A trackless rubber-tyred vehicle dynamic evolution scene adaptive control system and method

Abstract

The application discloses a trackless rubber-tyred vehicle dynamic evolution scene self-adaptive control system and method, which collects surrounding environment information first, constructs a dynamic and static space-time topology graph of an underground roadway, and models and analyzes a dynamic evolution and an unstructured underground roadway environment; a multi-sensor perception information value evaluation system suitable for the underground roadway environment is constructed to evaluate the value of data obtained by each sensor; a long-acting path planning scheme of the trackless rubber-tyred vehicle is established to realize self-adaptive long-acting decision of the trackless rubber-tyred vehicle in a dynamic scene; a trajectory prediction scheme is established in combination with historical trajectory information, a pre-control command is sent to the trackless rubber-tyred vehicle according to the trajectory prediction information; and a trajectory control scheme is established to enable the trackless rubber-tyred vehicle to track the planned trajectory in real time; through the above process, the working efficiency of the trackless rubber-tyred vehicle in the coal mine is improved, and the safety of the trackless rubber-tyred vehicle operation is ensured.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation technology in coal mines, specifically an adaptive control system and method for the dynamic evolution of trackless rubber-tired vehicles. Background Technology

[0002] Trackless rubber-tired vehicles are rubber-tired transport vehicles used underground, running on the floor of mine tunnels. They are versatile, flexible, safe, adaptable, and easy to load and unload, reducing transshipment steps. They significantly improve auxiliary transportation and overall mine productivity. However, the complex and variable underground environment of coal mines leads to frequent accidents involving trackless rubber-tired vehicles. Therefore, realizing intelligent and unmanned operation of these vehicles can reduce the accident rate and has significant practical implications.

[0003] The underground environment of coal mines is characterized by confined space and dynamic, complex evolution. Visibility is significantly reduced due to water mist, dust, and insufficient lighting. Furthermore, the narrow and low tunnels limit the operating space of trackless rubber-tired vehicles, leading to frequent safety accidents. Currently, due to the diverse interference factors and dynamically evolving environment of trackless rubber-tired vehicles in coal mines, trajectory planning in their trajectory control methods is lagging, and this lag cannot provide a high degree of guarantee for the safe operation of these vehicles.

[0004] Based on this, the research direction of this invention is to provide a new control method for trackless rubber-tired vehicles that can adjust the trajectory planning in real time according to the dynamic evolution environment faced by the trackless rubber-tired vehicles in underground coal mines, and control the trackless rubber-tired vehicles to adapt to driving, thereby not only improving the working efficiency of trackless rubber-tired vehicles in underground coal mines, but also ensuring the safety of trackless rubber-tired vehicle operation. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention provides an adaptive control system and method for the dynamic evolution of trackless rubber-tired vehicles, which can effectively solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a dynamic evolution scene adaptive control system for trackless rubber-wheeled vehicles, including a dynamic scene evolution characteristic analysis module, a complex scene perception information evaluation module, a trajectory adaptive long-term decision-making module, and a trajectory tracking control module.

[0007] The dynamic scene evolution characteristic analysis module is used to construct a dynamic and static spatiotemporal topology map of underground roadways, model and analyze the dynamically evolving and unstructured underground roadway environment, and obtain comprehensive scene information.

[0008] The complex scene perception information evaluation module, based on information gain and spatiotemporal correlation analysis, constructs a multi-sensor perception information value evaluation system suitable for underground roadway environments, and evaluates the value of data obtained by each sensor.

[0009] The trajectory adaptive long-term decision module is used to establish a long-term path planning scheme for trackless rubber-tired vehicles based on the DSAC (Distributional SoftActor-Critic) algorithm, so as to realize adaptive long-term decision-making in dynamic scenarios of trackless rubber-tired vehicles.

