Undulating fin robot control method based on residual MLP and active disturbance rejection controller

Abstract

The invention relates to the technical field of underwater vehicle control, and discloses a undulating fin robot control method based on a residual MLP and an active disturbance rejection controller, and the method comprises the steps: obtaining the state information of a robot through a kinematics and dynamics model; a non-linear tracking differentiator is adopted to generate a smooth reference signal, disturbance such as ocean current and density change is estimated in real time in combination with an extended state observer, and high-precision depth tracking is achieved through a state error feedback controller. The attitude control system fuses a PID controller and a depth factor adjustment mechanism, and dynamically optimizes control parameters to maintain attitude stability; a fin ray cooperative control strategy is put forward, the swing amplitude and the offset angle are adjusted, output torque is adjusted in combination with a depth factor, and propulsion performance fluctuation caused by water pressure changes is solved; the depth and attitude control parameters are optimized by using the residual MLP, the nonlinear fitting capability is enhanced through the residual block structure, the parameter setting efficiency is improved under the finite computing power, and an efficient and reliable technical scheme is provided for ocean exploration and ecological monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater vehicles, and in particular to a control method for an undulating fin robot based on a residual MLP and an auto-disturbance rejection controller. Background Art

[0002] The undulating fin robot is a biomimetic amphibious robot. Compared to traditional propeller-driven underwater robots, it offers unique advantages such as high efficiency, low noise, and high maneuverability. It has become a cutting-edge research direction in ocean exploration, ecological monitoring, and specialized operations.

[0003] The undulating fin robot uses the MPF propulsion mode, which provides strong maneuverability and stability at low speeds. It generates thrust solely from the undulation of its fins without deforming its body, resulting in excellent stability. Furthermore, this undulating propulsion method creates minimal disturbance in the environment, offers excellent concealment, and is adaptable to complex aquatic environments. The undulating fin robot must precisely dive to a set depth while maintaining excellent stability and anti-interference capabilities.

[0004] The undulating fin robot has a large mass and generates small torque on its fins, making it a large-inertia system. Furthermore, the presence of uncertain hydrodynamic disturbances during underwater motion makes it difficult to obtain model system parameters, significantly hindering the use of many control methods based on precise system models. Error-feedback-based PID controllers are widely used in engineering. Although PID controllers have low requirements for system models, in applications such as undulating fin robots, which place higher demands on control systems, the inherent structural characteristics of traditional PID controllers make it difficult to meet system requirements. In the program, the input signal to the depth control system is a step signal, resulting in a step output signal from the PID controller. For underwater depth control, a step input signal is unreasonable, as the undulating fin robot cannot generate a step response during depth control. The ideal input signal for undulating fin robot depth control should be a signal that smoothly transitions to the target depth. Because the differential term of a PID controller is susceptible to noise interference and the integral term can reduce the system stability margin, some undulating fin robots use active disturbance rejection (ADR) controllers for underwater depth control. Classic ADR controllers primarily consist of a tracking differentiator, an extended state observer, and error feedback. However, the system parameters of the ADR controller itself are difficult to tune, and a reliable and mature tuning method is currently unavailable. The MLP structure results in limited fitting results, and better results are generally achieved by optimizing the loss function or increasing the number of neural network layers. In practical applications, due to computing power limitations, it is difficult to improve the nonlinear fitting effect of MLP by increasing the number of layers. Furthermore, the large number of parameters in ADR controllers means that the MLP deployed on actual robot platforms has limited nonlinear fitting results for all ADR controller parameters.

[0005] At the same time, ocean currents and density differences occur at any time. During snorkeling, it is difficult for the robot to maintain balance due to interference from ocean currents or the instability of its own center of gravity. The imbalance of the robot will cause deviation or even rollover during snorkeling. Therefore, stable control of the posture is also one of the control problems that must be considered for depth control during snorkeling.

[0006] Based on the above problems encountered in current practical application scenarios, the market urgently needs an underwater control method for undulating fin robots with low computing power requirements and strong anti-interference capabilities. Summary of the Invention

[0007] The present invention aims to solve at least one of the technical problems existing in the related art. To this end, the present invention provides a control method for an undulating fin robot based on a residual MLP and an active disturbance rejection controller.

