A hybrid tractor multi-mode energy management method, device and medium

Abstract

The application discloses a hybrid tractor multi-mode energy management method, device and medium, and relates to the technical field of vehicle control. The method comprises the following steps: acquiring an agricultural task to set an initial operation mode; collecting real-time driving resistance; when the real-time driving resistance deviates from the threshold range of the initial operation mode within a continuous preset period, the initial operation mode is corrected to a corresponding target operation mode; fusing vehicle-mounted sensor data to construct a multi-dimensional observation vector; according to the target operation mode, activating a corresponding target intelligent agent in a pre-trained parallel intelligent agent set; inputting the multi-dimensional observation vector into the target intelligent agent to output an equivalent fuel consumption factor of electric energy; and substituting the equivalent fuel consumption factor of electric energy into a Hamilton function constructed by an equivalent fuel consumption minimum strategy to solve an optimal control instruction that minimizes the Hamilton function. The application improves fuel economy, operation stability and anti-interference ability in a complex farmland environment through the above method.

Description

Technical Field

[0001] This application relates to the field of vehicle control technology, and in particular to a multi-mode energy management method, device and medium for a hybrid tractor. Background Technology

[0002] Currently, energy management strategies for hybrid tractors mainly employ logic thresholding, instantaneous equivalent fuel consumption minimization, and artificial intelligence algorithms based on a single architecture. These methods typically rely on onboard sensors to monitor vehicle status in real time for passive response control, enabling torque distribution between the engine and motor. Some improvements introduce operating condition identification or short-term load prediction to enhance adaptability to changes in workload.

[0003] However, the existing methods still have the following significant limitations in actual farmland operation environments: First, they lack the ability to perceive the global task throughout the entire operation cycle and cannot utilize the prior information of smart agriculture platforms. This makes it difficult to achieve optimal energy allocation strategies throughout the entire timeframe, and during variable fertilization and sowing, power fluctuations caused by sudden power changes can easily lead to power interruptions or even engine shutdowns when turning into the soil at the edge of the field. Second, the severe vibrations and load noise caused by the unstructured environment of farmland seriously interfere with control signals. Existing methods directly use instantaneous load as the basis for control, which is easily misled by instantaneous spikes, resulting in frequent oscillations in torque commands, and cannot extract the true soil resistance characteristics from the noise. Third, for multi-mode operation conditions such as deep plowing, rotary tillage, and spraying, traditional single-model energy management strategies cannot take into account the energy consumption preferences and power protection needs of different modes, and lack a hierarchical management system that can dynamically switch and customize incentive mechanisms according to the operation type.

[0004] Based on the above analysis, the problems and shortcomings of the existing technology are as follows: Existing energy management methods for hybrid tractors lack global task awareness, have poor anti-interference capabilities, and are insufficient in multi-mode adaptability. Summary of the Invention

[0005] This application provides a multi-mode energy management method, device, and medium for hybrid tractors, which can solve the problems of lack of global task awareness, poor anti-interference ability, and insufficient multi-mode adaptability in existing hybrid tractor energy management methods.

[0006] In a first aspect, embodiments of this application provide a multi-mode energy management method for a hybrid tractor. The method includes: acquiring agricultural tasks to set an initial operating mode; collecting real-time driving resistance, and correcting to the corresponding target operating mode when the real-time driving resistance deviates from the threshold range of the initial operating mode within a continuous preset period; fusing on-board sensor data to construct a multi-dimensional observation vector, the multi-dimensional observation vector including the average tillage resistance, travel progress index, expected load change index, and field head space distance index; activating the corresponding target agent in a pre-trained set of parallel agents according to the target operating mode; inputting the multi-dimensional observation vector into the target agent and outputting an electric energy equivalent fuel consumption factor; substituting the electric energy equivalent fuel consumption factor into the Hamiltonian function constructed by the equivalent fuel consumption minimization strategy, and solving for the optimal control command that minimizes the Hamiltonian function.

[0007] In one implementation of this application, vehicle-mounted sensor data is fused to construct a multi-dimensional observation vector. Specifically, this includes: using a rolling window integral filtering algorithm to separate rolling resistance, air resistance, and acceleration resistance from the resultant force at the powertrain drive end to obtain the average tillage resistance, which characterizes soil cutting resistance; calculating the ratio of the length of the already completed path to the total length of the work path to obtain a travel progress index; acquiring a variable work prescription map from a smart agriculture cloud platform, extracting the trend of implement power change caused by the dependent variable work within a preset distance ahead of the current vehicle position to obtain an expected load change index; and extracting the real-time spatial distance between the current vehicle position and the next turnaround area in the work path to obtain a field-end spatial distance index.