[0010] The trajectory tracking control module contains a trajectory prediction control model and a trajectory tracking control model. The trajectory prediction control model combines historical trajectory information to establish a trajectory prediction scheme based on the Transformer method and issues pre-control commands to the trackless rubber-tired vehicle based on the trajectory prediction information. The trajectory tracking control model establishes a trajectory control scheme based on SMC (sliding mode control) to enable the trackless rubber-tired vehicle to track the planned trajectory in real time.

[0011] Furthermore, the sensors include lidar, millimeter-wave radar, ultrasonic radar, visible light cameras, and infrared cameras.

[0012] The control method of the adaptive control system for the dynamic evolution scenario of the trackless rubber-tired vehicle mentioned above includes the following steps:

[0013] Step 1: Use multiple sensors to collect information about the environment around the trackless rubber-tired vehicle, and construct a dynamic and static spatiotemporal topology map of the underground roadway. Construct a dynamic spatiotemporal topology map for the dynamic scene evolution during the operation of the trackless rubber-tired vehicle in the coal mine, and construct a static spatiotemporal topology map for the long-term evolution of the scene itself.

[0014] Step 2: Based on the gain of the collected information, construct a multi-sensor perception information value assessment system suitable for the working environment of trackless rubber-tired vehicles in underground roadways;

[0015] Step 3: Establish a long-term adaptive path planning scheme for trackless rubber-tired vehicles based on the DSAC-T (Distributional Soft Actor-Critic with Three Refinements) algorithm to realize adaptive path planning and obstacle avoidance of trackless rubber-tired vehicles in complex and dynamic evolution scenarios in coal mines;

[0016] Step 4: Combine the historical trajectory information of the trackless rubber-tired vehicle to establish a trajectory prediction scheme for the trackless rubber-tired vehicle based on the Transformer algorithm. Issue pre-control commands to the trackless rubber-tired vehicle based on the trajectory prediction information to improve the control flexibility and adaptability of the trackless rubber-tired vehicle in complex dynamic scenarios.

[0017] Step 5: Establish a trajectory control scheme based on SMC (sliding mode control) to achieve trajectory tracking of the trackless rubber-tired vehicle;

[0018] Step Six: Calculate the error of the pre-control command in Step Four and the error of trajectory tracking in Step Five. This is used to correct the long-term adaptive path planning scheme of the trackless rubber-tired vehicle in Step Three, enhance the trajectory adjustment of the trackless rubber-tired vehicle in complex dynamic evolution scenarios, and ensure the safety and driving efficiency of the trackless rubber-tired vehicle in complex dynamic evolution scenarios.

[0019] Furthermore, the scenario state model used in step one is as follows:

[0020] G t =(V t E t ,X t )

[0021] Among them, V t These represent environmental state nodes within a tunnel, signifying key areas within the environment, such as the main tunnel passage, branch tunnels, and equipment areas; E t Indicates spatiotemporal correlation, X t This represents the feature vector of each node.

[0022] Furthermore, step two specifically involves: acquiring the sensor information collected by the sensor, and using the Kullback-Leibler divergence to measure the relative entropy difference between the historical sensor information distribution and the collected sensor information distribution, as shown in the following formula.

[0023]

[0024] Where P(x) represents the historical sensor information distribution, Q(x) represents the currently acquired sensor information distribution, and D... KL The larger the value, the greater the difference between the collected data and historical data, indicating that the currently collected sensor information is more valuable.

[0025] Furthermore, step three specifically includes:

[0026] Initialize strategy π φ Distributed value network Z θ (s,a), target network θ'←θ; experience pool is Φ, initial velocity is α, entropy target is H, quantile parameter is τ. i According to the current strategy a:π φ (·|s) Executes the action, gains experience (s,a,r,s',d) and puts it into the experience pool; samples Φ=(s,a,r,s',d) from the experience pool Φ and calculates the target distribution.

[0027] Update the distributed value network Z θ :

[0028] Where d p For distribution metrics; update policy network π φ :

[0029] Update speed α:

[0030] Update the target network:

[0031] θ'←τθ+(1-τ)θ'

[0032] φ'←τφ+(1-τ)φ'

[0033] Training ends when the target network converges.