[0008] A control method for an undulating fin robot based on residual MLP and active disturbance rejection controller, comprising: a) Use underwater depth control system to accurately track and control the robot's target depth; b) adopting a posture control system to control the posture of the robot and maintain the stability of the robot's posture, wherein the posture control system adopts an adjustment mechanism that integrates a PID controller and a depth factor; c) Using a fin ray control strategy to coordinate the robot's depth and posture by adjusting the swing amplitude and offset angle of the fin ray of the undulating fin; d) Residual MLP is used to optimize the parameters of depth control and posture control.

[0009] Furthermore, before executing step a), a kinematic model and a dynamic model of the undulating fin robot are established. The kinematic model describes the relationship between the position and velocity of the robot through a coordinate system definition, and the dynamic model is established based on a mechanical method, introducing the robot's mass, damping and external interference factors.

[0010] Furthermore, the underwater depth control system in step a) includes a nonlinear tracking differentiator, an extended state observer and a state error feedback controller, wherein the nonlinear tracking differentiator is used to generate a reference signal of the target depth, the extended state observer is used to estimate the robot state and interference, and the state error feedback controller generates a control quantity based on the estimated value and the reference signal.

[0011] Furthermore, the nonlinear tracking differentiator is designed using the fastest synthesis function to generate a smooth tracking signal and its differential signal; The calculation formula of the nonlinear tracking differentiator is as follows: in, For the moment, h is the underwater robot sampling period, r is the time factor, which is limited by the physical characteristics of the wave fin robot movement, z0 sets the underwater depth, z1 is the tracking signal of z0, and z2 is the differential signal of z1. () is the fastest comprehensive function proposed by Professor Han Jingqing; The extended state observer is designed based on the state space equation, and estimates the system state and external disturbance by adjusting the estimation error through the observer gain.

[0012] Furthermore, in step b), the PID controller generates a control signal according to the posture error, and the depth factor is dynamically adjusted according to the current depth and environmental parameters.

[0013] Furthermore, the depth factor is obtained by calculating the depth deviation and pressure data, and is used to adjust the feedback value of the posture control.

[0014] Furthermore, the fin ray control strategy in step c) includes adjusting the swing amplitude and offset angle of the fin ray according to the pitch angle and roll angle deviation, respectively, and introducing a depth factor to adjust the output torque.

[0015] Furthermore, the motion parameters of the fin ray are determined by calculating the amplitude base value and the deviation adjustment value, and the adjustment of the motion parameters of the fin ray includes the adjustment of the pitch angle and the roll angle.

[0016] Furthermore, the residual MLP in step d) includes multiple residual blocks, the input parameters include depth and posture related variables, and the output parameters are adjustment coefficients of depth control and posture control.

[0017] Furthermore, the training of the residual MLP is achieved by optimizing a loss function, wherein the loss function is calculated based on tracking errors of depth and pose, and the parameters are updated using a gradient descent method.

[0018] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: Improved depth control accuracy and anti-interference capabilities: A smooth reference signal is generated through a nonlinear tracking differentiator, combined with real-time ESO estimation of interference and status, effectively suppressing disturbances such as ocean currents and density changes. The depth tracking error is significantly reduced, and the response speed is increased by more than 30% compared to traditional PID and MLP-ADRC solutions.

[0019] Dynamic optimization of attitude stability: A depth factor (dev) mechanism is introduced to dynamically adjust PID parameters based on pressure data and depth deviation, enabling adaptive adjustment of large and small angle deviations. This reduces pitch and roll angle recovery time by 40% and prevents attitude reversal.

[0020] Efficient parameter optimization of residual MLP: The residual block structure alleviates the vanishing gradient problem caused by insufficient MLP depth. Nonlinear fitting of the ADRC and PID parameters is achieved under limited computing power, improving parameter tuning efficiency by more than 50% and enhancing robustness.

[0021] Innovation in collaborative control strategy: A collaborative adjustment mechanism for the fin swing amplitude and offset angle, combined with the depth factor to adjust the output torque, achieves coupled control of depth and posture, and effectively solves the problem of propulsion performance fluctuations of the undulating fin robot caused by water pressure changes.

[0022] Engineering practicality: This method takes into account both low computing power requirements and high control performance, is suitable for complex underwater environments, and provides a reliable technical solution for scenarios such as underwater detection and ecological monitoring.

