LSTM-based tractor load prediction and hybrid control device and method

Abstract

The application relates to a tractor load prediction and hybrid control device and method based on an LSTM, which comprises a data acquisition unit, a load prediction unit, a hybrid control unit, an execution unit and a control unit ECU; the data acquisition unit is used for acquiring multi-source sensor data of the tractor; the load prediction unit and the hybrid control unit are both integrated in the control unit ECU, and the load prediction unit is used for processing the multi-source sensor data, and the hybrid control unit is used for generating a driving control strategy; the execution unit comprises a slip film controller used for adjusting a driving wheel slip rate and a driving control module used for adjusting parameters according to the control strategy. Through deep fusion of LSTM model time sequence feature extraction and physical model mechanism constraint, the application significantly improves the prediction accuracy of a sudden load and solves the problem of poor generalization ability of a single model.

Description

Technical Field

[0001] This invention belongs to the field of tractor control technology, and specifically relates to a tractor load prediction and hybrid control device and method based on LSTM. Background Technology

[0002] Tractor plowing is a core component of agricultural production, and its level of intelligence and green technology directly impacts national food security and sustainable agricultural development. Currently, my country is at a critical juncture of deep integration between agricultural modernization and intelligent manufacturing, and the national government has introduced a series of policies to promote the high-end, intelligent, and green development of agricultural machinery.

[0003] In terms of policy guidance, it is clearly stated that the transformation and upgrading of agricultural mechanization towards full-process, comprehensive, high-quality, and efficient development should be accelerated, with a focus on developing intelligent agricultural machinery and equipment to improve operational precision and efficiency. Simultaneously, intelligent manufacturing is prioritized, encouraging the use of artificial intelligence, big data, and other technologies to transform traditional manufacturing, with agricultural machinery being a key application scenario. Furthermore, under the "dual-carbon" target framework, the state is vigorously promoting the application of new and clean energy in agricultural machinery. Diesel-electric hybrid tractors, as a transitional technology, combine economic efficiency and environmental friendliness, making them a key area of ​​policy support. These policies provide clear policy guidance and market impetus for the research and application of intelligent tractor control technology.

[0004] In terms of the current state of the industry, with the rapid development of sensor technology, the Internet of Things, and artificial intelligence, intelligent agricultural machinery has become a new battleground for global competition. Internationally, companies such as John Deere and CNHJ have launched intelligent tractors with autonomous driving and variable-operation capabilities, the core of which lies in high-precision environmental perception and adaptive control. Domestically, leading companies such as China YTO Group and Lovol Heavy Industry are also actively developing intelligent tractors, with some products already achieving Level 2 autonomous driving. However, there is still room for improvement in their dynamic load adaptability in complex farmland environments.

[0005] Specifically, in plowing operations, existing intelligent control technologies face two main challenges: First, the accuracy of load prediction is insufficient. During plowing, factors such as soil moisture, terrain undulations, and crop residues cause drastic load fluctuations. Traditional methods based on single sensors or fixed models cannot accurately predict sudden load changes in real time, which can easily lead to excessive slippage of the drive wheels, reducing traction efficiency and increasing fuel consumption. Second, the control strategy is not adaptable. Most existing systems use fixed control logic or simple feedback adjustment, which cannot dynamically optimize control parameters according to load changes, resulting in large operational impacts, poor fuel economy, and difficulty in meeting the requirements of refined and low-loss operations.

[0006] At the technological research level, academia and industry are actively exploring solutions. In load prediction, Long Short-Term Memory (LSTM) networks have been introduced due to their advantage in processing time-series data; however, purely data-driven models are susceptible to sample limitations and may deviate from physical laws. Regarding control strategies, Model Predictive Control (MPC) can provide forward-looking optimization but relies on an accurate model; reinforcement learning can learn adaptively, but online training carries high risks. Therefore, hybrid control methods that integrate data-driven approaches with physical models and combine MPC with offline reinforcement learning have become a current research hotspot and trend.

[0007] In conclusion, under the strong impetus of national policies and the rapid iteration of industry technologies, the development of an intelligent control system capable of accurately predicting sudden load changes in real time and generating adaptive control strategies is of great practical significance and urgent need for improving the level of intelligent tractor operation in my country and achieving cost reduction, efficiency improvement and green production in agriculture. Summary of the Invention

[0008] To address the problems of existing technologies, this invention proposes a tractor load prediction and hybrid control device and method based on LSTM. The objectives of this invention and the technical problems it solves are achieved through the following technical solutions.

