Data-driven trajectory tracking control system and method for unmanned mine truck
By constructing a multi-source dataset of the mining area and adopting a data-driven approach using TDNN and DGPR models, combined with longitudinal and lateral collaborative control, the accuracy and robustness issues of trajectory tracking control for unmanned mining trucks in open-pit mines were solved, achieving high-precision and stable trajectory tracking results and improving the safety and reliability of unmanned mining trucks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENHUA BEIDIAN SHENGLI ENERGY
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-29
AI Technical Summary
Existing unmanned mining truck trajectory tracking and control methods are not accurate enough and have poor robustness under complex working conditions in open-pit mines. The simplification of traditional physical models leads to large errors, and the model parameters rely on manual calibration, which is costly and inconsistent. They also fail to effectively cope with load changes and sudden changes in road surface adhesion coefficient.
A multi-source dataset of the mining area was constructed, and dynamic modeling was performed using a time-delay neural network (TDNN) and a deep Gaussian process regression (DGPR) model. Combined with longitudinal and lateral collaborative control, model uncertainty constraints were introduced, and a data-driven trajectory tracking control system adapted to the characteristics of mining trucks was designed.
It achieves high-precision and high-stability trajectory tracking under complex working conditions in open-pit mines, reduces the prediction errors of yaw rate and lateral velocity, improves the driving safety and reliability of unmanned mining trucks, and reduces labor costs and parameter calibration requirements.
Smart Images
Figure CN122111003A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of autonomous mining truck trajectory tracking technology, and particularly relates to a data-driven unmanned mining truck trajectory tracking control system and method. Background Technology
[0002] Open-pit mining environments are characterized by rugged terrain (frequent gravel roads and potholes), significant gradient variations, and drastic fluctuations in road surface adhesion coefficients (significant differences between dry / rainy / dusty weather). Furthermore, unmanned mining trucks experience dynamic load changes under both full-load and empty conditions, placing stringent demands on the accuracy and robustness of trajectory tracking control. Trajectory tracking control is a core component of autonomous driving for unmanned mining trucks, and its performance directly determines the truck's transportation efficiency, driving safety, and operational reliability.
[0003] Existing trajectory tracking control methods for unmanned mining trucks are mostly based on traditional physical modeling approaches. These methods simplify the vehicle dynamics model (such as two-degree-of-freedom or seven-degree-of-freedom models) and combine it with algorithms such as PID control and linear time-varying model predictive control (LTV-MPC) to achieve control. These methods rely on accurate physical parameters and mechanism analysis, and require manual calibration of a large number of model parameters to adapt to the mining environment.
[0004] Existing physical model-based trajectory tracking and control methods for unmanned mining trucks have the following problems: First, traditional physical models require extensive simplification of vehicle dynamics (such as linearizing tire characteristics and ignoring the effects of load transfer), which cannot accurately characterize the nonlinear and strongly coupled dynamic behavior under complex working conditions in mining areas, resulting in large trajectory tracking errors, especially in low-adhesion road surfaces and large-curvature turning conditions, which are prone to instability risks. Second, the model parameters rely on manual calibration. The parameters need to be readjusted for different road conditions (such as gravel roads and dirt roads) and load conditions (full load / empty load) in the mining area, which consumes a lot of manpower. Moreover, the calibration results are affected by subjective experience and have poor consistency. Third, existing data-driven modeling methods are mostly applied to vehicle state estimation, rarely deeply integrated with control algorithms, and have limited model generalization ability. They are prone to overfitting in mining areas where the mining area is not covered, and cannot guarantee control stability. Fourth, traditional control algorithms do not consider the uncertainty of model output. When the road surface adhesion coefficient changes abruptly or the load switches, unpredictable control errors are likely to occur due to model mismatch, resulting in insufficient safety.
[0005] The Chinese invention patent with publication number CN116819969A and invention title "A method for longitudinal and lateral coordinated control of unmanned mining truck path tracking in following mode" is based on model predictive control algorithm. It achieves trajectory tracking by establishing a simplified model of vehicle dynamics, but does not consider the impact of nonlinear factors such as changes in mining load and sudden changes in road surface adhesion coefficient on the accuracy of the model, resulting in insufficient control robustness.