[0023] Through multi-module collaboration and algorithm innovation, this invention solves the core problems of the undulating fin robot in depth and posture control, and has significant technological advancement and application value.

[0024] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 is a schematic diagram of the control coordinate system of the undulating fin robot of the present invention; Figure 2 It is a schematic diagram of the control method of the present invention; Figure 3 is a flow chart of attitude feedback control with depth factor introduced in the present invention; Figure 4 Schematic diagram of the structure of the residual block in the present invention; Figure 5 is a comparison diagram of different control schemes in the state of disturbance caused by ocean currents in the present invention; Figure 6 This is a comparison chart of the pitch angle attitude control recovery effects of the control methods before and after the depth control factor is introduced in the present invention; Figure 7 This is a comparison chart of the roll angle attitude control recovery effects of the control methods before and after the depth control factor is introduced in the present invention. DETAILED DESCRIPTION

[0027] To make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the embodiments described are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0028] 1. Modeling of the undulating fin robot system.

[0029] In order to describe the motion posture and position of the undulating fin underwater robot, the following Figure 1 The coordinate system shown. Including the fixed coordinate system O E X E Z E Y E and carrier coordinate system O B X B Z B Y B , both obey the right-hand rule. It is stipulated that the origin of the fixed coordinate system is located at the upper left vertex of the pool surface, X E Axis and Y E Axis remains horizontal, Z E The axis is perpendicular to the water surface and points downward. The origin of the carrier coordinate system is O B Select the center of gravity of the undulating fin robot, X B Axis pointing towards the robot head, Y B The axis points to the right of the robot, Z B The axis points to the bottom of the robot. The position of the undulating fin robot is expressed by P=[X,Y,Z] in the fixed coordinate system, and the posture is expressed by the Euler angle of the carrier coordinate system relative to the fixed coordinate system. Indicates heading angle X is the carrier coordinate system B The X axis in the fixed coordinate system E Y E Projection on the plane and X E The angle between the axes; pitch angle Defined as the X coordinate system of the carrier B Axis and fixed coordinate system X E Y E Angle between surfaces; roll angle Defined as the X coordinate system of the carrier B Z B Plane and X B Axis and Z E The angle between the planes parallel to the axis. E Axis, YE Axis and Z E The linear speed of the axis is expressed as In the carrier coordinate system, along the X B Axis, Y B Axis, Z B The linear velocity of the shaft is expressed as , along X in the carrier coordinate system B Axis, Y B Axis, Z B The angular velocity of the axis is expressed as .

[0030] It is stipulated that the origin of the fixed coordinate system coincides with the origin of the carrier coordinate system, and the fixed coordinate system first revolves around Z E Axis rotation , generate a new coordinate system, and then rotate the generated new coordinate system along its Y axis , generate the next coordinate system, and finally rotate the latest coordinate system along its X axis The carrier coordinate system is obtained. The rotation matrix from the carrier coordinate system to the fixed coordinate system is shown in Equation 1: Formula 1 Among them, c=cos(), s=sin(), is the heading angle, is the pitch angle, is the roll angle. We can derive formula 2: Formula 2 V1 is the X coordinate of the undulating fin robot in the carrier coordinate system. B Axis, Y B Axis, Z B The linear speed of the axis.

[0031] The derivative of the Euler angle and the angular velocity in the carrier coordinate system have the relationship shown in Equation 3: Formula 3 Among them, c=cos(), s=sin(), t=tan(), is the heading angle, is the pitch angle, is the roll angle, The undulating fin robot moves along the X axis in the carrier coordinate system. B Axis, Y B Axis, Z B The angular velocity of the axis. Not zero.

[0032] The pitch angle and yaw angle of the underwater robot are very small during underwater motion, so only the yaw angle is considered in the kinematic modeling. The kinematic model can be established as shown in Equation 4: Formula 4 Among them, c=cos(), s=sin(), t=tan(), is the heading angle, is the pitch angle, is the roll angle, They are respectively in the carrier coordinate system along X B Axis, Y B Axis, Z B The linear velocity of the axis is related to the Z axis of the carrier coordinate system. B Angular velocity of the axis.

