AUV navigation intelligent system and method

The AUV navigation system addresses navigation lag by employing RBF and DDPG controllers to independently control depth and trajectory, enhancing adaptability and calculation speed for real-time underwater operations.

CN115639747BActive Publication Date: 2024-12-03ZHEJIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211210220.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-30
Publication Date
2024-12-03
Estimated Expiration
2042-09-30

AI Technical Summary

Technical Problem

Existing AUV navigation methods face excessive calculation loads due to the complexity of underwater environments, leading to long calculation times and navigation lag, making real-time navigation impossible.

Method used

An AUV navigation system utilizing a depth control module with an RBF network adaptive sliding mode controller and vertical thruster, and a trajectory control module with a DDPG controller and horizontal thruster, to independently control depth and trajectory, respectively, enhancing adaptability and reducing calculation complexity.

Benefits of technology

The system significantly improves calculation speed and enables autonomous navigation in unknown environments by leveraging RBF and DDPG algorithms for adaptive learning, ensuring precise and efficient underwater operations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115639747B_ABST
    Figure CN115639747B_ABST
Patent Text Reader

Abstract

The present invention discloses an AUV navigation intelligent system and method, wherein the AUV navigation intelligent system is disclosed, including a depth control module and a trajectory control module, wherein the depth control module includes an RBF network adaptive sliding film controller and a vertical thruster, wherein the RBF network adaptive sliding film controller is used to convert depth data into vertical thrust data and input the vertical thrust data into the vertical thruster, and the vertical thruster is used to provide the AUV with thrust perpendicular to the horizontal plane; the trajectory control module includes a DDPG controller and a horizontal thruster, wherein the DDPG controller is used to convert the plane position data of the AUV into horizontal thrust data and input the horizontal reasoning data into the horizontal thruster, and the horizontal thruster is used to provide the AUV with thrust parallel to the horizontal plane. The present invention also discloses an AUV navigation intelligent method, which adopts the aforementioned AUV navigation intelligent system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of underwater robots, and in particular to an AUV navigation intelligent system and method. Background Art

[0002] In recent years, due to the continuous reduction of land resources, the ocean, which is rich in mineral resources and fishery resources, has become the focus of resource development. Due to the particularity of the marine environment, it is necessary to use autonomous underwater robots (AUVs) to detect and develop marine resources. The autonomous navigation capability of AUVs is an important prerequisite for the smooth implementation of underwater operations, and then the exploitation of marine resources in the underwater environment can be implemented, avoiding the safety risks brought by artificial underwater operations; that is, navigation technology is the core technology of AUVs.

[0003] In the prior art, a multi-AUV collaborative navigation method based on reinforcement learning is disclosed in a Chinese patent with publication number CN113029123A, which includes the following steps: Step 1: In an offline state, for the collaborative navigation needs of a single-master and multiple-slave multi-AUV, the master AUV uses the pre-planned trajectories of each slave AUV to learn the action value function Q corresponding to each slave AUV through the Q-learning method; Step 2: In an offline state, according to the action value function corresponding to each slave AUV learned in step 1, the master AUV selects the optimal action for each sub-stage in the navigation process to execute and obtain the optimal trajectory; Step 3: In the actual navigation process, the master AUV performs navigation calculations based on the optimal trajectory obtained in step 2, and the slave AUV performs navigation calculations based on the pre-planned trajectory and their respective navigation devices to achieve single-master and multiple-slave multi-AUV collaborative navigation.

[0004] Another example is the all-source navigation system and method for AUV disclosed in Chinese patent with publication number CN105547300A, which includes: a basic navigation module, which obtains the motion data of the AUV and calculates the navigation position of the AUV based on the motion data; a real-time terrain acquisition module, which obtains the real-time terrain data of the location of the AUV; an accurate terrain acquisition module, which matches the navigation position of the AUV and the real-time terrain data with the reference terrain data to obtain the accurate terrain data of the location of the AUV; a correction data acquisition module, which obtains the slope data of the AUV from the seabed as correction data; a navigation position correction module, which uses the slope data to correct the navigation position of the AUV; and an all-source navigation calculation module, which performs all-source navigation calculation based on the corrected navigation position of the AUV and the accurate terrain data of the location of the AUV to obtain the AUV all-source navigation data.

