Underwater propulsion control method based on AI

Abstract

The present application provides an AI-based underwater propulsion control method, which is applied to the underwater propulsion system of an underwater robot. The underwater propulsion system includes k thrusters. The method includes: obtaining the real-time state and expected state of the underwater robot; determining the dynamic equation and dynamic model of the underwater propulsion system based on the real-time state of the underwater robot; determining the driving signal of the underwater propulsion system based on the real-time state and expected state of the underwater robot, as well as the dynamic equation and dynamic model of the underwater propulsion system, through a control law based on a neural network, so as to drive the operation of each thruster in the underwater propulsion system through the driving signal. The dynamic equation and dynamic model of the underwater propulsion system can accurately describe the dynamic behavior of the underwater propulsion system, thereby improving the control accuracy. The introduction of neural networks enables the control strategy to adapt to complex and changeable underwater environments, and further optimizes the stability of the control on the basis of ensuring control efficiency.

Description

Technical Field

[0001] The present application relates to the field of underwater robot driving technology, and more specifically, to an AI-based underwater propulsion control method. Background Art

[0002] As the core component of an underwater robot, the underwater propulsion system mainly provides the necessary thrust to ensure that the underwater robot can operate stably in the water according to the preset trajectory and speed. The underwater propulsion system is usually composed of multiple thrusters, each of which has the ability to independently control the thrust size and direction. By collaboratively adjusting the output of each thruster, the system can achieve all-round motion control of the underwater robot.

[0003] Due to the complexity and dynamics of the underwater environment, the propulsion system needs to adopt an efficient and stable control strategy. Existing underwater thruster control strategies generally include PID control (relatively simple, not very suitable for complex underwater environments, but applicable to simple static water environments), adaptive control, and sliding mode control. However, in complex underwater environments, signal transmission error is an important factor affecting the control performance of the propulsion system, and underwater communication is limited by the complexity of the environment. Therefore, a control terminal (located on land) is usually used to control the target of the propulsion system (such as the state of movement, the destination of movement, etc.), and the underwater propulsion system generates a drive signal according to its own real-time state to achieve the target.

[0004] Sliding mode control is very suitable for controlling underwater robots in complex underwater environments, but due to the different types of underwater robots, the focus of their control is also different. Some underwater robots focus more on stability and need to avoid the impact of emergencies on underwater robots as much as possible (requiring more complex underwater drive control strategies, and the overall control efficiency of underwater robots will be lower). This type of underwater robot is more suitable for performing more precise underwater tasks. For underwater robots that do not need to perform precise underwater tasks, further improving control stability while ensuring control efficiency is an optimization direction that needs to be considered. Therefore, how to provide a more suitable underwater drive control scheme to ensure the control efficiency of underwater robots performing routine tasks (such as cleaning the outer surface of underwater devices and equipment) while improving control stability is a problem that needs to be solved in this field. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide an AI-based underwater propulsion control method to optimize the underwater drive control strategy through artificial intelligence and improve the stability of underwater drive control.

[0006] In order to achieve the above purpose, the embodiments of the present application are implemented in the following manner:

[0007] In a first aspect, an embodiment of the present application provides an AI-based underwater propulsion control method, which is applied to an underwater propulsion system of an underwater robot, wherein the underwater propulsion system includes k thrusters, and the method includes: obtaining a real-time state and an expected state of the underwater robot; determining a dynamic equation and a dynamic model of the underwater propulsion system based on the real-time state of the underwater robot; determining a driving signal of the underwater propulsion system through a control law based on a neural network based on the real-time state and the expected state of the underwater robot, as well as the dynamic equation and the dynamic model of the underwater propulsion system, so as to drive the operation of each thruster in the underwater propulsion system through the driving signal.

[0008] In combination with the first aspect, in a first possible implementation of the first aspect, obtaining the real-time state and the expected state of the underwater robot includes: obtaining the real-time state of the underwater robot, including parameters of the underwater robot in an inertial coordinate system and parameters in a propulsion system coordinate system, wherein the parameters in the inertial coordinate system are x, y, z are the x-coordinate, y-coordinate, and z-coordinate of the underwater robot in the inertial coordinate system. θ,μ are the rolling angles of the underwater robot in the inertial coordinate system. Pitch angle θ and yaw angle μ, parameters v in the propulsion system coordinate system = [abc α β γ] T ∈R 6×1 , a, b, c are the linear velocity components of the underwater robot on each axis in the propulsion system coordinate system, α, β, γ are the angular velocity components of the underwater robot on each axis in the propulsion system coordinate system; obtain the desired state of the underwater robot, including the target parameters of the underwater robot in the inertial coordinate system at the next time node, and the target parameters in the inertial coordinate system x d ,y d ,z d They are the target x-coordinate, target y-coordinate, and target z-coordinate of the underwater robot in the inertial coordinate system at the next time node. θ, μ are the target roll angle, target pitch angle and target yaw angle of the underwater robot in the inertial coordinate system at the next time node.