[0006] The Chinese invention patent with publication number CN116819969A and invention title "A Robust Path Tracking Control Method for Unmanned Mining Trucks" improves the anti-interference capability to a certain extent, but does not solve the problem of inaccurate dynamic characteristic representation caused by the simplification of traditional physical models. The tracking error is large under extreme working conditions such as high speed, heavy load, and low-adhesion road surface. Summary of the Invention
[0007] To address the problems existing in the prior art, this invention proposes a data-driven unmanned mining truck trajectory tracking control system and method. By constructing a mining area-specific multi-source dataset, designing a data-driven dynamic model adapted to the characteristics of mining trucks, and implementing a collaborative control strategy, this invention solves the problems of inaccurate representation, complex parameter calibration, and insufficient robustness of traditional physical models, thereby achieving high-precision and high-stability trajectory tracking control under complex mining conditions.
[0008] The technical solution of the present invention is as follows: A data-driven trajectory tracking control system for unmanned mining trucks includes a mining area dataset construction module, a data-driven dynamics modeling module, and a trajectory tracking collaborative control module. The mining area dataset construction module is used to collect real working data from mining trucks and preprocess it to obtain a multi-source dataset of the mining area. The data-driven dynamics modeling module includes a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model, which are used to predict the dynamic characteristics of mining trucks. The trajectory tracking collaborative control module includes a longitudinal controller and a lateral controller. It combines feedforward feedback longitudinal control with model prediction lateral control, introduces model uncertainty constraints, and outputs the optimal control command.
[0009] Preferably, the Time Delay Neural Network (TDNN) model includes an input layer, a time delay layer, a hidden layer, and an output layer, wherein the input of the input layer is the longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions and road slope In addition to the road surface adhesion coefficient, the time delay layer introduces historical information from three time points, and the output layer outputs the first derivative of the yaw rate. With the first derivative of the transverse velocity The Softplus activation function and the mean absolute error loss function are used.
[0010] Preferably, the deep Gaussian process regression (DGPR) model includes an input layer, a flattened layer, a hidden layer, and an output layer, wherein the input of the input layer is the longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions and road slope The output of the output layer, along with the road surface adhesion coefficient, consists of the mean lateral velocity, the mean yaw rate, the variance of the lateral velocity, and the variance of the yaw rate. The activation functions used are the Softplus activation function and the modified ELU function, and the loss function is the Gaussian negative log-likelihood loss.
[0011] This invention also provides a data-driven trajectory tracking and control method for unmanned mining trucks, comprising the following steps: Step 1: The mining area dataset construction module collects real working data of mining trucks under various working conditions, preprocesses the real working data, and obtains a multi-source dataset of the mining area. Step 2: The data-driven dynamics modeling module establishes a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model, and trains them using the multi-source dataset from the mining area; Step 3: The trajectory tracking collaborative control module is designed with a longitudinal controller, a lateral controller, control constraints, and a control switching strategy; Step 4: The trajectory tracking and collaborative control module receives the desired trajectory output from the high-precision map, combines it with the real-time feedback data of the mining truck status from the sensors, and calculates the optimal front wheel steering angle. With longitudinal force Control commands are output to the steering and power braking systems for execution; sensors update the mining truck status data in real time and feed it back to the data-driven dynamics modeling module to dynamically update the model parameters.
[0012] Preferably, the various working conditions include dry gravel road, rain-soaked dirt road, fully loaded uphill driving, unloaded downhill driving, ISO double lane change and figure-eight curve driving.
[0013] Preferably, the actual working data includes longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load condition, road slope and road surface adhesion coefficient.
[0014] Preferably, the preprocessing in step 1 includes: using a second-order Butterworth low-pass filter to filter out sensor vibration noise, analyzing data distribution based on the stable envelope boundary of the mining truck, and supplementing missing state space data using a three-degree-of-freedom physical model to ensure the integrity and effectiveness of the dataset.
[0015] Preferably, the longitudinal controller is based on a feedforward feedback control method, with the feedforward component adjusting according to the mine truck load status and road slope. Calculate the desired longitudinal force The feedback section dynamically adjusts the feedback gain based on the relationship between tire slip ratio and sideslip angle; the lateral controller is based on the model predictive control (MPC) algorithm, embedding a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model respectively, with the objective function... It includes lateral deviation, heading deviation, control increment, and model uncertainty constraints.