[0005] However, when the AUV navigation method disclosed in the above patent is specifically applied to AUV, due to the complexity of the underwater environment, the entire calculation amount is too large, and the excessive calculation amount will lead to a long calculation time, and the navigation will have serious lag, that is, real-time navigation cannot be performed. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention provides an AUV navigation intelligent system and method.

[0007] An AUV navigation intelligent system includes a depth control module for controlling the depth of the AUV in water and a trajectory control module for controlling the motion trajectory of the AUV on a plane parallel to the horizontal plane.

[0008] The depth control module includes an RBF network adaptive sliding film controller and a vertical thruster. The RBF network adaptive sliding film controller is used to convert depth data into vertical thrust data and input the vertical thrust data into the vertical thruster. The vertical thruster is used to provide the AUV with thrust perpendicular to the horizontal plane.

[0009] The trajectory control module includes a DDPG controller and a horizontal thruster. The DDPG controller is used to convert the plane position data of the AUV into horizontal thrust data and input the horizontal reasoning data into the horizontal thruster. The horizontal thruster is used to provide the AUV with thrust parallel to the horizontal plane.

[0010] Preferably, the control rate of the RBF network adaptive synovial controller is:

[0011]

[0012] Among them, the control rate T wd is the expected thrust, m is the mass of the AUV, is the additional mass matrix of the AUV in the depth direction, λ is the coefficient of the RBF network adaptive sliding film controller, is the depth error change rate, is the second-order derivative of the desired depth, Z w is the viscous hydrodynamic coefficient, is the instantaneous depth change rate, is the RBF network weight, is the Gaussian function output of the RBF network, η is a constant greater than zero, D is the upper bound of the absolute value of the heave disturbance force of the system, and sat(s) is the saturation function.

[0013] The reasoning process of control rate is as follows:

[0014] Let the desired depth be z d , the instantaneous depth is z, and the error between the expected depth and the instantaneous depth is e z, the depth error change rate is therefore:

[0015] e z =z d -z,

[0016] The RBF network adaptive sliding mode controller designs the sliding surface of the depth control system through the dynamic characteristics expected by the depth control system. After the depth control system reaches the sliding mode, the tracking error gradually converges to zero, and the convergence speed can be adjusted by the sliding surface parameter matrix. The key to defining the design of the RBF network adaptive sliding mode controller is to define the sliding surface as s, and

[0017] In the actual sliding mode controller, due to the influence of factors such as time lag and inertia, after the state of the depth control system reaches the sliding mode surface, it does not slide on the sliding mode surface but crosses back and forth. This phenomenon is called chattering. In order to weaken the chattering by the reaching law, the control rate of the sliding mode controller is set as:

[0018]

[0019] In the formula, sgn(s) is the sign function. By adjusting the parameters λ, D, η, the quality of the arrival process is guaranteed, thereby effectively weakening the high-frequency jitter of the control signal.

[0020] The symbolic function sgn(s) is expressed as follows:

[0021]

[0022] Due to the existence of the sign function sgn(s), chattering still occurs in the sliding mode. In order to further reduce the chattering of the sliding mode controller, the saturation function sat(s) is used instead of the sign function sgn(s) to eliminate the instantaneous changes caused by the switching control.

[0023] The saturation function sat(s) is expressed as follows:

[0024]

[0025] Where Δ is the boundary layer, constant Δ>0,

[0026] The smaller Δ is, the closer the saturation function is to the sign function. After simulation and experiment, Δ=1 is the most suitable. Replacing the sign function in the control rate expression with the saturation function, the control rate expression can be obtained:

[0027]

[0028] The random interference force in the underwater environment and the system time delay make it impossible for the AUV to accurately obtain the change of the dynamic model during the depth control process. The RBF network can approximate any nonlinear function in a compact and arbitrary precision. Therefore, this paper uses the RBF network to approximate the model change. In the control law The change of the term is the uncertainty of the model change, so the control rate can be expressed as:

[0029]

[0030] From the nonlinear approximation characteristics of the RBF network, we know that there is an ideal network weight W * , so that:

[0031]

[0032] In the formula, is the network Gaussian basis function output, ε is the network approximation error, |ε|≤ε N . Then the RBF network output is:

[0033]

[0034] In the formula, is the RBF network weight;

[0035] In summary, the control law of the RBF network adaptive sliding mode controller is:

[0036]

[0037] The stability of the synovial controller is analyzed below.