[0009] In combination with the first possible implementation manner of the first aspect, in a second possible implementation manner of the first aspect, determining a dynamic equation and a dynamic model of an underwater propulsion system based on a real-time state of the underwater robot includes: determining a dynamic equation of the underwater propulsion system based on the real-time state of the underwater robot:

[0010]

[0011] in, is the derivative of Φ(t), Φ(t) is the function of parameter Φ with respect to time t, J∈R 6×6 represents the rotation matrix, which is used to realize the conversion between the inertial coordinate system and the propulsion system coordinate system, v(t) is the function of the parameter v with respect to time t; and, based on the real-time state of the underwater robot, the dynamic model of the underwater propulsion system is determined:

[0012]

[0013] in, is the derivative of v(t), f = -M -1 [C+D],f∈R 6×6 is a system parameter matrix, M∈R 6×6 is the matrix reflecting mass and inertia, C∈R 6×6 is the Coriolis centripetal force matrix, D = M -1 d, D∈R 6×1 is the matrix reflecting the system damping and external disturbance, d∈R 6×1 is an external disturbance, G∈R 6×1 is the matrix reflecting gravity, buoyancy and their torque, E∈R 6×k is the thruster configuration matrix, u(t)=[u 1 ,u 2 ,...,u k ] T ∈R k×1 , is the driving signal.

[0014] In combination with the second possible implementation method of the first aspect, in a third possible implementation method of the first aspect, based on the real-time state and expected state of the underwater robot and the dynamic model of the underwater propulsion system, a driving signal of the underwater propulsion system is determined by a control law based on a neural network, including: determining the error vector at the current time node based on the real-time state and expected state of the underwater robot; calculating the driving signal of the underwater propulsion system by a control law based on a neural network based on the error vector at the current time node and the dynamic model of the underwater propulsion system.

[0015] In combination with the third possible implementation of the first aspect, in a fourth possible implementation of the first aspect, based on the real-time state and the expected state of the underwater robot, determining the error vector at the current time node includes: based on the expected state Φ of the underwater robot d , the target parameter v of the underwater robot in the propulsion system coordinate system is determined by using the dynamic equation of the underwater propulsion system d , v d =[α d b d c d α d βd γ d ] T ∈R 6×1 , a d ,b d ,c d They are the target linear velocity components of the underwater robot on each axis in the propulsion system coordinate system, α d ,β d ,γ d are the target angular velocity components of the underwater robot on each axis in the propulsion system coordinate system; based on the real-time state and expected state of the underwater robot, the error vector e(t) at the current time node is determined:

[0016] e(t)=v d (t)-v(t),

[0017] Where e(t)∈R 6×1 is the error vector, e(t), v d v(t) and v(t) are the corresponding parameters expressed as functions of time t.

[0018] In combination with the fourth possible implementation manner of the first aspect, in a fifth possible implementation manner of the first aspect, the process of constructing a control law based on a neural network is as follows: constructing an integral sliding surface:

[0019]

[0020] Where s(t) is the integral sliding surface, λ is a positive number used to adjust the influence of the integral term; based on the integral sliding surface, the equivalent control term u is determined eq (t) and the switching control item u sw (t); Based on the real-time state and expected state of the underwater robot and the error vector e(t), the compensation term u is determined by the neural network nn (t); Based on this, the designed control law is:

[0021] u(t)=u eq (t)+u sw (t)+u nn (t),

[0022] Where u(t)∈R k×1 is the driving signal, u eq (t)∈R k×1 is an equivalent control term, u sw (t)∈R k×1 To switch the control item, u nn (t)∈R k×1 For compensation.

[0023] In combination with the fifth possible implementation manner of the first aspect, in a sixth possible implementation manner of the first aspect, based on the integral sliding surface, an equivalent control term u is determined. eq (t) is as follows:

[0024] Taking the derivative of s(t), we get:

[0025]

[0026] make Substituting the dynamic equations and kinetic models into In, we get:

[0027]

[0028] Here, 0 is the zero vector;

[0029] Calculate and solve, and get:

[0030]

[0031] Among them, u eq (t) is an equivalent control item.