[0008] In one implementation of this application, the method further includes: inversely calculating the average tillage resistance, and combining the current vehicle speed, drive wheel slip rate, and vehicle mass to obtain the soil compaction index; binding the soil compaction index with the current GPS spatiotemporal label and transmitting it back to the smart agriculture cloud platform to update the variable operation prescription map; constructing a digital twin model based on the soil compaction index; and solving the optimal battery state of charge reference trajectory based on the digital twin model and sending it to the vehicle-mounted edge system as the tracking target for the battery state of charge deviation term in the target agent's reward function.

[0009] In one implementation of this application, after finding the optimal control instruction that minimizes the Hamiltonian function, the method further includes: before executing the optimal control instruction, determining whether the optimal control instruction fully satisfies the powertrain physical constraint set; if it satisfies the powertrain physical constraint set, it is issued and executed normally; if it does not satisfy the powertrain physical constraint set, a degradation mode is automatically triggered, and a preset fallback control strategy is called to replace the optimal control instruction.

[0010] In one implementation of this application, a multi-dimensional observation vector is input into the target agent, and an energy-equivalent fuel consumption factor is output. Specifically, the initial operating modes include heavy load mode, medium load mode, and light load mode. The energy-equivalent fuel consumption factor is obtained by training a differentiated reward function, wherein the differentiated reward function is a weighted sum of a power penalty term, a fuel consumption rate term, and a battery state of charge deviation term, and the weight coefficients in the weighted sum are differentiated according to the heavy load mode, medium load mode, and light load mode corresponding to the target agent.

[0011] In one implementation of this application, during the training phase, the agent is configured with a battery state of charge reference trajectory corresponding to the operating mode; in heavy load mode, the battery state of charge reference trajectory maintains a high plateau area throughout; in light load mode, the battery state of charge reference trajectory is a ramp trajectory that decreases linearly according to the travel progress index.

[0012] In one implementation of this application, after obtaining the mean tillage resistance characterizing soil cutting resistance, the method further includes: simultaneously calculating the standard deviation of the instantaneous resistance value within the window during the rolling window integral filtering process; calculating the ratio of the standard deviation to the stationarity threshold, which is defined as the signal confidence index; when the signal confidence index is lower than a preset threshold, determining that the current mean tillage resistance signal is affected by vibration noise, proportionally attenuating the normalized weight of the mean tillage resistance component in the multi-dimensional observation vector, and proportionally allocating the attenuated weight to the expected load change index component; during the duration when the signal confidence index is lower than the preset threshold, limiting the update step size of the equivalent factor in the solution process of the minimum equivalent fuel consumption strategy to a preset small step size range.

[0013] In one implementation of this application, after obtaining the field head space distance index, the method further includes: when the field head space distance index is less than a distance threshold, determining that the tractor has entered the field head turning start stage, triggering the transient power pre-compensation mechanism of the power output shaft, reducing the torque contribution coefficient of the motor's power output shaft, and simultaneously increasing the output power ratio of the engine; when the tractor completes the field head turn, the field head space distance index becomes zero and the average tillage resistance recovers to a stable range, determining that the tractor is in the soil recovery stage, forcibly increasing the torque contribution coefficient of the motor's power output shaft, and calling the instantaneous peak power of the battery to compensate for the sudden increase in load on the power output shaft side.

[0014] Secondly, embodiments of this application also provide a multi-mode energy management device for a hybrid tractor, the device including at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: perform any one of the steps of a multi-mode energy management method for a hybrid tractor.

[0015] Thirdly, embodiments of this application also provide a non-volatile computer storage medium for multi-mode energy management of a hybrid tractor, storing computer-executable instructions configured to perform any one of the steps of a multi-mode energy management method for a hybrid tractor.