[0016] Preferably, the longitudinal force The expression is:
[0017] in, For feedforward longitudinal force, For the quality of mining cards, For speed feedback force, For speed feedback gain, For the desired longitudinal velocity, The longitudinal velocity is measured in real time by a sensor. The feedback force is based on tire characteristics; The objective function The expression is:
[0018] in, To predict the time domain, To control the time domain, e is a constant with the base of the natural logarithm. For the heading angle increment, For the front wheel steering angle increment, These are lateral deviation, heading deviation, and control increment weights, respectively. These are the variances of lateral velocity and yaw rate, respectively.
[0019] Preferably, the control switching strategy is as follows: when the deep Gaussian process regression (DGPR) model outputs the lateral velocity variance... or the rate of change in road surface adhesion coefficient If the condition is met, switch to DGPR-MPC mode; otherwise, use TDNN-MPC mode.
[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The data-driven unmanned mining truck trajectory tracking control system and method proposed in this invention constructs a multi-source dataset for the mining area, covering typical working conditions such as full load / empty load, different adhesion coefficients, and slope, providing comprehensive support for data-driven modeling and avoiding the representation bias caused by the reliance on simplification assumptions in traditional models.
[0021] 2. The data-driven unmanned mining truck trajectory tracking control system and method proposed in this invention adopts a dual-model architecture of TDNN and DGPR. The TDNN model accurately captures the temporal dynamic characteristics of the mining truck, while the DGPR model realizes uncertainty prediction. This dual guarantee ensures the modeling accuracy under complex working conditions. Compared with traditional physical models, the prediction errors of yaw rate and lateral velocity are significantly reduced.
[0022] 3. The data-driven unmanned mining truck trajectory tracking control system and method proposed in this invention adopts a longitudinal and lateral coordinated control strategy, combining model predictive control and uncertainty constraints. Under double lane change and large curvature turning conditions, the peak lateral error is reduced compared with nonlinear model predictive control (NMPC), and the peak lateral error is significantly reduced under low-adhesion road surface conditions; the tracking accuracy is improved.
[0023] 4. The data-driven unmanned mining truck trajectory tracking control system and method proposed in this invention do not require manual calibration of model parameters. The model can adapt to different road conditions and load states in the mining area, reducing labor costs. It also has strong generalization ability and can maintain stable control even in uncovered working conditions.
[0024] 5. The data-driven unmanned mining truck trajectory tracking control system and method proposed in this invention have a dynamic switching mechanism for control modes to adapt to complex working conditions in mining areas, effectively avoiding control instability caused by sudden changes in road surface adhesion coefficient and load switching, and improving the driving safety and reliability of unmanned mining trucks. Attached Figure Description
[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly introduced below. The features and advantages of the present invention can be more clearly understood by referring to the accompanying drawings. The accompanying drawings are schematic and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 This is a schematic diagram of the data-driven unmanned mining truck trajectory tracking control system of the present invention.
[0027] Figure 2This is a flowchart of the construction process of a multi-source dataset for mining areas according to the present invention.
[0028] Figure 3 This is a diagram of the data-driven dynamics model architecture of the present invention.
[0029] Figure 4 This is a flowchart of the data-driven trajectory tracking and control method for unmanned mining trucks according to the present invention. Detailed Implementation
[0030] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.
[0031] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.
[0032] The data-driven trajectory tracking control system for unmanned mining trucks of this invention consists of three parts: a mining area dataset construction module, a data-driven dynamics modeling module, and a trajectory tracking collaborative control module. Figure 1 As shown.
[0033] The mining area dataset construction module is responsible for collecting and preprocessing mine truck dynamics data to form a complete dataset; the data-driven dynamics modeling module trains a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model based on the dataset to achieve accurate prediction of mine truck dynamics characteristics; the trajectory tracking collaborative control module combines feedforward feedback longitudinal control and model prediction lateral control, introduces model uncertainty constraints, and outputs optimal control commands.