[0032] In combination with the fifth possible implementation manner of the first aspect, in a seventh possible implementation manner of the first aspect, based on the integral sliding surface, the switching control item u is determined. sw (t) is as follows:

[0033] Designing a switch control sw (t) is a symbolic function that satisfies:

[0034] u sw (t) = -κsgn(s(t)),

[0035]

[0036] Among them, u sw (t) is the switching control term, κ∈R k×1 is a design parameter used to adjust the strength of switching control, sgn(s(t)) is a sign function, and 0 is a zero vector.

[0037] In combination with the seventh possible implementation manner of the first aspect, in an eighth possible implementation manner of the first aspect, the design parameter κ is determined by using a neural network model, wherein the neural network model is a multilayer perceptron.

[0038] In combination with the eighth possible implementation of the first aspect, in a ninth possible implementation of the first aspect, based on the real-time state and the expected state of the underwater robot and the error vector e(t), a compensation term u is determined by a neural network. nn(t), including: inputting the real-time state and the expected state, the error vector e(t) and the design parameter κ into the neural network model, performing nonlinear mapping through the neural network model, and finally determining the compensation term u nn (t) and output.

[0039] Beneficial effects:

[0040] 1. This scheme obtains the real-time state and expected state of the underwater robot, and further determines the dynamic equation and dynamic model of the underwater propulsion system; then based on the real-time state and expected state of the underwater robot, as well as the dynamic equation and dynamic model of the underwater propulsion system, the driving signal of the underwater propulsion system is determined through the control law based on the neural network, so as to drive the operation of each propeller in the underwater propulsion system through the driving signal. Through the state parameters obtained in real time, using the dynamic equation and dynamic model of the underwater propulsion system, the dynamic behavior of the underwater propulsion system (characterizing the underwater robot) can be accurately described, including factors such as mass, inertia, Coriolis centripetal force, damping and external disturbance, thereby improving the control accuracy. The introduction of neural networks enables the control strategy to automatically learn and adapt to the complex and changeable underwater environment, and further optimizes the stability of control on the basis of ensuring control efficiency.

[0041] 2. Using the error vector at the current time node and generating a drive signal based on the error vector through the neural network control law is conducive to the rapid tracking and accurate control of the target state. Based on the state parameters of the underwater robot in multiple dimensions (position, posture, speed, etc.) and the detailed dynamic characteristics of the propulsion system, the constructed dynamic model is conducive to making the control strategy more comprehensive and detailed. In addition, this scheme has strong flexibility and can be flexibly applied to different types of underwater robots. It only needs to adjust the relevant parameters and models according to the characteristics and task requirements of the specific robot. The control law optimized by the neural network can achieve accurate and efficient control. In addition, the system damping and external disturbances are taken into account in the dynamic model, so that the system can maintain good stability and anti-interference ability when facing complex underwater environments. This method can be seamlessly integrated with the existing underwater robot platform, which is convenient for upgrading and transformation of the existing system.

[0042] 3. Based on sliding mode control (with the advantage of efficient control), the integral term is introduced into the sliding surface, so that the system can accumulate past error information, which helps to eliminate static errors and improve the steady-state accuracy of the system, and increase the robustness of the system to external disturbances and parameter changes. The equivalent control term derived based on the integral sliding surface can drive the system state to the sliding surface, directly relying on the dynamic equations and dynamic models of the system design, and can provide precise control. The switching control term adopts the form of a symbolic function, which can quickly adjust the size and direction of the control force according to the sign of the error, thereby accelerating the convergence speed of the system state and reducing the time to reach the sliding surface. At the same time, by adjusting the design parameter κ, the intensity of the switching control can be flexibly controlled to adapt to different control requirements. At the same time, the real-time state, the desired state, the error vector and the design parameter κ are nonlinearly mapped using a neural network to generate compensation terms, and finally a complete control law is constructed. Neural networks (such as multilayer perceptrons) can learn and adapt to the uncertainties and nonlinear factors in the system dynamics, and make certain adjustments as the design parameter κ changes, thereby providing more precise and flexible compensation control. It can not only further improve the control accuracy and dynamic performance, but also effectively reduce or even eliminate the jitter problem caused by switching control, and further improve the stability of control.

[0043] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 Schematic diagram of the propulsion drive framework of the underwater propulsion system in the underwater robot.

[0046] Figure 2 A flowchart of an AI-based underwater propulsion control method provided in an embodiment of the present application.

[0047] Figure 3 Schematic diagram of the neural network model. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0049] See also Figure 1 , Figure 1 Schematic diagram of the propulsion drive framework of the underwater propulsion system in the underwater robot.