[0016] The multi-mode energy management method, device, and medium for hybrid tractors provided in this application construct a dual judgment system of task prior guidance and real-time data self-correction, and introduce a load moving average algorithm based on a time window. This effectively filters out instantaneous vibration noise in farmland operations, achieves accurate quantification of the actual soil operating resistance, significantly improves the accuracy and robustness of different working mode recognition such as deep plowing and rotary tillage, and avoids frequent oscillations in control commands. Furthermore, it constructs a three-in-one spatiotemporal prior perception system integrating macroscopic travel progress, mesoscopic prescription map expected load, and microscopic field turning distance, integrating global operation progress, forward... The introduction of variable operation sudden change trend and transient impact prediction at the field head into the energy optimization process enables the controller to have a global spatiotemporal view, realizing advance power replenishment before entering the high-intensity operation area and power compensation for transient impacts when entering the soil at the field head, which greatly improves energy utilization efficiency. Independent reinforcement learning agents are trained for three operation modes: heavy load, medium load, and light load. Through the design of differentiated reward functions, the reliability constraints of power output are strengthened in the heavy load mode, and the full utilization of electric energy is emphasized in the light load mode, which effectively solves the technical problem that a single model cannot take into account the energy consumption balance of various operating conditions. A two-layer control architecture of agent decision-making and ECMS (Equivalent Consumption Minimization Strategy) physical constraints is adopted. The upper layer is responsible for global energy efficiency optimization, and the lower layer combines the universal characteristic diagram of the engine and motor to strictly limit the physical boundaries of control commands, eliminating the risk of power interruption, overload, or battery overcharging and over-discharging caused by abnormal commands from a physical level. At the same time, the soil compaction is calculated by fusing the inverted real tillage resistance with the coordinates of the global positioning system and transmitted back to the cloud, dynamically correcting the digital twin model and variable operation prescription map. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a multi-mode energy management method for a hybrid tractor provided in this application embodiment; Figure 2 A schematic diagram of the working stroke ratio of a multi-mode energy management method for a hybrid tractor provided in this application embodiment; Figure 3A schematic diagram of a variable operation prescription for a multi-mode energy management method for a hybrid tractor provided in this application embodiment; Figure 4 A schematic diagram illustrating the estimated turning distance from the edge of the field for a multi-mode energy management method for a hybrid tractor provided in this application embodiment; Figure 5 A comparison diagram of reward mechanism preferences for a multi-mode energy management method for a hybrid tractor provided in this application embodiment; Figure 6 A strategy scheduling logic block diagram of a multi-mode energy management method for a hybrid tractor provided in this application embodiment; Figure 7 This is a schematic diagram illustrating the agent training convergence of a multi-mode energy management method for a hybrid tractor provided in an embodiment of this application. Figure 8 This is a schematic diagram of the internal structure of a multi-mode energy management device for a hybrid tractor provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] This application provides a multi-mode energy management method, device, and medium for hybrid tractors, which solves the problems of lack of global task awareness, poor anti-interference ability, and insufficient multi-mode adaptability in the existing hybrid tractor energy management methods.

[0020] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0021] Figure 1 A flowchart illustrating a multi-mode energy management method for a hybrid tractor provided in this application embodiment. Figure 1 As shown in the figure, the multi-mode energy management method for a hybrid tractor provided in this application embodiment specifically includes the following steps: Step 10: Obtain agricultural tasks to set the initial operation mode.

[0022] In this step, a pre-set list of agricultural tasks is obtained through the smart agriculture platform, and the working conditions are divided into heavy load mode: deep plowing and tillage; medium load mode: rotary tillage and sowing; and light load mode: spraying pesticides and fertilizing.

[0023] Step 20: Collect real-time driving resistance. When the real-time driving resistance deviates from the threshold range of the initial working mode within a continuous preset period, correct it to the corresponding target working mode.

[0024] In this step, the operating mode is dynamically calibrated by combining the driving resistance data monitored in real time by the sensors, so as to ensure that when the actual operating load deviates from the preset task threshold, the system can automatically identify and switch to the appropriate operating mode.

[0025] In practical implementation, to address differences in soil compaction, changes in plot boundaries, and task status, a sliding window-based online load verification algorithm is introduced: powertrain torque demand and tillage resistance are collected in real time and compared within a preset time window; a set of thresholds for mode switching is established. When the monitored actual load is continuous When a calculation cycle deviates from the preset threshold range of the current mode, it is determined that a condition drift has occurred; the controller updates the current mode based on the corrected load characteristics.

[0026] To prevent high-frequency switching of modes in boundary states, a hysteresis comparison logic is introduced. That is, mode switching must not only meet the threshold jump, but also the mode maintenance time after switching must exceed the preset steady-state threshold, thereby ensuring the smooth output of the power system and avoiding mechanical vibration and response distortion of the actuator caused by frequent mode switching.

[0027] Step 30: Integrate vehicle-mounted sensor data to construct a multi-dimensional observation vector, which includes the average tillage resistance, travel progress index, expected load change index, and field head space distance index.

[0028] In this step, to meet the real-time and accuracy requirements of the reinforcement learning decision layer for environmental state perception, heterogeneous data fusion is performed between vehicle-mounted multi-source sensor data and smart agriculture geographic information to construct a unified seven-dimensional observation vector. : ; Among them, the remaining battery power information Real-time vehicle speed Total torque demand And the core evolutionary quantity reflecting farmland operation load and spatiotemporal prior characteristics: mean tillage resistance. Work travel ratio Expected load increment Distance from the starting point of the road .

[0029] As an optional embodiment, the data from vehicle-mounted sensors are fused to construct a multi-dimensional observation vector, which may specifically include: Step 301: Using a rolling window integral filtering algorithm, rolling resistance, air resistance and acceleration resistance are separated from the resultant force at the drive end of the powertrain to obtain the mean value of tillage resistance characterizing soil cutting resistance.

[0030] In this step, to filter out high-frequency noise and spurious torque fluctuations in the mechanical transmission system caused by uneven farmland surfaces, a rolling window integral filtering algorithm is used to extract the actual soil cutting resistance. ; In the formula, The preset sliding time window; and These are rolling resistance and air resistance, respectively. For the overall quality of the tractor; The instantaneous acceleration of the tractor; This represents the resultant force exerted by the powertrain on the drive end. By removing non-agricultural impedance from the total driving force, this operator achieves a steady-state characterization of real agricultural loads under complex random disturbances.