[0034] Specifically, trajectory tracking control methods include: Step 1: Data Acquisition and Preprocessing like Figure 2 First, real-world working data was collected from mining trucks in open-pit mines under various operating conditions. The collected data included the longitudinal velocity of the mining trucks. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions, tire force, and road slope And road surface adhesion coefficient, etc. The working conditions covered include dry gravel roads, dirt roads after rain, fully loaded uphill driving, unloaded downhill driving, ISO double lane change and figure-eight curve driving, etc.
[0035] Then, a second-order Butterworth low-pass filter is used to filter out sensor vibration noise. The data distribution is analyzed based on the vehicle stability envelope boundary, and the missing state space data is supplemented using a three-degree-of-freedom physical model to ensure the integrity and effectiveness of the dataset.
[0036] The second step is to construct the dynamic model: like Figure 3 As shown, the Time Delay Neural Network (TDNN) model is constructed by selecting the longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions and road slope The road surface adhesion coefficient is used as the input feature, and the first derivative of the yaw rate and the first derivative of the lateral velocity are used as the output. Historical information at three time points is introduced to improve the temporal modeling capability. The Softplus activation function and the mean absolute error (MAE) loss function are used. The model is trained by the Adam optimization algorithm, and the network structure is determined after hyperparameter optimization.
[0037] Deep Gaussian Process Regression (DGPR) Model Construction: Integrating the advantages of traditional machine learning and deep learning, an output uncertainty prediction branch is added to the model structure. The input features are consistent with the TDNN model, and the output includes the mean (mean of lateral velocity, mean of yaw rate) and variance (variance of lateral velocity and yaw rate) of the dynamic state at the next time step. A Gaussian negative log-likelihood loss function is used to accelerate the training speed, and the activation function adopts a combination of Softplus and modified ELU functions to ensure that the variance output is positive.
[0038] Step 3, Controller Design: Longitudinal controller design: Based on the feedforward feedback control method, the feedforward part calculates the desired longitudinal force according to the mine truck load and slope information, and the feedback part dynamically adjusts the feedback gain through the relationship between tire slip ratio and sideslip angle to avoid the problems of overheating during braking on long downhill slopes and insufficient acceleration during climbing. The formula is as follows:
[0039] in, For feedforward longitudinal force, For the quality of mining cards, For speed feedback force, For speed feedback gain, For the desired longitudinal velocity, The longitudinal velocity is measured in real time by a sensor. The feedback force is based on tire characteristics; Lateral controller design: Based on the Model Predictive Control (MPC) algorithm, TDNN and DGPR models are embedded respectively. The objective function includes lateral deviation, heading deviation, control increment, and model uncertainty constraints, as shown in the following formula:
[0040] in, To predict the time domain, To control the time domain, e is a constant with the base of the natural logarithm. For the heading angle increment, For the front wheel steering angle increment, These are lateral deviation, heading deviation, and control increment weights, respectively. These are the variances of lateral velocity and yaw rate, respectively.
[0041] Control constraint settings: Considering the characteristics of mining truck wheel track width and center of gravity height, tighten the center of gravity sideslip angle constraint. The front wheel steering angle and the range of steering angle increments are limited, and the longitudinal force constraint of the tire introduces a dynamic adjustment mechanism for the road adhesion coefficient.
[0042] Fourthly, the trajectory tracking and collaborative control module dynamically switches control modes according to the working conditions of the mining area, such as... Figure 4 As shown: Under high-adhesion road surface and constant speed driving conditions, the TDNN-MPC control mode is adopted to prioritize tracking accuracy; under low-adhesion road surface, load switching, and variable speed driving conditions, the DGPR-MPC control mode is adopted to improve robustness through uncertainty constraints. The control mode switching threshold is adaptively determined based on the road surface adhesion coefficient and the model prediction variance.
[0043] Fifth, after the mining truck completes its transportation task, it uploads the operating data to the data-driven dynamics modeling module. The data-driven dynamics modeling module receives the data, preprocesses it, updates the network, and sends the updated network to the mining truck. The truck receives the new network parameters and updates the predictive network used by the controller to achieve closed-loop control.