[0050] In this embodiment, the control terminal (such as the land control console of the underwater robot, the intelligent terminal, etc.) controls the operating state of the underwater robot (such as which destination to go to, at what speed to travel, the state to be achieved, etc.), and the underwater propulsion system of the underwater robot is equipped with k thrusters. Based on the operating state to be achieved, the signal is converted into a driving signal to control the k thrusters (part or all), thereby comprehensively realizing the drive control of the underwater robot.

[0051] In order to facilitate the understanding of this scheme, the AI-based underwater propulsion control method will be introduced in detail using the underwater propulsion system.

[0052] See also Figure 2 , Figure 2 Flow chart of the AI-based underwater propulsion control method provided in the embodiment of the present application. In this embodiment, the AI-based underwater propulsion control method is applied to the underwater propulsion system of the underwater robot, and may include step S10, step S20, and step S30.

[0053] First, the underwater propulsion system may execute step S10.

[0054] Step S10: Acquire the real-time state and expected state of the underwater robot.

[0055] In this embodiment, the underwater propulsion system can obtain the real-time state and expected state of the underwater robot.

[0056] The real-time state of the underwater robot includes the parameters of the underwater robot in the inertial coordinate system and the parameters in the propulsion system coordinate system, among which the parameter Φ in the inertial coordinate system is:

[0057]

[0058] Among them, x, y, z are the x-coordinate, y-coordinate, and z-coordinate of the underwater robot in the inertial coordinate system, respectively. θ,μ are the rolling angles of the underwater robot in the inertial coordinate system. The pitch angle θ and the yaw angle μ.

[0059] Parameter v of the underwater robot in the propulsion system coordinate system:

[0060] v=[abc α β γ] T ∈R 6×1 , (2)

[0061] Among them, a, b, c are the linear velocity components of the underwater robot on each axis in the propulsion system coordinate system, and α, β, γ are the angular velocity components of the underwater robot on each axis in the propulsion system coordinate system.

[0062] The desired state of the underwater robot includes the target parameters of the underwater robot in the inertial coordinate system at the next time node and the target parameters Φ in the inertial coordinate system d :

[0063]

[0064] Among them, x d ,y d ,z d They are the target x-coordinate, target y-coordinate, and target z-coordinate of the underwater robot in the inertial coordinate system at the next time node. θ, μ are the target roll angle, target pitch angle and target yaw angle of the underwater robot in the inertial coordinate system at the next time node.

[0065] In this embodiment, the main control unit of the underwater robot can analyze the control target, thereby generating the expected state of the underwater robot at each time node (such as the position and posture of the underwater robot at each time node, etc.) based on the planning method built into the main control unit. Of course, the expected state can also be generated by the control terminal based on the control target and the underwater environmental parameters, and then sent to the underwater robot, and the underwater robot obtains the expected state through analysis, which is not limited here. In addition, the parameters Φ, parameters v and other parameters in this embodiment are shown here in static form, and these parameters can be written as functions of time t, that is, dynamic forms, such as Φ(t), v(t), etc. This is essentially different forms of the same parameters in different scenarios, and the parameters Φ and parameters v obtained in real time are not written as functions of time t, but should not be regarded as a limitation of the present application.

[0066] After obtaining the real-time state and the expected state of the underwater robot, the underwater propulsion system may execute step S20.

[0067] Step S20: Based on the real-time state of the underwater robot, determine the dynamic equation and dynamic model of the underwater propulsion system.

[0068] In this embodiment, the underwater propulsion system can determine the dynamic equation of the underwater propulsion system based on the real-time state of the underwater robot:

[0069]

[0070] in, is the derivative of Φ(t), Φ(t) is the function of parameter Φ with respect to time t, J∈R 6×6Represents the rotation matrix, which is used to realize the conversion between the inertial coordinate system and the propulsion system coordinate system. v(t) is the function of parameter v with respect to time t.

[0071] And, the underwater propulsion system can determine the dynamic model of the underwater propulsion system based on the real-time state of the underwater robot:

[0072]

[0073] in, is the derivative of v(t), f = -M -1 [C+D],f∈C 6×6 is a system parameter matrix, M∈R 6×6 is the matrix reflecting mass and inertia, C∈R 6×6 is the Coriolis centripetal force matrix, D = M -1 d, D∈R 6×1 is the matrix reflecting the system damping and external disturbance, d∈R 6×1 is an external disturbance, G∈R 6×1 is the matrix reflecting gravity, buoyancy and their torque, E∈R 6×j is the thruster configuration matrix, u(t)=[u 1 ,u 2 ,...,u k ] T ∈R k×1 , is the driving signal. The driving signal u(t) at this time is undetermined and needs to be determined according to the control law later.