[0031] Step 302: Calculate the ratio of the length of the completed path to the total length of the completed path to obtain the travel progress index.

[0032] In this step, the macro-time prior is: the ratio of work trips. Calculate the ratio of the current position to the total length of the preset task path to provide the system with a task progress reference throughout its entire lifecycle.

[0033] The task travel ratio is used to elevate the local instantaneous state of the vehicle to a global perspective of the entire path planning: ; In the formula, The cumulative length of the current vehicle's completed path. The total path length preset in the task description allows the agent to dynamically adjust the power release rate based on the remaining space in the task, improving the overall economy of energy allocation. Figure 2 As shown.

[0034] Step 303: Obtain the variable operation prescription map from the smart agriculture cloud platform, extract the trend of agricultural implement power change caused by the dependent variable operation within a preset distance in front of the current vehicle position, and obtain the expected load change index.

[0035] In this step, the mesoscopic space prior is used for the expected load increment: based on the variable operation prescription map issued from the cloud, the expected variable of fertilization or sowing within 100 meters ahead of the current location is extracted as the change in farm implement power due to fertilization or sowing, which is used for feedforward control to guide the system's power feedforward allocation. Figure 3 As shown.

[0036] Step 304: Extract the real-time spatial distance between the current vehicle position and the next U-turn area in the operation path to obtain the U-turn spatial distance index.

[0037] In this step, the microscopic spatial prior is the turning distance from the edge of the terrain: the real-time spatial distance between the vehicle and the turning area at the edge of the terrain is extracted and used for predictive control of transient impacts at the edge of the terrain. Figure 4 As shown.

[0038] As an optional embodiment, the method may further include: inversely calculating the average tillage resistance, and combining the current vehicle speed, drive wheel slip rate and total vehicle mass to obtain the soil compaction index.

[0039] In this step, during operation, the vehicle-mounted edge system integrates the average real tillage resistance obtained from the lossless inversion, the vehicle speed at the corresponding moment, the drive wheel slip rate, and the overall vehicle mass. Through the vehicle dynamics inverse evolution algorithm, it inversely calculates the high-precision soil compaction index of the current plot.

[0040] After binding the soil compaction index with the current GPS spatiotemporal label, it is transmitted back to the smart agriculture cloud platform to update the variable operation prescription map.

[0041] In this step, the system automatically binds the calculated soil compaction index to the real-time spatiotemporal tag provided by the vehicle-mounted high-precision global positioning system and transmits it back to the smart agriculture cloud platform via the wireless communication network.

[0042] A digital twin model was constructed based on the soil compaction index.

[0043] In this step, after the cloud platform collects the soil compaction data stream of the measured plots, it dynamically corrects the cloud-based farmland digital twin model and then automatically updates the variable operation prescription map for the next round of agricultural operations.

[0044] Based on the digital twin model, the optimal battery state of charge reference trajectory is solved and sent to the vehicle edge system as the tracking target of the battery state of charge deviation term in the reward function of the target agent.

[0045] In this step, during operation, the vehicle-mounted edge system integrates the average real tillage resistance obtained from the inversion extraction, the vehicle speed at the corresponding time, and the real-time driving efficiency to calculate a high-precision soil compaction index. It then binds a real-time GPS spatiotemporal tag and transmits it back to the smart agriculture cloud platform via a wireless communication network. The system uses the transmitted measured soil data to dynamically correct the digital twin system and automatically updates the variable operation prescription map for the next round of operations.

[0046] As an optional embodiment, after obtaining the mean tillage resistance characterizing soil cutting resistance, the method may further include: simultaneously calculating the standard deviation of the instantaneous resistance values ​​within the window during the rolling window integral filtering process; and calculating the ratio of the standard deviation to the stationary threshold, which is defined as the signal confidence index.

[0047] In this step, during the rolling window integral filtering process, the standard deviation of the instantaneous resistance values ​​within the window is calculated simultaneously. Specifically, within each sliding time window, the standard deviation σF is calculated using all instantaneous resistance sample values ​​within the window, reflecting the dispersion of the resistance signal within the window, i.e., the intensity of vibration noise.