[0034] Where s represents the current state of the trackless rubber-wheeled vehicle, obtained by collecting environmental information around the vehicle and then processing it through a state encoder; a represents the output speed information of the trackless rubber-wheeled vehicle; r is the reward function, which consists of the time it takes for the trackless rubber-wheeled vehicle to reach the target point, whether a collision occurs, and whether the path is smooth. The shorter the time to reach the target point, the fewer collisions with obstacles, and the smoother the path, the greater the reward, and vice versa. The trackless rubber-wheeled vehicle selects the operation with the largest Q value based on the real-time environmental information fed back by the environmental assessment module, and obtains the corresponding α value, which is the real-time speed information of the trackless rubber-wheeled vehicle.

[0035] Furthermore, step five specifically includes:

[0036] ① Establish the system dynamic equations

[0037]

[0038] Where, x, Let P(x,t) represent position and velocity, b(x,t) represent control gain, and d(t) represent disturbance Pd(t)P≤D;

[0039] ② Definition of trajectory tracking error

[0040] Tracking error:

[0041] ③ Slipform surface design

[0042] Design the sliding surface s to constrain error dynamics: Where λ is the control convergence rate;

[0043] ④ Calculate the equivalent control and design the switching control, as well as the total control input;

[0044]

[0045] u(t)=u eq (t)+u sw (t)

[0046] Where K is the gain for suppressing disturbances, K≥1.5D, and Φ is the boundary layer thickness;

[0047] ⑤ Stability conditions: Convergence conditions

[0048] Compared with existing technologies, this invention first collects surrounding environmental information to construct a dynamic and static spatiotemporal topology map of underground roadways, and models and analyzes the dynamically evolving and unstructured underground roadway environment to obtain comprehensive scene information. Based on information gain and spatiotemporal correlation analysis, it constructs a multi-sensor perception information value assessment system suitable for underground roadway environments to evaluate the value of data obtained by each sensor. It establishes a long-term path planning scheme for trackless rubber-tired vehicles based on the DSAC algorithm to achieve adaptive long-term decision-making for trackless rubber-tired vehicles in dynamic scenarios. Combining historical trajectory information, it establishes a trajectory prediction scheme based on the Transformer method and issues pre-control commands to the trackless rubber-tired vehicles based on the trajectory prediction information. It establishes a trajectory control scheme based on SMC (sliding mode control) to enable the trackless rubber-tired vehicles to track the planned trajectory in real time. Through the above process, this invention enables the trackless rubber-tired vehicles to adjust trajectory planning in real time according to the dynamically evolving environment faced by the trackless rubber-tired vehicles in coal mines and control the trackless rubber-tired vehicles to drive adaptively, thereby not only improving the working efficiency of trackless rubber-tired vehicles in coal mines, but also ensuring the safety of trackless rubber-tired vehicle operation. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0050] The present invention will be further described below.

[0051] An adaptive control system for dynamic evolution scenarios of a trackless rubber-tired vehicle includes a dynamic scenario evolution characteristic analysis module, a complex scenario perception information evaluation module, a trajectory adaptive long-term decision-making module, and a trajectory tracking control module.

[0052] The dynamic scene evolution characteristic analysis module is used to construct a dynamic and static spatiotemporal topology map of underground roadways, model and analyze the dynamically evolving and unstructured underground roadway environment, and obtain comprehensive scene information.

[0053] The complex scene perception information evaluation module, based on information gain and spatiotemporal correlation analysis, constructs a multi-sensor perception information value evaluation system suitable for underground roadway environments, and evaluates the value of data obtained by each sensor; the sensors include lidar, millimeter-wave radar, ultrasonic radar, visible light camera and infrared camera.