[0074] After the dynamic equation and the dynamic model of the underwater propulsion system are determined, the underwater propulsion system may execute step S30.

[0075] Step S30: Based on the real-time state and expected state of the underwater robot, as well as the dynamic equation and dynamic model of the underwater propulsion system, a driving signal of the underwater propulsion system is determined through a control law based on a neural network, so as to drive the operation of each propeller in the underwater propulsion system through the driving signal.

[0076] In this embodiment, the underwater propulsion system can determine the driving signal of the underwater propulsion system based on the real-time state and expected state of the underwater robot, the dynamic equation and dynamic model of the underwater propulsion system, and the control law based on the neural network.

[0077] Exemplarily, the underwater propulsion system may determine the error vector at the current time node based on the real-time state and expected state of the underwater robot.

[0078] Specifically, the underwater propulsion system can be based on the desired state Φ of the underwater robot d, using the dynamic equation of the underwater propulsion system (Formula (4)), determine the target parameter v of the underwater robot in the propulsion system coordinate system d ,

[0079] v d =[a d b d c d α d β d γ d ] T ∈R 6×1 , (6)

[0080] Among them, a d ,b d ,c d They are the target linear velocity components of the underwater robot on each axis in the propulsion system coordinate system, α d ,β d ,γ d They are the target angular velocity components of the underwater robot on each axis in the propulsion system coordinate system.

[0081] Then, the underwater propulsion system can determine the error vector e(t) at the current time node based on the real-time state and expected state of the underwater robot:

[0082] e(t)=v d (t)-v(t), (7)

[0083] Where e(t)∈R 6×1 is the error vector, e(t), v d (t), v(t) are the corresponding parameters (i.e., e, v d , v) is written as a function of time t.

[0084] After determining the error vector e(t) at the current time node, the underwater propulsion system can further calculate the driving signal of the underwater propulsion system based on the error vector at the current time node and the dynamic model of the underwater propulsion system through a control law based on a neural network.

[0085] For ease of understanding, the construction process of the control law based on neural network is first introduced here.

[0086] Since this solution is based on sliding mode control and introduces neural network for optimization, the construction process of the control law based on neural network in this embodiment is as follows:

[0087] Construct an integral sliding surface:

[0088]

[0089] Among them, s(t) is the integral sliding surface, and λ is a positive number used to adjust the influence of the integral term.

[0090] Then, based on the integral sliding surface, the equivalent control term u can be determined eq (t) and the switching control item u sw (t).

[0091] Since the equivalent control term u eq (t) is mainly used for equivalent control items to keep the system state on the sliding surface. When the system state has reached the sliding surface, maintaining the system state on the sliding surface can be understood as providing a control force that matches the current system state to offset the uncertainty within the system and the impact of external disturbances to ensure that the system can slide along the sliding surface, thereby achieving stable control of the system state and ensuring the stability and robustness of the system.

[0092] Therefore, the system state is on the sliding surface during equivalent control, that is, s(t) = 0, and to determine the equivalent control term u eq (t)∈R k×1 , we can take the derivative of s(t) and get:

[0093]

[0094] Re-order Substitute the dynamic equation (Formula (4)) and the kinetic model (Formula (5)) into In, we get:

[0095]

[0096] Here, 0 is a zero vector and is of order 6×1.

[0097] Calculate and solve, and get:

[0098]

[0099] Among them, u eq (t)∈R k×1 is an equivalent control term. And the parameter v d (t) can be used to calculate the parameter Φ d (t) is determined by the dynamic equation of the underwater propulsion system (formula (4)) (i.e., by calculating the first derivative) and is considered to be known; and Then the expected state parameter Φ d (t) Take the second-order derivative to get (and the desired state is known through the built-in planning algorithm), so, Also considered as known. In addition, it can be estimated by the difference method This is not a limitation.

[0100] The switch control u sw (t) is mainly used to provide a sufficiently large control force when the system state is far away from the sliding surface, so that the system state can quickly reach the sliding surface within a limited time. When the system is disturbed by external factors or the system parameters change, the switching control item can quickly adjust the control force to ensure that the system state can return to the sliding surface, thereby maintaining the stability and control performance of the system. Therefore, it is necessary to determine the switching control item u sw (t)∈R k×1 , usually using symbolic functions.