[0048] The ratio of the standard deviation to the stationarity threshold is defined as the signal confidence index C. conf : When the signal confidence index is lower than the preset threshold, it is determined that the current average tillage resistance signal is affected by vibration noise. The normalized weight of the average tillage resistance component in the multi-dimensional observation vector is reduced proportionally, and the weight released by the reduction is distributed proportionally to the expected load change index component. In this step, σth is a preset stability threshold, representing the upper limit of the standard deviation of the resistance signal under stable operating conditions. When C conf When C is less than 1, it indicates that the current signal quality is good; when C... conf A value greater than or equal to 1 indicates severe noise interference in the signal. When the signal confidence index is below a preset threshold, it is determined that the current average tillage resistance signal is affected by vibration noise. In this case, the normalized weight of the average tillage resistance component in the multi-dimensional observation vector is proportionally attenuated, with the specific attenuation coefficient set to α = 1 / C. conf The weight of attenuation release is proportionally allocated to the expected load change index component, thereby realizing dynamic reconstruction of the observation space and increasing the decision-making weight of prescription map prior information in a severe vibration environment.

[0049] During the duration when the signal confidence index is below a preset threshold, the update step size of the equivalent factor in the process of solving the equivalent fuel consumption minimization strategy is limited to a preset small step size range.

[0050] In this step, during the duration when the signal confidence index is below a preset threshold, the update step size of the equivalent factor in the process of solving the equivalent fuel consumption minimization strategy is limited to a preset small step size range. That is, the maximum change of the equivalent factor s output by the upper-level agent in each control cycle is forced not to exceed the preset limit value, so as to avoid the equivalent factor from fluctuating violently due to the unreliability of the resistance signal and to ensure the stability of the solution of the underlying equivalent fuel consumption minimization strategy.

[0051] As an optional embodiment, after obtaining the spatial distance index, the method may further include: When the distance index at the edge of the field is less than the distance threshold, it is determined that the tractor has entered the starting stage of turning at the edge of the field, triggering the transient power pre-compensation mechanism of the power output shaft, reducing the torque contribution coefficient of the motor's power output shaft, and simultaneously increasing the output power ratio of the engine.

[0052] In this step, when the distance indicator at the edge of the field is less than a preset distance threshold, it is determined that the tractor is about to enter the starting stage of a turn at the edge of the field. At this time, the system triggers the transient power pre-compensation mechanism of the power output shaft: First, it reduces the torque contribution coefficient of the motor's power output shaft, so that the motor reduces the power output to the power output shaft side; at the same time, it synchronously increases the output power ratio of the engine, so that the engine bears the main load of the power output shaft. In this way, during the process of the tractor lifting the implement to turn at the edge of the field, the mechanical load on the power output shaft side is actively released, avoiding drastic fluctuations in bus voltage or engine stalling due to sudden load changes.

[0053] When the tractor completes the turn at the edge of the field, the field distance index becomes zero, and the average tillage resistance returns to a stable range, the tractor is determined to be in the soil recovery stage. The torque contribution coefficient of the motor's power output shaft is forcibly increased, and the instantaneous peak power of the battery is used to compensate for the sudden increase in load on the power output shaft side.

[0054] In this step, when the tractor completes the turn at the edge of the field, the distance indicator at the edge of the field becomes zero (meaning the vehicle has left the field area), and the average tillage resistance returns to a stable range (e.g., the fluctuation amplitude is less than the preset stable fluctuation threshold over several consecutive control cycles), the tractor is determined to be in the soil recovery phase. At this time, the system executes the opposite power compensation action: forcibly increasing the torque contribution coefficient of the motor's power output shaft and utilizing the instantaneous peak power of the battery to compensate for the sudden increase in load on the power output shaft side caused by the implement re-entering the soil. Through the above-mentioned transient power pre-compensation and post-compensation linkage mechanism, power interruption at the moment of soil entry is effectively avoided, ensuring the continuity of operation and agronomic accuracy.

[0055] Step 40: Activate the corresponding target agent in the pre-trained set of parallel agents according to the target job mode.

[0056] In this step, the mid-level decision module deploys three parallel reinforcement learning agents, corresponding to the three modes of heavy load, medium load and light load respectively. Each agent takes the observation vector as input and outputs the optimal energy-equivalent fuel consumption factor (i.e. equivalent factor) in real time, transforming the complex global energy consumption optimization problem into a dynamic parameter adjustment problem.

[0057] Step 50: Input the multi-dimensional observation vector into the target agent and output the electric energy equivalent fuel consumption factor.

[0058] In this step, each agent uses the constructed seven-dimensional observation vector As the input signal, it undergoes feature mapping and action space search within a multi-layer sensing network, ultimately outputting a continuous-domain action value, namely the electrical energy equivalent fuel consumption factor. It serves as a weighting factor for the value of oil and electricity within the system, and is defined as the equivalent mass of fuel converted per unit of battery electrical energy. The magnitude of the value directly determines the degree of energy saving or the intensity of energy release during instantaneous optimization by the underlying controller. For example, when the agent outputs a high value... When the output is low, the system tends to prioritize using the engine to conserve electricity; conversely, when the output is low... At that time, the system tends to increase the electric motor assist ratio to reduce fuel consumption.