[0038] First, establish the Lyapunov equation V:

[0039]

[0040]

[0041] We can get:

[0042]

[0043] We can get:

[0044]

[0045] The output of the sliding membrane controller

[0046] T wd =T wd :

[0047]

[0048] According to the Lassalle invariance principle, when When t→∞, s=0, so the closed depth control system is asymptotically stable. When t→∞, s→0. The constant η determines its convergence speed. For the sliding membrane controller, the larger the upper bound D of the absolute value of the vertical disturbance force, the stronger the robustness of the system.

[0049] That is, the parameters and characteristics of the system depend only on the sliding surface and have nothing to do with external interference, so the sliding mode controller has strong robustness.

[0050] Preferably, the horizontal propeller includes a first propeller and a second propeller, and the thrusts provided by the first propeller and the second propeller are opposite or the same in direction.

[0051] Preferably, the DDPG network of the DDPG controller is trained using the following reward function:

[0052] r=r1+r2+r3

[0053]

[0054] r2=-100(|x t |≥200|||y t |≥200)

[0055]

[0056] Among them, r1 is the reward for the AUV to be close to the destination, r2 is the penalty for the AUV to be far away from the destination, r3 is the penalty for the distance between the AUV and the destination when the AUV deviates from the navigation trajectory and the penalty for the force deviation of the horizontal thruster causing the AUV to have an unsteady speed, and x t ,y t is the coordinate of the AUV’s position on the running track at time t, θ t is the speed direction of the AUV at time t, L t and R t is the thrust of the first and second thrusters at time t.

[0057] Preferably, the depth control module further comprises a depth sensor for detecting the instantaneous depth of the AUV and feeding back the depth to the RBF network adaptive sliding film controller.

[0058] An AUV navigation intelligent method, using any of the above AUV navigation intelligent systems, comprises the following steps:

[0059] The AUV enters the water and starts the AUV navigation intelligent system. The depth control module first controls the depth of the AUV. When the error between the instant depth of the AUV and the expected depth is zero, the trajectory control module controls the motion trajectory of the AUV. During the process of trajectory control, if the instant depth of the AUV deviates from the expected depth, the trajectory control module will suspend work until the depth control module controls the depth of the AUV to reach the expected depth. After that, the trajectory control module will control the running trajectory of the AUV until the AUV reaches the destination.

[0060] An AUV, using the AUV navigation intelligent system as described above for navigation

[0061] Compared with the prior art, the present invention is beneficial in that:

[0062] The depth control system and trajectory control system are used to control the depth and plane trajectory of the AUV respectively, which can significantly improve the calculation speed. At the same time, the depth control system and trajectory control system use RBF network and DDPG algorithm respectively to give the underwater robot the ability of autonomous learning, improve the robot's adaptability in working in different environments, and realize the autonomous navigation of the AUV in unknown environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A module diagram of the depth control system of the AUV navigation intelligent system provided by the present invention;

[0064] Figure 2 A module diagram of the trajectory control system of the AUV navigation intelligent system provided by the present invention;

[0065] Figure 3 A schematic diagram of the sliding membrane surface of the depth control system of the AUV navigation intelligent system provided by the present invention;

[0066] Figure 4 is the image of the saturation function sat(s);

[0067] Figure 5 This is a logic block diagram of the AUV navigation intelligent method provided by the present invention. DETAILED DESCRIPTION

[0068] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0069] like Figure 1 , 2 As shown, the AUV navigation intelligent system includes a depth control module for controlling the depth of the AUV in water and a trajectory control module for controlling the motion trajectory of the AUV on a plane parallel to the horizontal plane.

[0070] The depth control module includes an RBF network adaptive sliding film controller and a vertical thruster. The RBF network adaptive sliding film controller is used to convert depth data into vertical thrust data and input the vertical thrust data into the vertical thruster. The vertical thruster is used to provide the AUV with thrust perpendicular to the horizontal plane. The depth control module also includes a depth sensor for detecting the instantaneous depth of the AUV and feeding back the depth to the RBF network adaptive sliding film controller.

[0071] The trajectory control module includes a DDPG controller and a horizontal thruster. The DDPG controller is used to convert the plane position data of the AUV into horizontal thrust data and input the horizontal reasoning data into the horizontal thruster. The horizontal thruster is used to provide the AUV with thrust parallel to the horizontal plane (T r ); The horizontal propeller includes a first propeller and a second propeller, and the thrust directions provided by the first propeller and the second propeller are opposite or the same.