[0101] Exemplarily, the design switching control item u sw (t) is a symbolic function that satisfies:

[0102] u sw (t) = -κsgn(s(t)), (12)

[0103]

[0104] Among them, u sw (t)∈R k×1 is the switching control term, κ∈R k×1 is a design parameter used to adjust the strength of switching control. sgn(s(t)) is a sign function, 0 is a zero vector, the 0 corresponding to s(t) is a 6×1 order zero vector, and the 0 corresponding to sgn(s(t)) is a k×1 order zero vector.

[0105] In order to better adapt the intensity of switching control and maintain the stability and control performance of system control, it is necessary to determine the design parameter κ. The design parameter κ can generally be determined by empirical setting or adaptive method. This scheme introduces a neural network to determine this design parameter κ. In addition, since the switching control term u sw (t) High-frequency oscillation or chattering may occur during the control process. Therefore, based on this, this scheme also uses a neural network to determine a compensation term u nn (t)∈R k×1 , in order to slow down or even eliminate the jitter problem caused by switching control as much as possible.

[0106] For ease of understanding, the neural network model used in this solution is first introduced here. This solution uses a multi-layer perceptron (you can also choose a feedforward neural network, a radial basis neural network, or a long short-term memory network, etc., which is not limited here) to build a neural network model. Figure 3 As shown, the neural network model includes an input layer, a first hidden layer, a first output layer, a second hidden layer and a second output layer. The input layer receives input data (e.g., parameters Φ(t), v(t), Φ d (t), vd (t), e(t), s(t), The first hidden layer (including multiple sub-hidden layers) is then mapped to the output of the first output layer (the output of the first output layer is the design parameter κ, which is a k×1 order vector); at the same time, the output of the first output layer (the design parameter κ) is combined with the input data of the input layer (consisting of parameters Φ(t), v(t), Φ d (t), v d (t), e(t), s(t), After being spliced ​​again, the new input data is input to the second hidden layer (including multiple sub-hidden layers), and after being processed, it is mapped to the second output layer output, and the output switching control item u sw (t) The corresponding compensation term u nn (t), this compensation term plays a compensatory role in the switching control to effectively reduce or even eliminate the jitter problem caused by the switching control.

[0107] Of course, the construction and training of the neural network model can be built through the deep learning framework Tensorflow, but the collection of training data is best divided into two parts of the training data set according to the design of the model (each part of the training data set also reserves a part, which is merged for subsequent joint training). One part of the training data set is responsible for training the first part (that is, the model structure formed by the input layer, the first hidden layer and the first output layer is back-propagated according to the design parameter κ of its output), and the other part of the training data set is responsible for training the second part (that is, the model structure formed by the comprehensive input layer composed of the input layer + the first output layer, the second hidden layer and the second output layer is back-propagated according to the compensation term u output by it). nn (t) perform back propagation training), complete the training of the first part and the second part separately, and then perform joint training, and finally obtain a trained neural network model, which is built into the underwater propulsion system of the underwater robot to realize the drive control of the underwater robot.

[0108] Based on this, the underwater propulsion system can be based on the equivalent control term u eq (t), switch control item u sw (t) and the compensation term u nn (t), the complete control law is constructed as:

[0109] u(t)=u eq (t)+u sw (t)+u nn (t), (14)

[0110] Where u(t)∈Rk×1 is the driving signal, u eq (t)∈R k×1 is an equivalent control term, u sw (t)∈R k×1 To switch the control item, u nn (t)∈R k×1 It should be noted that in formula (14), although each term is connected by a plus sign, in essence, the equivalent control term u eq (t) To deal with the equivalent control part (i.e., the system state is on the sliding surface, i.e., s(t) = 0), the switching control item u sw (t) and the corresponding compensation term u nn (t) is used to realize the switching control part (i.e., the system state is not on the sliding surface, i.e., s(t)≠0), so the compensation term u nn (t) has another limiting condition, that is, the compensation term is needed only when s(t)≠0. When s(t)=0, the compensation term u nn (t) = 0. Then, the complete control law can also be written as:

[0111]

[0112] The above is an introduction to the control law of the neural network introduced in this scheme. On this basis, the underwater propulsion system can determine the driving signal u(t)∈R of the underwater propulsion system through the control law based on the neural network. k×1 , to control the k thrusters in the underwater propulsion system, thereby realizing the drive control of the underwater robot.