[0059] As an optional embodiment, a multi-dimensional observation vector is input into the target agent, and an energy-equivalent fuel consumption factor is output. Specifically, it may include: an initial operating mode including a heavy load mode, a medium load mode, and a light load mode; and an energy-equivalent fuel consumption factor is obtained by training a differentiated reward function, wherein the differentiated reward function is a weighted sum of a power penalty term, a fuel consumption rate term, and a battery state of charge deviation term, and the weight coefficients in the weighted sum are differentiated according to the heavy load mode, medium load mode, and light load mode corresponding to the target agent.

[0060] In this step, differentiated target assessment and incentive mechanisms are set according to the energy consumption characteristics of different modes; based on the digital operation instructions obtained from the smart agriculture management cloud platform, the system analyzes the current task type and presets the initial operation mode.

[0061] Heavy-duty mode: Suitable for high-torque conditions such as deep plowing, land clearing, and fully loaded trailer transportation; Medium load mode: Suitable for agricultural operations with moderate power loads, such as rotary tillage, sowing, and ridging; Light load mode: Suitable for low power demand conditions such as spraying, fertilizing, spreading fertilizer, and patrol inspection.

[0062] Heavy-load mode: The core focus is on maintaining stable power output and balancing battery state of charge. For example, when a microscopic prior indication indicates that the battery is about to be buried, the agent uses a high-penalty mechanism to prevent power interruption due to insufficient power.

[0063] Medium-load mode: Aiming at optimal overall energy consumption, the system is guided to find the best balance between the engine's high-efficiency range and the electric motor's auxiliary power. For example, if the mesoscopic prior prediction indicates a high-load prescription area ahead, the excitation system adjusts the equivalent factor in advance to increase the speed to compensate for the power consumption, reserving an energy margin.

[0064] Light load mode: The excitation system utilizes electrical energy more actively. For example, when the stroke ratio is nearing completion and there is no heavy load area ahead, it maximizes power consumption to achieve full energy utilization.

[0065] To enable the agent to achieve an optimal balance between energy consumption and power performance for different agricultural tasks, this application customizes reward evaluation functions for three operating modes, quantifying energy management objectives into traceable numerical indicators. Reward Function Design Logic Defined as the weighted sum of the power penalty, fuel economy penalty, and SOC deviation: ; in, These are weighting coefficients that are dynamically adjusted according to different working conditions. This represents the amount of unmet total power demand. Engine fuel consumption rate; This is the preset target value for the battery's state of charge.

[0066] Heavy-load job mode: Set weight The primary optimization objective is power preservation. When the microscopic prior indicates that the vehicle is about to finish turning and re-enter the soil, or when the system detects that the total power demand is not fully met, [the system will] [take action]. The item is assigned a very high weighting coefficient, which forces the intelligent agent to output a high impedance equivalent factor for pre-charging before entering the soil, and to call up the full-capacity battery to release peak power at the moment of entering the soil, so as to avoid losing power and stalling.

[0067] Medium load operation mode: Set weight Focusing on optimizing total system energy consumption, when a high-intensity variable operation zone is predicted to be approaching through meso-level prior analysis, the system is guided to adjust its power consumption in advance. Pre-charging allows for a voltage margin, enabling the engine and motor to work together in the optimal fuel consumption range, thus ensuring the agronomic precision and energy efficiency of fertilization and sowing.

[0068] Light load operation mode: Set weight When the travel ratio indicates the task is nearing completion and there are no heavy-load conditions in the subsequent prescription map, the system proactively reduces the power consumption maintenance constraint, incentivizing the system to more actively consume the remaining power, thereby maximizing the global economy of energy allocation. The reward mechanism preferences under the three different modes are compared as follows: Figure 5 As shown.

[0069] Step 60: Substitute the electric energy equivalent fuel consumption factor into the Hamiltonian function constructed by the equivalent fuel consumption minimization strategy, and solve for the optimal control command that minimizes the Hamiltonian function.

[0070] In this step, the lower execution layer receives the equivalent factor sent down from the middle layer. By constructing a Hamiltonian function through an equivalent fuel consumption minimization strategy, the optimal control command is solved under the physical constraints of the battery and powertrain. The system integrates hard constraint limiting logic to ensure that the calculated target torque of the engine and motor is always within the mechanical safety limits, thus ensuring the absolute safety of tractor operation while tapping into the system's energy-saving potential.

[0071] The lower control layer serves as the secure execution foundation of the system, receiving the equivalent factor output in real time by the middle-layer intelligent agent. The torque distribution of the power source is achieved by solving for the instantaneous minimum value of the Hamiltonian function.

[0072] The system is based on the dynamic equivalent factor issued by the intelligent agent. Construct an evaluation index for instantaneous equivalent energy consumption, using the Hamiltonian function. : ; in, This is the system control command vector; Instantaneous fuel consumption rate; The charging and discharging power of the battery terminal; For fuels with low calorific value coefficient, through The cost of instantaneous power consumption of electricity is converted into the cost of fuel consumption in a homogeneous manner.