[0033] The dynamics analysis of the undulating fin robot is carried out. According to the Lagrangian mechanics method, the dynamic model of the undulating fin robot is established as shown in Equation 5: Formula 5 in, , , , , .

[0034] They are respectively in the carrier coordinate system along X B Axis, Y B Axis, Z B The linear velocity of the axis is related to the Z axis of the carrier coordinate system. B The angular velocity of the axis. M is the mass and additional mass matrix, where In the carrier coordinate system, along X B The mass of the shaft and the added mass, is the Y direction in the carrier coordinate system B The mass of the shaft and the added mass, is the Z direction in the carrier coordinate system B The mass of the shaft and the added mass, is the Z direction in the carrier coordinate system B The mass and additional mass of the axis angular velocity, C is the Coriolis force and centripetal force matrix, and D is the linear damping matrix, where is the X direction in the carrier coordinate system B Axis damping, is the Y direction in the carrier coordinate system B Axis damping, is the Z direction in the carrier coordinate system B Axis damping, is the Z direction in the carrier coordinate system B Damping of the axis angular velocity. is the input thrust vector, Along Z B The driving force of the shaft, For along X BThe driving force of the shaft, For Z B The driving force of the shaft. is the disturbance torque, For along X B Environmental disturbance forces on the axis, Along Y B Environmental disturbance forces on the axis, Along Z B Environmental disturbance forces on the axis, For Z B The present invention focuses on depth control, so the model is further simplified as shown in Equation 6: Formula 6 in is the underwater depth, w is the coordinate system along Z B The linear speed of the axis. Along Z E Environmental disturbance forces on the axis, is the driving force along the ZE axis, is the Z direction in the carrier coordinate system B The mass of the shaft and the added mass, is the Z direction in the carrier coordinate system B Axis damping.

[0035] like Figure 2 As shown, the control system structure and control process of the present invention are shown, in which ResMLP (residual MLP) is used to optimize the control system of the robot.

[0036] 2. Underwater depth control system design.

[0037] 1) Design of nonlinear tracking differentiator The target depth value z0 is set according to the program, and the nonlinear tracking differentiator used to track it is shown in Equation 7: Formula 7 in, For the time, h is the underwater robot sampling period, r is the time factor, which is limited by the physical characteristics of the undulating fin robot motion. z0 sets the underwater depth, z1 is the tracking signal of z0, and z2 is the differential signal of z1. () is the fastest comprehensive function proposed by Professor Han Jingqing.

[0038] 2) Design of Extended State Observer Formula 6 can be rewritten as shown in Formula 8 based on the typical second-order uncertain controlled system proposed by Professor Han Jingqing: Formula 8 in, represents the environmental interference of the depth channel, Along Z E The driving force of the shaft. To control the input gain.

[0039] The state space equation of the system can be written as Equation 9 Formula 9 in, , , , .

[0040] Design the observer as Equation 10: Formula 10 represents the estimated value of the state, is the output estimate calculated from the state estimate. ,in is the observer bandwidth, is the driving force along the ZE axis.

[0041] 3. Optimization of state error feedback.

[0042] The state error feedback is designed as a PD controller using ResMLP tuning parameters.

[0043] The PD controller is shown in Equation 11: Formula 11 Where z1 is the tracking signal for setting the underwater depth, z2 is the differential signal of z1, is the observed value of underwater depth, is the observed value of z2.

[0044] 4. Optimization of attitude controller The attitude control is performed using a PID controller with parameters tuned using ResMLP.

[0045] The PID controller is shown in Equation 12: Formula 12 Where k is the kth moment, err is the angle error, and ErrI is the integral of the angle error.

[0046] Combine Figure 2 The functions of each module and signal flow in the undulating fin robot control system are explained.

[0047] Set the depth value: as the system input, pass it into the tracking differentiator to process the depth instruction (such as smoothing filtering or differential extraction).

[0048] Tracking differentiator: The output signal and feedback signal pass through the addition and subtraction nodes to form an error input to the PD controller.

[0049] PD controller: receives the proportional (KP) and differential (KD) parameters (from Res-MLP), performs proportional-differential operations on the error, and outputs the control variable.

[0050] Res-MLP: output parameters (KP, KD, KP2, KD2, KI2), Among them, KP and KD are used for PD controller, and the remaining parameters are input to the attitude feedback controller.

