching rules, algorithm performance trade-offs and unsolved bottlenecks of hybrid tractor EMS, providing systematic theoretical references and clear research roadmaps for developing high-performance hybrid tractor energy control systems and facilitating the advancement of low-carbon sustainable green agriculture and dual-carbon strategic implementation.