[0072] Preferably, the control rate of the RBF network adaptive synovial controller is:

[0073]

[0074] Among them, the control rate T wd is the expected thrust, m is the mass of the AUV, is the additional mass matrix of the AUV in the depth direction, λ is the coefficient of the RBF network adaptive sliding film controller, is the depth error change rate, is the second-order derivative of the desired depth, Z w is the viscous hydrodynamic coefficient, is the instantaneous depth change rate, is the RBF network weight, is the Gaussian function output of the RBF network, η is a constant greater than zero, D is the upper bound of the absolute value of the heave disturbance force of the system, and sat(s) is the saturation function.

[0075] The reasoning process of control rate is as follows:

[0076] Let the desired depth be z d , the instantaneous depth is z, and the error between the expected depth and the instantaneous depth is e z , the depth error change rate is therefore:

[0077] e z =z d -z,

[0078] The RBF network adaptive sliding mode controller designs the sliding surface of the depth control system through the dynamic characteristics expected by the depth control system. After the depth control system reaches the sliding mode, the tracking error gradually converges to zero, and the convergence speed can be adjusted by the sliding surface parameter matrix. The key to defining the design of the RBF network adaptive sliding mode controller is to define the sliding surface as s. The sliding surface is as follows Figure 3 shown, and

[0079] In the actual sliding mode controller, due to the influence of factors such as time lag and inertia, after the state of the depth control system reaches the sliding mode surface, it does not slide on the sliding mode surface but crosses back and forth. This phenomenon is called chattering. In order to weaken the chattering by the reaching law, the control rate of the sliding mode controller is set as:

[0080]

[0081] In the formula, sgn(s) is the sign function. By adjusting the parameters λ, D, η, the quality of the arrival process is guaranteed, thereby effectively weakening the high-frequency jitter of the control signal.

[0082] The symbolic function sgn(s) is expressed as follows:

[0083]

[0084] Due to the existence of the sign function sgn(s), chattering still occurs in the sliding mode. In order to further reduce the chattering of the sliding mode controller, the saturation function sat(s) is used instead of the sign function sgn(s) to eliminate the instantaneous changes caused by the switching control.

[0085] The saturation function sat(s) is expressed as follows:

[0086]

[0087] Where Δ is the boundary layer, constant Δ>0,

[0088] like Figure 4 As shown in the figure, the smaller Δ is, the closer the saturation function is to the sign function. After simulation and experiment, Δ=1 is the most suitable. Replacing the sign function in the control rate expression with the saturation function, the control rate expression can be obtained:

[0089]

[0090] The random interference force in the underwater environment and the system time delay make it impossible for the AUV to accurately obtain the change of the dynamic model during the depth control process. The RBF network can approximate any nonlinear function in a compact and arbitrary precision. Therefore, this paper uses the RBF network to approximate the model change. In the control law The change of the term is the uncertainty of the model change, so the control rate can be expressed as:

[0091]

[0092] From the nonlinear approximation characteristics of the RBF network, we know that there is an ideal network weight W * , so that:

[0093]

[0094] In the formula, is the network Gaussian basis function output, ε is the network approximation error, ε≤ε N . Then the RBF network output is:

[0095]

[0096] In the formula, is the RBF network weight;

[0097] In summary, the control law of the RBF network adaptive sliding mode controller is:

[0098]

[0099] The stability of the synovial controller is analyzed below.

[0100] First, establish the Lyapunov equation V:

[0101]

[0102]

[0103] We can get:

[0104]

[0105] We can get:

[0106]

[0107] The output of the sliding membrane controller

[0108] T wd =T wd :

[0109]

[0110] According to the Lassalle invariance principle, when When t→∞, s=0, so the closed depth control system is asymptotically stable. When t→∞, s→0. The constant η determines its convergence speed. For the sliding membrane controller, the larger the upper bound D of the absolute value of the vertical disturbance force, the stronger the robustness of the system.

[0111] That is, the parameters and characteristics of the system depend only on the sliding surface and have nothing to do with external interference, so the sliding mode controller has strong robustness.