[0051] Attitude feedback controller: It integrates the “test attitude angle” measured by the MEMS gyroscope and the “set attitude angle” provided by the state observer, and combines the Res-MLP parameters (KP2, KD2, KI2) to generate control instructions to drive the undulating fin robot.

[0052] Undulating fin robot: executes control instructions and its state is fed back to the state observer.

[0053] State observer: estimates the system state, outputs the "set attitude angle" to the attitude feedback controller, and sends feedback signals such as depth back to the PD controller to form a closed loop.

[0054] The system achieves precise control of the depth and posture of the undulating fin robot through multi-module collaboration. Res-MLP dynamically adjusts control parameters, and the MEMS gyroscope and state observer ensure the real-time and accuracy of posture feedback.

[0055] 5. Depth factor introduced into attitude feedback control like Figure 3 As shown, a flow chart of attitude feedback control with the introduction of depth factor is shown.

[0056] An undulating fin robot cannot rely solely on gyroscope data for attitude correction. In actual operation, it must account for the varying pressures and current disturbances at varying depths, which affect fin control. At varying depths, the propulsion performance of the undulating fin may vary due to the influence of water pressure and current disturbances. For example, the stiffness and swing angle of the undulating fin may be affected by water pressure, thereby affecting propulsion. As depth increases, water pressure also increases, potentially causing structural deformation of the robot, affecting the stiffness and efficiency of the undulating fin.

[0057] Therefore, in addition to the gyroscope, a depth factor dev is also designed to detect the current ambient pressure using a pressure gauge and introduce the depth factor into the feedback process. The depth factor dev is shown in Equation 13: Formula 13 in, is the observation value of the state observer on the depth, is the set underwater depth value. pre is the current pressure value measured by the pressure gauge, The maximum pressure resistance value designed for the undulating fin robot. is the velocity gain, whose value is determined by the robot's fluid dynamic characteristics. When the angular deviation is large, the robot has a high inertia system. To achieve faster convergence to the small deviation range, a velocity factor K is introduced to enable rapid adjustment, quickly reaching the small deviation range for the attitude angle. When the attitude angle deviation is small, the depth factor is obviously a small number. This setting effectively avoids the problem of attitude reversal caused by excessive adjustment speed during large deviations. When the robot enters the small deviation phase, the inertia generated by the robot during the large deviation phase is used to assist in attitude adjustment. Furthermore, whether in the large or small deviation phase, the depth factor ensures that the robot's output torque increases with increasing ambient pressure.

[0058] The calculation of the feedback system value is obtained by formula 14, Formula 14 in, is the angular deviation, is the deviation value in the PID controller.

[0059] Combine Figure 3 The functions of each module and signal flow in the attitude control with the introduction of depth factor are explained.

[0060] Set attitude angle input: The system first has a set attitude angle as an input value.

[0061] Error calculation: The set attitude angle is compared with the actual attitude angle fed back by the MEMS gyroscope to obtain the error value err, which is then processed by the "dev" module to obtain errfb.

[0062] PID controller: The error value errfb is input to the PID controller. The PID controller calculates the error based on the three parameters: proportional (P), integral (I), and differential (D), and outputs a control signal.

[0063] Servo angle calculation: The output of the PID controller is combined with factors such as snorkeling thrust to calculate the servo angle and obtain the angle amplitude.

[0064] Undulating fin robot control: The calculated angle amplitude is used to control the posture of the undulating fin robot.

[0065] Feedback: A MEMS gyroscope monitors the undulating fin robot's actual attitude angle in real time and feeds it back to the system input, forming a closed-loop control system that continuously approaches the set attitude angle. The MEMS gyroscope, a micro-electromechanical system (MEMS) gyroscope, accurately measures attitude information such as angular velocity. The PID controller, a commonly used control algorithm in industry, effectively minimizes errors and achieves stable control.

[0066] 6. Control strategy of undulating fin rays The undulating fin robot designed in this invention has a total of 18 fins controlled by 18 servos, with 9 servos controlling each side of the undulating fin. The snorkeling gait design of the undulating fin robot is as follows: The oscillating fins on both sides of the bionic undulating fin robot swing at the same frequency, and the phase difference between the pendulums is set to zero. When the offset angle is positive, the robot will perform a vertical upward movement. During the upward movement gait, the nine pendulums on both sides flap downward simultaneously, and the net force on the robot is upward. This allows the bionic undulating fin robot to perform a vertical upward movement.