[0044] Example 1 Step 1: The mining area dataset construction module performs data acquisition and preprocessing. (1) Real Data Set Acquisition: Real data set acquisition involves using a data acquisition platform to receive real working data of open-pit mine trucks under various operating conditions. The data acquired by the platform includes longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions, tire forces (longitudinal and lateral forces), and road slope And the road surface adhesion coefficient. The working conditions covered include dry gravel road (μ=0.7), rain-soaked dirt road (μ=0.4), fully loaded uphill (10° gradient), unloaded downhill (12° gradient), ISO double lane change and figure-eight curve driving, etc. A total of 3,000 sets of valid real vehicle data were obtained in this step.
[0045] (2) Data preprocessing: A second-order Butterworth low-pass filter (cutoff frequency 10Hz) was used to filter out high-frequency vibration noise of the suspension. The vehicle stability envelope boundary was calculated based on the maximum value of the rear wheel slip angle. The missing state space region in the simulation data was supplemented and divided into training set, validation set and test set in an 8:1:1 ratio.
[0046] Step 2, the data-driven dynamics modeling module trains a dual model: (1) TDNN model training: Input feature selection: vertical velocity lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load status (full load = 1, no load = 0) and road slope And the road surface adhesion coefficient, the output is the first derivative of the yaw rate. With the first derivative of the transverse velocity The network structure is set as input layer → time delay layer (3 historical time points) → hidden layer (2 layers, 128 neurons per layer) → output layer; the activation function is Softplus, the loss function is MAE, the optimization algorithm is Adam, the initial learning rate is 0.0001, the batch size is 1000, and it converges after 500 training rounds. The prediction error on the validation set is 0.0016.
[0047] (2) DGPR model training: The input features are the same as those of the TDNN model. The output layer contains 4 neurons, which output the mean lateral velocity, the mean lateral angular velocity, and the variance of lateral velocity, respectively. and yaw rate variance The network structure is: input layer → flattened layer → hidden layer (2 layers, Softplus activation) → output layer (corrected ELU activation to ensure positive variance); the loss function is Gaussian negative log-likelihood loss, the training process loads TDNN model pre-trained weights, converges after 300 iterations, and the model uncertainty prediction error is less than 5%.
[0048] Step 3: The trajectory tracking cooperative control module designs and solves the control strategy: (1) Longitudinal controller parameter settings: The feedforward gain is dynamically adjusted according to the ore car mass and slope, and the speed feedback gain is adjusted accordingly. Longitudinal slip rate gain Side slip angle gain Braking response time τ=0.2s, safe stopping distance .
[0049] (2) Horizontal controller parameter settings: prediction time domain Control time domain Sampling step size 20ms; front wheel steering angle constraint Incremental corner constraint Weighting coefficients ,
[0050] (3) Control mode switching threshold: When the DGPR model output variance or the rate of change in road surface adhesion coefficient If the condition is met, switch to DGPR-MPC mode; otherwise, use TDNN-MPC mode.
[0051] Step 4, Control command output and closed-loop feedback: The trajectory tracking and collaborative control module receives the desired trajectory (including longitude, latitude, elevation, and curvature information) output from the high-precision map, combines it with the real-time feedback of the mining truck's status data from the sensors, and solves for the optimal front wheel steering angle and longitudinal force command through the above control strategy, which is then output to the steering system and the power braking system. The sensors collect the actual driving status of the mining truck in real time and feed it back to the data-driven modeling module to dynamically update the model parameters, ensuring control accuracy and stability.
[0052] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0053] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.
[0054] In this invention, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The term "multiple" refers to two or more unless otherwise expressly defined.
[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A data-driven unmanned mining truck trajectory tracking control system, characterized in that, It includes a mining area dataset construction module, a data-driven dynamics modeling module, and a trajectory tracking and collaborative control module; The mining area dataset construction module is used to collect real working data from mining trucks and preprocess it to obtain a multi-source dataset of the mining area. The data-driven dynamics modeling module includes a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model, which are used to predict the dynamic characteristics of mining trucks. The trajectory tracking collaborative control module includes a longitudinal controller and a lateral controller. It combines feedforward feedback longitudinal control with model prediction lateral control, introduces model uncertainty constraints, and outputs the optimal control command.