[0113] In summary, the embodiment of the present application provides an AI-based underwater propulsion control method, which further determines the dynamic equation and dynamic model of the underwater propulsion system by acquiring the real-time state and expected state of the underwater robot; then based on the real-time state and expected state of the underwater robot, as well as the dynamic equation and dynamic model of the underwater propulsion system, the driving signal of the underwater propulsion system is determined through the control law based on the neural network, so as to drive the operation of each propeller in the underwater propulsion system through the driving signal. Through the state parameters obtained in real time, using the dynamic equation and dynamic model of the underwater propulsion system, the dynamic behavior of the underwater propulsion system (characterizing the underwater robot) can be accurately described, including factors such as mass, inertia, Coriolis centripetal force, damping and external disturbances, thereby improving the control accuracy. The introduction of neural networks enables the control strategy to automatically learn and adapt to complex and changeable underwater environments, and further optimizes the stability of control on the basis of ensuring control efficiency.

[0114] Using the error vector at the current time node and generating a drive signal based on the error vector through a neural network control law is conducive to rapid tracking and accurate control of the target state. Based on the state parameters of the underwater robot in multiple dimensions (position, posture, speed, etc.) and the detailed dynamic characteristics of the propulsion system, the constructed dynamic model is conducive to making the control strategy more comprehensive and detailed. In addition, this scheme has strong flexibility and can be flexibly applied to different types of underwater robots. It only needs to adjust the relevant parameters and models according to the characteristics and task requirements of the specific robot. The control law optimized by the neural network can achieve accurate and efficient control. In addition, the system damping and external disturbances are considered in the dynamic model, so that the system can maintain good stability and anti-interference ability when facing complex underwater environments. This method can be seamlessly integrated with the existing underwater robot platform, which is convenient for upgrading and transformation of the existing system.

[0115] Based on sliding mode control (with the advantage of efficient control), the integral term is introduced into the sliding surface, so that the system can accumulate past error information, which helps to eliminate static errors and improve the steady-state accuracy of the system, and increase the robustness of the system to external disturbances and parameter changes. The equivalent control term derived based on the integral sliding surface can drive the system state to the sliding surface, directly relying on the dynamic equations and dynamic models of the system design, and can provide precise control. The switching control term adopts the form of a symbolic function, which can quickly adjust the size and direction of the control force according to the sign of the error, thereby accelerating the convergence speed of the system state and reducing the time to reach the sliding surface. At the same time, by adjusting the design parameter κ, the intensity of the switching control can be flexibly controlled to adapt to different control requirements. At the same time, the real-time state, the desired state, the error vector and the design parameter κ are nonlinearly mapped using a neural network to generate compensation terms, and finally a complete control law is constructed. Neural networks (such as multilayer perceptrons) can learn and adapt to the uncertainties and nonlinear factors in the system dynamics, and make certain adjustments as the design parameter κ changes, thereby providing more precise and flexible compensation control. It can not only further improve the control accuracy and dynamic performance, but also effectively reduce or even eliminate the jitter problem caused by switching control, and further improve the stability of control.

[0116] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An AI-based underwater propulsion control method, characterized in that: An underwater propulsion system applied to an underwater robot, the underwater propulsion system comprising k thrusters, the method comprising: Obtain the real-time status and expected status of the underwater robot; Based on the real-time status of the underwater robot, the dynamic equation and dynamic model of the underwater propulsion system are determined; Based on the real-time state and expected state of the underwater robot, as well as the dynamic equation and dynamic model of the underwater propulsion system, a driving signal of the underwater propulsion system is determined through a control law based on a neural network, so as to drive the operation of each propeller in the underwater propulsion system through the driving signal; Obtain the real-time status and expected status of the underwater robot, including: Obtain the real-time status of the underwater robot, including the parameters of the underwater robot in the inertial coordinate system and the parameters in the propulsion system coordinate system. x, y, z are the x-coordinate, y-coordinate, and z-coordinate of the underwater robot in the inertial coordinate system. They are the roll angle of the underwater robot in the inertial coordinate system Pitch angle θ and yaw angle μ, parameters v = [abcαβγ] in the propulsion system coordinate system T ∈R 6×1 , a, b, c are the linear velocity components of the underwater robot on each axis in the propulsion system coordinate system, α, β, γ are the angular velocity components of the underwater robot on each axis in the propulsion system coordinate system; Obtain the desired state of the underwater robot, including the target parameters of the underwater robot in the inertial coordinate system at the next time node and the target parameters in the inertial coordinate system x d ,y d ,z d They are the target x-coordinate, target y-coordinate, and target z-coordinate of the underwater robot in the inertial coordinate system at the next time node. They are the target roll angle, target pitch angle, and target yaw angle of the underwater robot in the inertial coordinate system at the next time node; Based on the real-time status of the underwater robot, the dynamic equations and dynamic models of the underwater propulsion system are determined, including: Based on the real-time state of the underwater robot, the dynamic equation of the underwater propulsion system is determined: in, is the derivative of Φ(t), Φ(t) is the function of parameter Φ with respect to time t, J∈R 6×6 represents the rotation matrix, which is used to realize the transformation between the inertial coordinate system and the propulsion system coordinate system. v(t) is the function of the parameter v with respect to time t; And, based on the real-time state of the underwater robot, the dynamic model of the underwater propulsion system is determined: in, is the derivative of v(t), f = -M -1 [C+D],f∈R 6×6 is a system parameter matrix, M∈R 6×6 is the matrix reflecting mass and inertia, C∈R 6×6 is the Coriolis centripetal force matrix, D = M -1 d, D∈R 6×1 is the matrix reflecting the system damping and external disturbance, d∈R 6×1 is an external disturbance, G∈R 6×1 is the matrix reflecting gravity, buoyancy and their torque, E∈R 6×k is the thruster configuration matrix, u(t)=[u1,u2,...,u k ] T ∈R k×1 , is the driving signal.