[0054] The trajectory adaptive long-term decision module is used to establish a long-term path planning scheme for trackless rubber-tired vehicles based on the DSAC (Distributional SoftActor-Critic) algorithm, so as to realize adaptive long-term decision-making in dynamic scenarios of trackless rubber-tired vehicles.

[0055] The trajectory tracking control module contains a trajectory prediction control model and a trajectory tracking control model. The trajectory prediction control model combines historical trajectory information to establish a trajectory prediction scheme based on the Transformer method and issues pre-control commands to the trackless rubber-tired vehicle based on the trajectory prediction information. The trajectory tracking control model establishes a trajectory control scheme based on SMC (sliding mode control) to enable the trackless rubber-tired vehicle to track the planned trajectory in real time.

[0056] like Figure 1 As shown, the control method of the adaptive control system for the dynamic evolution scenario of the trackless rubber-tired vehicle includes the following steps:

[0057] Step 1: Use multiple sensors to collect information about the environment surrounding the trackless rubber-tired vehicle, and construct dynamic and static spatiotemporal topology maps of the underground roadway. Construct a dynamic spatiotemporal topology map for the dynamic scene evolution during the operation of the trackless rubber-tired vehicle in the coal mine, and a static spatiotemporal topology map for the long-term evolution of the scene itself. The scene state model used is as follows:

[0058] G t =(V t E t ,X t )

[0059] Among them, V t These represent environmental state nodes within a tunnel, signifying key areas within the environment, such as the main tunnel passage, branch tunnels, and equipment areas; E t Indicates spatiotemporal correlation, X t This represents the feature vector of each node.

[0060] Step 2: Based on the gain of the collected information, construct a multi-sensor perception information value assessment system suitable for the underground roadway environment of the trackless rubber-tired vehicle. Specifically, acquire the sensor information collected by the sensors, and use the Kullback-Leibler divergence to measure the relative entropy difference between the historical sensor information distribution and the collected sensor information distribution. The formula is as follows:

[0061]

[0062] Where P(x) represents the historical sensor information distribution, Q(x) represents the currently acquired sensor information distribution, and D... KL The larger the value, the greater the difference between the collected data and historical data, indicating that the currently collected sensor information is more valuable.

[0063] Step 3: Establish a long-term adaptive path planning scheme for trackless rubber-tired vehicles based on the DSAC-T (Distributive Soft Actor-Critic with Three Refinements) algorithm, realizing adaptive path planning and obstacle avoidance for trackless rubber-tired vehicles in complex and dynamically evolving scenarios in coal mines. Specifically:

[0064] Initialize strategy π φ Distributed value network Z θ (s,a), target network θ'←θ; experience pool is Φ, initial velocity is α, entropy target is H, quantile parameter is τ. i According to the current strategy a:π φ (·|s) Executes the action, gains experience (s,a,r,s',d) and puts it into the experience pool; samples Φ=(s,a,r,s',d) from the experience pool Φ and calculates the target distribution.

[0065] Update the distributed value network Z θ :

[0066] Where d p For distribution metrics; update policy network π φ :

[0067] Update speed α:

[0068] Update the target network:

[0069] θ'←τθ+(1-τ)θ'

[0070] φ'←τφ+(1-τ)φ'

[0071] Training ends when the target network converges.

[0072] Where s represents the current state of the trackless rubber-wheeled vehicle, obtained by collecting environmental information around the vehicle and then processing it through a state encoder; a represents the output speed information of the trackless rubber-wheeled vehicle; r is the reward function, which consists of the time it takes for the trackless rubber-wheeled vehicle to reach the target point, whether a collision occurs, and whether the path is smooth. The shorter the time to reach the target point, the fewer collisions with obstacles, and the smoother the path, the greater the reward, and vice versa. The trackless rubber-wheeled vehicle selects the operation with the largest Q value based on the real-time environmental information fed back by the environmental assessment module, and obtains the corresponding α value, which is the real-time speed information of the trackless rubber-wheeled vehicle.