2. The unmanned mining truck trajectory tracking control system according to claim 1, characterized in that, The Time Delay Neural Network (TDNN) model includes an input layer, a time delay layer, a hidden layer, and an output layer, wherein the input of the input layer is the longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions and road slope In addition to the road surface adhesion coefficient, the time delay layer introduces historical information from three time points, and the output layer outputs the first derivative of the yaw rate. With the first derivative of the transverse velocity The Softplus activation function and the mean absolute error loss function are used.
3. The unmanned mining truck trajectory tracking control system according to claim 1, characterized in that, The Deep Gaussian Process Regression (DGPR) model comprises an input layer, a Flattened layer, hidden layers, and an output layer, wherein the input to the input layer is the longitudinal velocity. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load conditions and road slope The output of the output layer, along with the road surface adhesion coefficient, consists of the mean lateral velocity, the mean yaw rate, the variance of the lateral velocity, and the variance of the yaw rate. The activation functions used are the Softplus activation function and the modified ELU function, and the loss function is the Gaussian negative log-likelihood loss.
4. A control method based on the unmanned mining truck trajectory tracking control system according to any one of claims 1-3, characterized in that, Includes the following steps: Step 1: The mining area dataset construction module collects real working data of mining trucks under various working conditions, preprocesses the real working data, and obtains a multi-source dataset of the mining area. Step 2: The data-driven dynamics modeling module establishes a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model, and trains them using the multi-source dataset from the mining area; Step 3: The trajectory tracking collaborative control module is designed with a longitudinal controller, a lateral controller, control constraints, and a control switching strategy; Step 4: The trajectory tracking and collaborative control module receives the desired trajectory output from the high-precision map, combines it with the real-time feedback data of the mining truck status from the sensors, and calculates the optimal front wheel steering angle. With longitudinal force Control commands are output to the steering and power braking systems for execution; sensors update the mining truck status data in real time and feed it back to the data-driven dynamics modeling module to dynamically update the model parameters.
5. The control method according to claim 4, characterized in that, The various operating conditions include dry gravel road, rain-soaked dirt road, fully loaded uphill driving, unloaded downhill driving, ISO double lane change and figure-eight curve driving.
6. The control method according to claim 5, characterized in that, The actual working data includes longitudinal speed. lateral velocity yaw rate Front wheel steering angle Longitudinal acceleration Load condition, road slope and road surface adhesion coefficient.
7. The control method according to claim 6, characterized in that, The preprocessing in step 1 includes: using a second-order Butterworth low-pass filter to filter out sensor vibration noise, analyzing data distribution based on the stable envelope boundary of the mining truck, and using a three-degree-of-freedom physical model to supplement missing state space data to ensure the integrity and validity of the dataset.
8. The control method according to claim 7, characterized in that, The longitudinal controller is based on a feedforward feedback control method, with the feedforward component adjusting according to the mine truck load status and road slope. Calculate the desired longitudinal force The feedback section dynamically adjusts the feedback gain based on the relationship between tire slip ratio and sideslip angle; the lateral controller is based on the model predictive control (MPC) algorithm, embedding a time-delay neural network (TDNN) model and a deep Gaussian process regression (DGPR) model respectively, with the objective function... It includes lateral deviation, heading deviation, control increment, and model uncertainty constraints.
9. The control method according to claim 8, characterized in that, The longitudinal force The expression is: in, For feedforward longitudinal force, For the quality of mining cards, For speed feedback force, For speed feedback gain, For the desired longitudinal velocity, The longitudinal velocity is measured in real time by a sensor. The feedback force is based on tire characteristics; The objective function The expression is: in, To predict the time domain, To control the time domain, e is a constant with the base of the natural logarithm. For the heading angle increment, For the front wheel steering angle increment, These are lateral deviation, heading deviation, and control increment weights, respectively. These are the variances of lateral velocity and yaw rate, respectively.
10. The control method according to claim 9, characterized in that, The control switching strategy is as follows: when the Deep Gaussian Process Regression (DGPR) model outputs the lateral velocity variance... or the rate of change in road surface adhesion coefficient If the condition is met, switch to DGPR-MPC mode; otherwise, use TDNN-MPC mode.