[0073] To ensure execution safety, optimization solutions based on hard constraints apply strict physical constraints imposed on the dynamic system by the lower-level control layer. Find the optimal solution internally: ; constraint set The system covers the following key physical limitations: constraints imposed by the current battery state of charge level and peak charge / discharge power limits; constraints imposed by the engine's universal characteristic curve and dynamic torque response physical limits; and constraints imposed by the motor's field weakening characteristics, steady-state and dynamic torque response ranges. The system searches for the optimal power output under the given conditions by traversing the discrete solution space within the feasible region in real time. Control command to reach the minimum value.

[0074] As an optional embodiment, after solving for the optimal control command that minimizes the Hamiltonian function, the method may further include: before executing the optimal control command, determining whether the optimal control command fully satisfies the powertrain physical constraint set; if it satisfies the powertrain physical constraint set, it is issued and executed normally; if it does not satisfy the powertrain physical constraint set, a degradation mode is automatically triggered, and a preset fallback control strategy is called to replace the optimal control command.

[0075] In this step, the optimized target control quantity is converted by the configuration adaptation layer and then sent to the vehicle controller in real time via the controller area network bus. The underlying system integrates anomaly truncation logic: it can only be executed if and only if the allocation result fully meets the powertrain safety limits; if the optimization command exceeds the limits due to communication failure or instantaneous load change, the system will automatically trigger the degradation mode and execute a rule-based fallback control strategy, thereby fundamentally eliminating the risks of powertrain overload, engine shutdown, or battery overcharging and over-discharging while leveraging intelligent optimization performance.

[0076] As an optional embodiment, before activating the corresponding target agent in the pre-trained set of parallel agents, the method may further include: Step 01: During the training phase, the agent is configured with a battery state of charge reference trajectory corresponding to the operating mode; Step 02: In heavy load mode, the battery state of charge reference trajectory maintains a high plateau region throughout; Step 03: In light load mode, the battery state of charge reference trajectory is a ramp trajectory that linearly decreases according to the travel progress index.

[0077] The system pre-constructs three independent deep reinforcement learning agents, named the heavy-load mode agent, the medium-load mode agent, and the light-load mode agent, respectively. Each agent is trained in an offline simulation environment for reinforcement learning based on the different agricultural operation load distributions of heavy, medium, and light, and learns the optimal action probability distribution under the corresponding working conditions.

[0078] The decision-making layer identifies signals based on real-time dynamic feedback of operating modes and activates the corresponding target agents in real time through the scheduling logic unit. When the system undergoes a change in operating conditions, the scheduler employs a soft-smooth switching logic to smoothly transition the decision outputs of the agents before and after the change, preventing step fluctuations in control factors. The decision logic block diagram is as follows: Figure 6 As shown.

[0079] The equivalent fuel consumption minimization strategy, built in Matlab and incorporating the soft actor-critic algorithm, is illustrated in the agent training convergence diagram below. Figure 7 As shown.

[0080] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a multi-mode energy management device for a hybrid tractor, the structure of which is as follows: Figure 8 As shown.

[0081] Figure 8 This application provides a schematic diagram of the internal structure of a multi-mode energy management device for a hybrid tractor. (See attached diagram.) Figure 8 As shown, the device includes: At least one processor 801; And a memory 802 that is communicatively connected to at least one processor; The memory 802 stores instructions that can be executed by at least one processor, which are executed by at least one processor 801 to enable at least one processor 801 to: perform any one of the steps of a hybrid tractor multi-mode energy management method.

[0082] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium for multi-mode energy management of a hybrid tractor stores computer-executable instructions, which are configured to execute any one of the steps of a multi-mode energy management method for a hybrid tractor.

[0083] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0084] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0085] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0086] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0087] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0088] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0089] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0090] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0091] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0092] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0093] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A multi-mode energy management method for a hybrid tractor, characterized in that, The method includes: Obtain agricultural tasks to set the initial operation mode; Real-time driving resistance is collected, and when the real-time driving resistance deviates from the threshold range of the initial operation mode within a continuous preset period, it is corrected to the corresponding target operation mode. By integrating data from vehicle-mounted sensors, a multi-dimensional observation vector is constructed, which includes the average tillage resistance, travel progress index, expected load change index, and field head space distance index. According to the target operation mode, activate the corresponding target agent in the pre-trained set of parallel agents; The multi-dimensional observation vector is input into the target intelligent agent, and the electric energy equivalent fuel consumption factor is output. Substitute the electrical energy equivalent fuel consumption factor into the Hamiltonian function constructed by the equivalent fuel consumption minimization strategy, and solve for the optimal control command that minimizes the Hamiltonian function.