[0073] Step 4: Combining historical trajectory information of the trackless rubber-tired vehicle, establish a trajectory prediction scheme based on the Transformer algorithm. Issue pre-control commands to the trackless rubber-tired vehicle based on the trajectory prediction information to improve its control flexibility and adaptability in complex dynamic scenarios. The specific formula is as follows:

[0074] Y = Transformer(X)

[0075] Where X = (x1, x2, ... x n Y represents the historical trajectory information of the trackless rubber-tired vehicle; Y represents the predicted trajectory of the trackless rubber-tired vehicle.

[0076] Step 5: Establish a trajectory control scheme based on SMC (sliding mode control) to achieve trajectory tracking of the trackless rubber-tired vehicle, specifically as follows:

[0077] ① Establish the system dynamic equations

[0078]

[0079] Where, x, Let P(x,t) represent position and velocity, b(x,t) represent control gain, and d(t) represent disturbance Pd(t)P≤D;

[0080] ② Definition of trajectory tracking error

[0081] Tracking error:

[0082] ③ Slipform surface design

[0083] Design the sliding surface s to constrain error dynamics: Where λ is the control convergence rate, which is 0.6 in this embodiment;

[0084] ④ Calculate the equivalent control and design the switching control, as well as the total control input;

[0085]

[0086] u(t)=u eq (t)+usw (t)

[0087] Where K is the gain for suppressing disturbances, K≥1.5D, and Φ is the boundary layer thickness, which is taken as 0.3 in this embodiment;

[0088] ⑤ Stability conditions: Convergence conditions

[0089] Step Six: Calculate the errors of the pre-control command in Step Four and the trajectory tracking in Step Five. These errors are used to correct the long-term adaptive path planning scheme for the trackless rubber-tired vehicle in Step Three, enhancing the vehicle's trajectory adjustment in complex dynamic evolution scenarios and ensuring its safety and efficiency. The formula used in this step is:

[0090] r1 = Y 真 -Y, r = r yuan +r1+r2;

[0091] Where r1 is the trajectory prediction error reward function, r2 is the trajectory tracking error reward function, and r yuan The reward function consists of the time it takes for the trackless rubber-wheeled vehicle to reach the target point, whether a collision occurs, and whether the path is smooth. The above formula forms an auxiliary task to modify the long-term adaptive path planning scheme of the trackless rubber-wheeled vehicle.

[0092] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A control method for an adaptive control system of a trackless rubber-tired vehicle in a dynamic evolution scenario, characterized in that, The control system includes a dynamic scene evolution characteristic analysis module, a complex scene perception information evaluation module, a trajectory adaptive long-term decision-making module, and a trajectory tracking control module. The dynamic scene evolution characteristic analysis module is used to construct a dynamic and static spatiotemporal topology map of the underground roadway, modeling and analyzing the dynamically evolving and unstructured underground roadway environment to obtain comprehensive scene information. The complex scene perception information evaluation module, based on information gain and spatiotemporal correlation analysis, constructs a multi-sensor perception information value assessment system suitable for the underground roadway environment, evaluating the value of data obtained by each sensor. The trajectory adaptive long-term decision-making module is used to establish a long-term path planning scheme for the trackless rubber-tired vehicle, realizing adaptive long-term decision-making under dynamic scenarios. The trajectory tracking control module internally includes a trajectory prediction control model and a trajectory tracking control model. The trajectory prediction control model combines historical trajectory information to establish a trajectory prediction scheme and issues pre-control commands to the trackless rubber-tired vehicle based on the trajectory prediction information. A trajectory tracking control model is established to create a trajectory control scheme for the trackless rubber-tired vehicle to track the planned trajectory in real time. This control method includes the following steps: Step 1: Use multiple sensors to collect information about the environment around the trackless rubber-tired vehicle, and construct a dynamic and static spatiotemporal topology map of the underground roadway. Construct a dynamic spatiotemporal topology map for the dynamic scene evolution during the operation of the trackless rubber-tired vehicle in the coal mine, and construct a static spatiotemporal topology map for the long-term evolution of the scene itself. Step 2: Based on the gain of the collected information, construct a multi-sensor perception information value assessment system suitable for the working environment of trackless rubber-tired vehicles in underground roadways; Step 3: Establish a long-term adaptive path planning scheme for trackless rubber-tired vehicles based on the DSAC-T algorithm to realize adaptive path planning and obstacle avoidance of trackless rubber-tired vehicles in complex and dynamic evolution scenarios in coal mines; Step 4: Combining the historical trajectory information of the trackless rubber-tired vehicle, establish a trajectory prediction scheme for the trackless rubber-tired vehicle based on the Transformer algorithm. Issue pre-control commands to the trackless rubber-tired vehicle based on the trajectory prediction information to improve the control flexibility and adaptability of the trackless rubber-tired vehicle in complex dynamic scenarios. Step 5: Establish a trajectory control scheme based on SMC to achieve trajectory tracking of the trackless rubber-wheeled vehicle; Step Six: Calculate the error of the pre-control command in Step Four and the error of trajectory tracking in Step Five. This is used to correct the long-term adaptive path planning scheme of the trackless rubber-tired vehicle in Step Three, enhance the trajectory adjustment of the trackless rubber-tired vehicle in complex dynamic evolution scenarios, and ensure the safety and driving efficiency of the trackless rubber-tired vehicle in complex dynamic evolution scenarios.