2. The AI-based underwater propulsion control method according to claim 1, characterized in that: Based on the real-time state and expected state of the underwater robot and the dynamic model of the underwater propulsion system, the driving signal of the underwater propulsion system is determined through the control law based on the neural network, including: Based on the real-time state and expected state of the underwater robot, the error vector at the current time node is determined; Based on the error vector at the current time node and the dynamic model of the underwater propulsion system, the driving signal of the underwater propulsion system is calculated through the control law based on the neural network.

3. The AI-based underwater propulsion control method according to claim 2, characterized in that: Based on the real-time state and expected state of the underwater robot, the error vector at the current time node is determined, including: Based on the expected state Φ of the underwater robot d , the target parameter v of the underwater robot in the propulsion system coordinate system is determined by using the dynamic equation of the underwater propulsion system d , v d =[a d b d c d α d β d γ d ] T ∈R 6×1 , a d ,b d ,c d They are the target linear velocity components of the underwater robot on each axis in the propulsion system coordinate system, α d ,β d ,γ d They are the target angular velocity components of the underwater robot on each axis in the propulsion system coordinate system; Based on the real-time state and expected state of the underwater robot, the error vector e(t) at the current time node is determined: e(t)=v d (t)-v(t), Where e(t)∈R 6×1 is the error vector, e(t), v d v(t) and v(t) are the corresponding parameters expressed as functions of time t.

4. The AI-based underwater propulsion control method according to claim 3, characterized in that: The construction process of the control law based on neural network is as follows: Construct an integral sliding surface: Among them, s(t) is the integral sliding surface, and λ is a positive number used to adjust the influence of the integral term; Based on the integral sliding surface, the equivalent control term u is determined eq (t) and the switching control item u sw (t); Based on the real-time state and expected state of the underwater robot and the error vector e(t), the compensation term u is determined by the neural network. nn (t); Based on this, the designed control law is: u(t)=u eq (t)+u sw (t)+u nn (t), Where u(t)∈R k×1 is the driving signal, u eq (t)∈R k×1 is an equivalent control term, u sw (t)∈R k×1 To switch the control item, u nn (t)∈R k×1 For compensation.

5. The AI-based underwater propulsion control method according to claim 4, characterized in that: Based on the integral sliding surface, the equivalent control term u is determined eq (t) is as follows: Taking the derivative of s(t), we get: make Substituting the dynamic equations and kinetic models into In, we get: Here, 0 is the zero vector; Calculate and solve, and get: Among them, u eq (t) is an equivalent control item.

6. The AI-based underwater propulsion control method according to claim 4, characterized in that: Based on the integral sliding surface, the switching control term u is determined sw (t) is as follows: Designing a switch control sw (t) is a symbolic function that satisfies: u sw (t)=-κsgn(s(t)), Among them, u sw (t) is the switching control item, k∈R k×1 is a design parameter used to adjust the strength of switching control, sgn(s(t)) is a sign function, and 0 is a zero vector.

7. The AI-based underwater propulsion control method according to claim 6, characterized in that: The design parameter κ is determined by using a neural network model, where the neural network model is a multi-layer perceptron.

8. The AI-based underwater propulsion control method according to claim 7, characterized in that: Based on the real-time state and expected state of the underwater robot and the error vector e(t), the compensation term u is determined by the neural network. nn (t), including: The real-time state and expected state, error vector e(t) and design parameter k are input into the neural network model, and the compensation term u is finally determined by performing nonlinear mapping through the neural network model. nn (t) and output.

Images