[0009] According to the present invention, a tractor load prediction and hybrid control device based on LSTM is proposed, comprising a data acquisition unit, a load prediction unit, a hybrid control unit, an execution unit, and a control unit ECU; the data acquisition unit is used to acquire multi-source sensor data of the tractor; the load prediction unit and the hybrid control unit are both integrated into the control unit ECU, and the load prediction unit is used to process the multi-source sensor data, while the hybrid control unit is used to generate a drive control strategy; the execution unit includes a diaphragm controller for adjusting the slip rate of the drive wheels and a drive control module for adjusting parameters according to the control strategy.

[0010] Furthermore, the data acquisition unit includes a vehicle speed sensor mounted on the tractor chassis for measuring tillage speed, a drive wheel speed sensor mounted on the drive axle for measuring drive wheel speed, a steering angle sensor mounted on the tractor for calculating tillage depth, an upper tie rod sensor mounted on the upper tie rod pin for measuring the upper tie rod force, a lower tie rod sensor mounted on the lower tie rod pin for measuring the lower tie rod force, an engine torque sensor mounted on the engine output shaft for measuring torque, a soil moisture sensor mounted on the plowing mechanism for measuring soil moisture, and a battery status detection module for detecting the battery status of the diesel-electric hybrid system.

[0011] Furthermore, the load prediction unit includes a data preprocessing module for processing multi-source sensor data, an LSTM encoding module for encoding time-series data, and a data model coupling module for generating the final load prediction value.

[0012] Furthermore, the hybrid control unit includes an MPC module for optimizing control sequences and an offline reinforcement learning module for generating drive instructions.

[0013] A control method for a tractor load prediction and hybrid control device based on LSTM includes the following steps: S1, Data Acquisition and Processing: Acquire data from multiple source sensors and perform preprocessing; S2, Load Forecasting: Predicting tractor load based on LSTM and physical models; S3, Hybrid Control: Combines the MPC module and the offline reinforcement learning module to generate driving control strategies; S4, slip ratio control: Adjust the slip ratio based on the synovial adaptive method to optimize traction performance.

[0014] Furthermore, in step S2, the specific steps for load prediction are as follows: first, an LSTM network is used to encode the job time series data, and then, combined with the nonlinear dynamic model of the dynamic system and the longitudinal-transverse coupled nonlinear dynamic model, the predicted value of sudden load is output.

[0015] Furthermore, in step S3, the specific steps of hybrid control are as follows: the control sequence is optimized based on model prediction by the MPC module, the strategy is trained based on historical data by the offline reinforcement learning module, and the two are combined to generate the final control strategy.

[0016] Furthermore, in step S3, the specific steps of hybrid control are as follows: the control sequence is optimized based on model prediction by the MPC module, the strategy is trained based on historical data by the offline reinforcement learning module, and the two are combined to generate the final control strategy.

[0017] In summary, the present invention has the following advantages: 1. By deeply integrating the temporal feature extraction of the LSTM model with the mechanism constraints of the physical model, the prediction accuracy of the mutation load is significantly improved, and the problem of poor generalization ability of the single model is solved.

[0018] 2. By combining forward optimization of the MPC module with empirical decision-making through offline reinforcement learning, the system can respond in advance to load changes within the next 5 seconds, effectively reducing the impact of dynamic loads and optimizing fuel consumption.

[0019] 3. Through sliding mode adaptive control, the speed and robustness of slip ratio tracking are ensured, and the intelligent operation level of the unit in complex terrain is improved.

[0020] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0021] Figure 1 This is a structural block diagram of a tractor load prediction and hybrid control device based on LSTM according to the present invention.

[0022] Figure 2 This is a flowchart illustrating the operation of the load prediction unit in this invention.

[0023] Figure 3 This is a flowchart illustrating the operation of the hybrid control unit in this invention. Detailed Implementation

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0025] Please refer to Figure 1. Figure 2 , Figure 3 A tractor load prediction and hybrid control device based on LSTM includes a data acquisition unit, a load prediction unit, a hybrid control unit, an execution unit, and a control unit ECU. The data acquisition unit is used to collect multi-source sensor data from the tractor, specifically including a vehicle speed sensor mounted on the tractor chassis for measuring tillage speed, a drive wheel speed sensor mounted on the drive axle for measuring drive wheel speed, a steering angle sensor mounted on the tractor for calculating tillage depth, an upper tie rod sensor mounted on the upper tie rod pin for measuring the force of the upper tie rod, a lower tie rod sensor mounted on the lower tie rod pin for measuring the force of the lower tie rod, an engine torque sensor mounted on the engine output shaft for measuring torque, a soil moisture sensor mounted on the plowing mechanism for measuring soil moisture, and a battery status detection module for detecting the battery status of the diesel-electric hybrid system.