2. The control method according to claim 1, characterized in that, The sensors include lidar, millimeter-wave radar, ultrasonic radar, visible light cameras, and infrared cameras.

3. The control method according to claim 1, characterized in that, The scenario state model used in step one is: in, These represent environmental state nodes within the tunnel, signifying key areas within the environment. Indicates spatiotemporal relationships, This represents the feature vector of each node.

4. The control method according to claim 1, characterized in that, Step two specifically involves: acquiring the sensor information collected by the sensor, and using the Kullback-Leibler divergence to measure the relative entropy difference between the historical sensor information distribution and the collected sensor information distribution, as shown in the following formula. in, This indicates the distribution of historical sensing information from the sensor. This indicates the distribution of currently collected sensor information. The larger the value, the greater the difference between the collected data and historical data, indicating that the currently collected sensor information is more valuable.

5. The control method according to claim 1, characterized in that, Step three specifically involves: Initialization strategy Distributed value network Target network The experience pool is initial velocity The entropy target is H, and the quantile parameter is... According to the current strategy Perform actions to gain experience Add to experience pool; Remove from experience pool sampling Calculate the target distribution : Update Distributed Value Network : in Measure the distribution; update the policy network. : Update speed : Update the target network: Training ends when the target network converges. Where s represents the current state of the trackless rubber-tired vehicle, which is obtained by collecting environmental information around the trackless rubber-tired vehicle and then passing it through a state encoder; The output speed information for the trackless rubber-wheeled vehicle is provided; r is the reward function, which represents the time it takes for the trackless rubber-wheeled vehicle to reach the target point, whether a collision occurs, and whether the path is smooth. The shorter the time to reach the target point, the fewer collisions with obstacles, and the smoother the path, the greater the reward; conversely, the smaller the reward function. The trackless rubber-wheeled vehicle selects the operation with the highest Q value based on the real-time environmental information fed back by the environmental assessment module, and obtains the corresponding reward. The value represents the real-time speed information of the trackless rubber-wheeled vehicle.

6. The control method according to claim 1, characterized in that, Step five specifically involves: ① Establish the system dynamic equations in, For position and velocity, To control the gain, For disturbance ; ② Definition of trajectory tracking error Tracking error: ; ③ Slipform surface design Design the sliding surface s to constrain error dynamics: ;in To control the convergence rate; ④ Calculate the equivalent control and design the switching control, as well as the total control input; in, To suppress the gain of the disturbance, , Boundary layer thickness; ⑤ Stability conditions: Convergence conditions .

Images