[0112] The DDPG network of the DDPG controller is trained using the following reward function:

[0113] r=r1+r2+r3

[0114]

[0115] r2=-100(|x t |≥200|||y t |≥200)

[0116]

[0117] Among them, r1 is the reward for the AUV to be close to the destination, r2 is the penalty for the AUV to be far away from the destination, r3 is the penalty for the distance between the AUV and the destination when the AUV deviates from the navigation trajectory and the penalty for the force deviation of the horizontal thruster causing the AUV to have an unsteady speed, and x t ,y t is the coordinate of the AUV’s position on the running track at time t, θ t is the speed direction of the AUV at time t, L t and R t is the thrust of the first and second thrusters at time t.

[0118] like Figure 5 As shown, the AUV navigation intelligent method, using the aforementioned AUV navigation intelligent system, includes the following steps:

[0119] The AUV enters the water and starts the AUV navigation intelligent system. The depth control module first controls the depth of the AUV. When the error between the AUV's instant depth and the expected depth is After it is zero, the trajectory control module controls the motion trajectory of the AUV. During the trajectory control process, if the AUV If it is not zero, the trajectory control module will suspend its operation until the depth of the AUV controlled by the depth control module reaches the desired depth. Then the trajectory control module will control the running trajectory of the AUV until the error Δg between the instant position of the AUV and the destination is zero.

Claims

1. An AUV intelligent navigation system, characterized in that: It includes a depth control module for controlling the depth of the AUV in the water and a trajectory control module for controlling the motion trajectory of the AUV on a plane parallel to the horizontal plane. The depth control module includes an RBF network adaptive sliding mode controller and a vertical thruster. The RBF network adaptive sliding mode controller is used to convert depth data into vertical thrust data and input the vertical thrust data into the vertical thruster. The vertical thruster is used to provide the AUV with thrust perpendicular to the horizontal plane. The control rate of the RBF network adaptive sliding mode controller is: Among them, T wd is the control rate, i.e. the expected thrust, m is the mass of the AUV, is the additional mass matrix of the AUV in the depth direction, λ is the coefficient of the RBF network adaptive sliding mode controller, is the depth error change rate, is the second-order derivative of the desired depth, Z w is the viscous hydrodynamic coefficient, is the instantaneous depth change rate, is the RBF network weight, is the Gaussian function output of the RBF network, η is a constant greater than zero, D is the upper bound of the absolute value of the heave disturbance force of the system, and sat(s) is the saturation function; the value of the saturation function sat(s) is as follows: Where s is the sliding surface defined by the RBF network adaptive sliding mode controller, and e z is the error of instant depth; Δ is a constant of 1; The trajectory control module includes a DDPG controller and a horizontal thruster, wherein the DDPG controller is used to convert the plane position data of the AUV into horizontal thrust data and input the horizontal reasoning data into the horizontal thruster, and the horizontal thruster is used to provide the AUV with thrust parallel to the horizontal plane; The horizontal propeller comprises a first propeller and a second propeller, wherein the thrusts provided by the first propeller and the second propeller are opposite in direction or the same in direction; The DDPG network of the DDPG controller is trained using the following reward function: r=r1+r2+r3 r2=-100(|x t |≥200∥|y t |≥200) Among them, r1 is the reward for the AUV to be close to the destination, r2 is the penalty for the AUV to be far away from the destination, r3 is the penalty for the distance between the AUV and the destination when the AUV deviates from the navigation trajectory and the penalty for the force deviation of the horizontal thruster causing the AUV to have an unsteady speed, and x t ,y t is the coordinate of the AUV’s position on the running track at time t, θ t is the speed direction of the AUV at time t, l t and R t is the thrust of the first and second thrusters at time t.

2. The AUV intelligent navigation system according to claim 1, characterized in that: The depth control module also includes a depth sensor for detecting the instantaneous depth of the AUV and feeding back the depth to the RBF network adaptive sliding mode controller.

3. An AUV intelligent navigation method, characterized in that: The AUV intelligent navigation system according to any one of claims 1 to 2 comprises the following steps: The AUV enters the water and starts the AUV intelligent navigation system. The depth control module first controls the depth of the AUV. When the error between the instantaneous depth of the AUV and the expected depth is zero, the trajectory control module controls the motion trajectory of the AUV. During the process of controlling the trajectory, if there is a deviation between the instantaneous depth of the AUV and the expected depth, the trajectory control module will suspend work until the depth control module controls the depth of the AUV to reach the expected depth. After that, the trajectory control module will control the running trajectory of the AUV until the AUV reaches the destination.

4. An AUV, characterized in that: Navigation is performed using the AUV intelligent navigation system as described in any one of claims 1 to 2.