[0026] Both the load prediction unit and the hybrid control unit are integrated within the control unit (ECU). The load prediction unit processes multi-source sensor data, specifically including a data preprocessing module and a data model coupling module. The data preprocessing module performs denoising, normalization, and feature extraction on the multi-source sensor data to eliminate dimensional influences. The LSTM encoding module encodes the time-series data, employing a two-layer LSTM network with 128 neurons per layer. This allows for modeling load change characteristics based on the time-series data and outputting preliminary load prediction values. The data model coupling module generates the final abrupt load prediction value by coupling the LSTM output with the nonlinear dynamic model of the power system and the longitudinal-transverse coupled nonlinear dynamic model, and then using a weighted fusion formula to calculate the final load prediction value F_final. F_final = α•F_LSTM + (1-α)•F_physics; Where F_LSTM is the initial predicted value output by LSTM, F_physics is the theoretical value output by the dynamic system and the coupled dynamic model, and α is the dynamic adjustment coefficient (range 0.6~0.8).

[0027] The hybrid control unit is used to generate drive control strategies, specifically including an MPC module and an offline reinforcement learning module. The MPC module takes the aforementioned predicted sudden load and unit status as inputs and uses a quadratic programming algorithm to optimize the control sequence within the next 5-second time window, ensuring the foresight and stability of the control. The offline reinforcement learning module uses the SAC (Soft Actor-Critic) algorithm, taking the predicted load and unit status as inputs and the throttle opening and steering angle as outputs, to train the control strategy based on historical operating data, generate drive commands that adapt to load changes (different site conditions), and combine the control sequence to generate the final control strategy, thus balancing real-time performance and adaptability.

[0028] The execution unit specifically includes a synovial controller and a drive control module. The synovial controller is designed based on Lyapunov stability theory and uses a synovial adaptive control method to adjust the slip ratio of the drive wheels according to the control strategy to maintain traction efficiency. The drive control module can adjust parameters such as throttle opening and steering angle according to the control strategy.

[0029] A control method for a tractor load prediction and hybrid control device based on LSTM includes the following steps: S1, Data Acquisition and Processing: The sampling frequency is 100Hz, specifically for tillage speed, drive wheel speed, tillage depth, upper lever force, lower lever force, engine torque, battery status, and soil moisture. The collected data is preprocessed, including noise reduction (median filtering to eliminate outliers), normalization (mapping each parameter to the [0,1] interval), and feature extraction (extracting feature parameters, including mean, variance, rate of change, etc.).

[0030] S2, Load Prediction: Specifically, the preprocessed job time series data is input into the LSTM network, which encodes and extracts features from the job time series data. At the same time, the ReLU activation function is used to output the preliminary load prediction value. The preliminary load prediction value is input into the physical model coupling module, which is corrected by combining the nonlinear dynamic model of the dynamic system and the longitudinal-transverse coupled nonlinear dynamic model, and outputs the final sudden load prediction value.

[0031] S3, Hybrid Control: Specifically, based on load forecasts and unit status, the MPC module optimizes the control sequence within the future time window; the offline reinforcement learning module uses historical data to train the SAC algorithm offline and generate drive commands that adapt to load changes; the two are combined to generate the final control strategy.

[0032] S4, slip ratio control: Specifically, a sliding mode controller is designed based on Lyapunov stability theory, a sliding film surface is constructed, the deviation between the current slip ratio and the optimal slip ratio is calculated in real time, the drive wheel torque is adjusted by the drive wheel slip ratio to maintain optimal traction efficiency, and the control strategy is executed to adjust parameters such as throttle opening and steering angle.

[0033] The present invention was further verified using the above-described device and control method on the Matlab / Simulink simulation platform: 1. Construct a tractor dynamics model and a plowing load model; 2. Collect historical assignment data to train the LSTM network and offline reinforcement learning strategy; 3. Conduct simulation experiments to verify the accuracy of load prediction and the effectiveness of the control strategy; 4. Conduct simulation experiments to compare the performance differences between the proposed solution and the single LSTM solution and the single MPC solution.