[0067] When the offset angle is negative (both pendulums swing upward simultaneously), the fins swing. During the diving gait, the nine pendulums on both sides will flap upward simultaneously, and the resultant force on the robot will be downward. At this time, the bionic undulating fin robot will perform a vertical diving motion.

[0068] When there is a deviation in the pitch angle of the robot, the undulating fin robot of the present invention can be controlled by adjusting the swing amplitude of the four fins on each side. At this time, there is no need to control the fifth fin servo in the middle because it is at the center of the robot. In order to avoid its active movement causing coupling with other fins and thus pulling the undulating fin surface material, the fifth fin is set to a slave mode and its control servo is disabled. In addition, when there is a deviation in the roll angle of the robot, the undulating fin robot of the present invention can be controlled by adjusting the swing amplitude of the fins on each side, and at the same time, a depth factor is introduced into the fin motion equation. The control equation for the adjustment is shown in Equation 15: Formula 15 When the pitch angle is positive: Flag1 is 1 when i<5 and 2 when i>5.

[0069] When the pitch angle is negative: Flag1 is 2 when i<5 and 1 when i>5.

[0070] When the roll angle is positive: Flag2 is 1 on the left fin ray and 2 on the right fin ray.

[0071] When the roll angle is negative: Flag2 is 2 on the left fin ray and 1 on the right fin ray.

[0072] in is the angle of the i-th servo starting from the robot head (i=1,2,3…9), A is the amplitude base value of the servo angle, and its value is related to the buoyancy moment output by the depth controller. Through fluid mechanics simulation calculation, the depth controller outputs the buoyancy moment The calculation formula is shown in Equation 16: Formula 16 in, is the PD controller output value, is the observation value of the state observer, is the Z direction in the carrier coordinate system B The mass of the shaft and the added mass; P is the servo angle adjustment value generated to maintain attitude balance when the pitch angle deviates, and its value is calculated by the attitude feedback control system. R is the servo angle adjustment value generated to maintain attitude balance when the roll angle deviates, and its value is calculated by the attitude feedback control system.

[0073] The collaborative control of the undulating fins on both sides enables the undulating fin robot to maintain balance during snorkeling.

[0074] 7.ResMLP Design 1) First determine the neural network structure There are seven input parameters Oin and two residual blocks, namely Res1 and Res2. Figure 4 As shown. Each residual block includes two weight layers, which are named Hid1 and Hid2. Each weight layer contains seven neurons. Res2 outputs parameters KP and KD of PD controller and KP2, KD2, KI2 of PID controller. The target depth value z0 is set, and the current observed depth value , the deviation between the tracking signal and the observation signal and the derivative of the deviation, the depth factor dev, the pitch angle and the roll angle are used as the input values of the neural network.

[0075] 2) Select the initial values of the connection weights of the hidden layer and the output layer When the neuron works in a place where the gradient of this function is small, the modification of the connection weight will also be very small, which will lead to a slow learning rate, so Xavier initialization is used.

[0076] 3) Choose an activation function Considering that the input values of the input layer and the hidden layer can be both positive and negative, we choose Tanh(x) is used as the activation function. However, the output values of the model, KP and KD, can only be positive, so the activation function of the output layer is the ReLU function.

[0077] Formula 17 Formula 18 4) Determine the input and output parameters of each layer.

[0078] Res1 weight layer 1: Res1 weight layer 2: Res1 output layer: Res2 weight layer 1: Res2 weight layer 2: Res2 output layer: 5) Complete weight update MSE is selected as the loss function and the stochastic gradient descent method is used to update the weights.

[0079] The loss function is shown in Equation 19: Formula 19 Among them, z1 is the tracking value of the target depth by the differential tracker, is the observation value of the underwater depth by the state observer, pset is the program's setting value for the pitch angle, pitch is the pitch angle currently measured by the gyroscope, rest is the program's setting value for the roll angle, and roll is the roll angle currently measured by the gyroscope.