2. The multi-mode energy management method for a hybrid tractor according to claim 1, characterized in that, The fusion of vehicle sensor data to construct a multi-dimensional observation vector specifically includes: By using a rolling window integral filtering algorithm, rolling resistance, air resistance, and acceleration resistance are separated from the resultant force at the drive end of the powertrain to obtain the mean value of tillage resistance, which characterizes soil cutting resistance. The ratio of the length of the completed work path to the total length of the work path is calculated to obtain the travel progress index; The variable operation prescription map is obtained from the smart agriculture cloud platform, and the trend of the change in farm implement power caused by the dependent operation within a preset distance in front of the current vehicle position is extracted to obtain the expected load change index. Extract the real-time spatial distance between the current vehicle position and the next U-turn area in the work path to obtain the U-turn spatial distance index.

3. The multi-mode energy management method for a hybrid tractor according to claim 2, characterized in that, The method further includes: The average tillage resistance is calculated in reverse, and the soil compaction index is obtained by combining the current vehicle speed, drive wheel slip rate and vehicle mass. After binding the soil compaction index with the current GPS spatiotemporal label, it is transmitted back to the smart agriculture cloud platform to update the variable operation prescription map; Based on the soil compaction index, a digital twin model is constructed; Based on the digital twin model, the optimal battery state of charge reference trajectory is solved and sent to the vehicle edge system as the tracking target for the battery state of charge deviation term in the reward function of the target agent.

4. The multi-mode energy management method for a hybrid tractor according to claim 1, characterized in that, After finding the optimal control command that minimizes the Hamiltonian function, the method further includes: Before executing the optimal control command, it is determined whether the optimal control command fully satisfies the set of powertrain physical constraints; If the set of powertrain physical constraints is met, the command and execution will proceed normally. If the set of powertrain physical constraints is not met, a degradation mode is automatically triggered, and a preset fallback control strategy is invoked to replace the optimal control command.

5. The multi-mode energy management method for a hybrid tractor according to claim 1, characterized in that, The step of inputting the multi-dimensional observation vector into the target intelligent agent and outputting the electrical energy equivalent fuel consumption factor specifically includes: The initial operating modes include heavy load mode, medium load mode and light load mode; The equivalent fuel consumption factor of electric energy is obtained by training a differentiated reward function, wherein the differentiated reward function is a weighted sum of a power penalty term, a fuel consumption rate term, and a battery state of charge deviation term, and the weight coefficients in the weighted sum are differentiated according to the heavy load mode, medium load mode, and light load mode corresponding to the target agent.

6. A multi-mode energy management method for a hybrid tractor according to claim 5, characterized in that, Before activating the corresponding target agent in the pre-trained set of parallel agents, the method further includes: During the training phase, the agent is configured with a battery state of charge reference trajectory corresponding to the operating mode; In the heavy load mode, the battery state of charge reference trajectory maintains a high plateau region throughout the entire process; In the light load mode, the battery state of charge reference trajectory is a ramp trajectory that decreases linearly according to the travel progress index.

7. A multi-mode energy management method for a hybrid tractor according to claim 2, characterized in that, After obtaining the mean tillage resistance, which characterizes soil cutting resistance, the method further includes: During the rolling window integral filtering process, the standard deviation of the instantaneous resistance value within the window is calculated simultaneously. The ratio of the standard deviation to the stationarity threshold is calculated and defined as the signal confidence index; When the signal confidence index is lower than a preset threshold, it is determined that the current average tillage resistance signal is affected by vibration noise. The normalized weight of the average tillage resistance component in the multi-dimensional observation vector is reduced proportionally, and the weight released by the reduction is distributed proportionally to the expected load change index component. During the duration when the signal confidence index is below a preset threshold, the update step size of the equivalent factor in the process of solving the equivalent fuel consumption minimization strategy is limited to a preset small step size range.

8. A multi-mode energy management method for a hybrid tractor according to claim 2, characterized in that, After obtaining the spatial distance index, the method further includes: When the distance index at the edge of the field is less than the distance threshold, it is determined that the tractor has entered the starting stage of turning at the edge of the field, triggering the transient power pre-compensation mechanism of the power output shaft, reducing the torque contribution coefficient of the motor's power output shaft, and simultaneously increasing the output power ratio of the engine. When the tractor completes the turn at the edge of the field, the field edge spatial distance index becomes zero, and the average tillage resistance returns to a stable range, the tractor is determined to be in the soil recovery stage. The torque contribution coefficient of the motor's power output shaft is forcibly increased, and the instantaneous peak power of the battery is used to compensate for the sudden increase in load on the power output shaft side.

9. A multi-mode energy management device for a hybrid tractor, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Perform the steps of the multi-mode energy management method for a hybrid tractor as described in any one of claims 1-8.

10. A non-volatile computer storage medium for multi-mode energy management of a hybrid tractor, storing computer-executable instructions, characterized in that, The computer-executable instructions are set as follows: Perform the steps of the multi-mode energy management method for a hybrid tractor as described in any one of claims 1-8.

Images