[0034] Referring to the experimental results of the control method of the present invention and the traditional control method in Table 1 below, it can be seen that the present invention has the following effects: 1. Load prediction accuracy: The proposed solution reduces prediction error by approximately 15%-20% compared to a single LSTM model; 2. Impact intensity index: reduced by approximately 25%-30% compared to traditional control methods; 3. Fuel consumption: Approximately 10%-15% lower than traditional control methods; 4. Hardware-in-the-loop testing verifies that the model accuracy and policy real-time performance meet control requirements.

[0035] Based on the above test results, the present invention can effectively predict sudden load changes, reduce the impact and fuel consumption under dynamic loads, and improve the intelligence level of tractors.

[0036] The above description is merely a preferred embodiment of the present invention. Any simple modifications, equivalent changes, and alterations made by those skilled in the art to the above embodiments without departing from the scope of the present invention and based on the technical essence of the present invention shall still fall within the scope of the present invention.

Claims

1. A tractor load prediction and hybrid control device based on LSTM, characterized in that: It includes a data acquisition unit, a load prediction unit, a hybrid control unit, an execution unit, and a control unit (ECU). The data acquisition unit is used to collect multi-source sensor data from the tractor; Both the load prediction unit and the hybrid control unit are integrated into the control unit ECU. The load prediction unit is used to process multi-source sensor data, and the hybrid control unit is used to generate drive control strategies. The execution unit includes a diaphragm controller for adjusting the slip rate of the drive wheels and a drive control module for adjusting parameters according to the control strategy.

2. The LSTM-based tractor load prediction and hybrid control device according to claim 1, characterized in that: The data acquisition unit includes a vehicle speed sensor mounted on the tractor chassis for measuring tillage speed, a drive wheel speed sensor mounted on the drive axle for measuring drive wheel speed, a steering angle sensor mounted on the tractor for calculating tillage depth, an upper tie rod sensor mounted on the upper tie rod pin for measuring the upper tie rod force, a lower tie rod sensor mounted on the lower tie rod pin for measuring the lower tie rod force, an engine torque sensor mounted on the engine output shaft for measuring torque, a soil moisture sensor mounted on the plowing mechanism for measuring soil moisture, and a battery status detection module for detecting the battery status of the diesel-electric hybrid system.

3. The LSTM-based tractor load prediction and hybrid control device according to claim 1, characterized in that: The load prediction unit includes a data preprocessing module for processing multi-source sensor data, an LSTM encoding module for encoding time-series data, and a data model coupling module for generating the final load prediction value.

4. The LSTM-based tractor load prediction and hybrid control device according to claim 1, characterized in that: The hybrid control unit includes an MPC module for optimizing control sequences and an offline reinforcement learning module for generating drive instructions.

5. A control method for a tractor load prediction and hybrid control device based on LSTM, comprising the tractor load prediction and hybrid control device based on LSTM as described in any one of claims 1-4, characterized in that, Includes the following steps: S1, Data Acquisition and Processing: Acquire data from multiple source sensors and perform preprocessing; S2, Load Forecasting: Predicting tractor load based on LSTM and physical models; S3, Hybrid Control: Combines the MPC module and the offline reinforcement learning module to generate driving control strategies; S4, slip ratio control: Adjust the slip ratio based on the synovial adaptive method to optimize traction performance.

6. The control method for a tractor load prediction and hybrid control device based on LSTM according to claim 5, characterized in that: In step S2, the specific steps for load prediction are as follows: First, the LSTM network is used to encode the job time series data, and then the nonlinear dynamic model of the dynamic system and the longitudinal-transverse coupled nonlinear dynamic model are combined to output the predicted value of sudden load.

7. The control method for a tractor load prediction and hybrid control device based on LSTM according to claim 5, characterized in that: In step S3, the specific steps of hybrid control are as follows: the control sequence is optimized based on model prediction by the MPC module, the strategy is trained based on historical data by the offline reinforcement learning module, and the two are combined to generate the final control strategy.

8. The control method for a tractor load prediction and hybrid control device based on LSTM according to claim 5, characterized in that: In step S4, the specific steps of slip ratio control are as follows: based on the slip ratio controller, a slip ratio surface is constructed, and the slip ratio of the drive wheel is adjusted to maintain the optimal traction efficiency.

Images