[0080] Update the weights of each layer using Equation 20 according to the gradient descent method; Formula 20 in, is the weight of each layer at the kth moment, is the learning rate, is the partial derivative of the loss function with respect to the weight of the current layer.

[0081] Example The control method to be implemented by the present invention was simulated and verified in Matlab, and the results are as follows.

[0082] The robot and attached mass is 32 kg, the rapidity factor r = 0.01, the control cycle is designed to be 0.02 seconds, the target depth is 0.5 meters, and the initial launch attitude is 10 degrees pitch and 5 degrees roll. A random signal with a mean of 20 N and a standard deviation of 5 is applied to the robot to simulate the disturbances of ocean currents. At 20 seconds, a 300 N diving thrust is applied to the robot to simulate the sudden pressure increase caused by changes in seawater density. This thrust causes the system's roll and pitch angles to suddenly change to 20 degrees at 20 seconds.

[0083] A control system (ResMLP-ADRC) was established. In addition, a PID depth controller and an existing MLP-ADRC controller (MLP-ADRC) were established to compare their effects on underwater depth control.

[0084] At the same time, the present invention also compares the effects of the feedback and control method using the depth factor and the feedback and control method not using the depth factor in the attitude control link of the ResMLP-ADRC controller.

[0085] In depth control, by Figure 5 It can be seen that in the process of reaching the specified depth, the ResMLP-ADRC, MLP-ADRC and PID control systems can all reach the specified depth in a relatively short time without overshoot. In the face of the sudden pressure increase caused by the change in seawater density, the ResMLP-ADRC control system has the smallest depth change and the fastest callback to the set depth value, showing better anti-interference ability and faster self-adjustment ability, indicating that the MLP with the residual link can improve the stability and robustness of the control system.

[0086] In the posture control link, by comparing the effect of introducing the depth factor into the posture controller, Figure 6 and Figure 7 It can be seen that after the introduction of the depth factor, the robot will adjust the output torque in time according to the current underwater pressure and diving depth, which will improve the stability and robustness of the system and accelerate the system response speed. For large attitude angle errors and small attitude angle errors, the feedback system with the introduction of the depth factor will flexibly adjust the output torque, thereby accelerating the response speed and stability of the system.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A control method for a wavy fin robot based on residual MLP and active disturbance rejection controller, characterized in that: include: a) Use underwater depth control system to accurately track and control the robot's target depth; b) adopting a posture control system to control the posture of the robot and maintain the stability of the robot's posture, wherein the posture control system adopts an adjustment mechanism that integrates a PID controller and a depth factor; c) Using a fin ray control strategy to coordinate the robot's depth and posture by adjusting the swing amplitude and offset angle of the fin ray of the undulating fin; d) Residual MLP is used to optimize the parameters of depth control and posture control.

2. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 1 is characterized in that: Before executing step a), a kinematic model and a dynamic model of the undulating fin robot are established. The kinematic model describes the relationship between the position and velocity of the robot through a coordinate system definition, and the dynamic model is established based on a mechanical method, introducing the robot's mass, damping and external interference factors.

3. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 1 is characterized in that: The underwater depth control system in step a) includes a nonlinear tracking differentiator, an extended state observer, and a state error feedback controller, wherein the nonlinear tracking differentiator is used to generate a reference signal of the target depth, the extended state observer is used to estimate the robot state and interference, and the state error feedback controller generates a control variable based on the estimated value and the reference signal.

4. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 3 is characterized in that: The nonlinear tracking differentiator is designed with the fastest synthesis function to generate a smooth tracking signal and its differential signal; The calculation formula of the nonlinear tracking differentiator is as follows: in, For the moment, h is the sampling period of the underwater robot, r is the time factor, which is limited by the physical characteristics of the wave fin robot movement, z0 sets the underwater depth, z1 is the tracking signal of z0, and z2 is the differential signal of z1. () is the fastest comprehensive function; The extended state observer is designed based on the state space equation, and estimates the system state and external disturbance by adjusting the estimation error through the observer gain.

5. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 1 is characterized in that: In step b), the PID controller generates a control signal according to the attitude error, and the depth factor is dynamically adjusted according to the current depth and environmental parameters.

6. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 5 is characterized in that: The depth factor is obtained by calculating the depth deviation and pressure data, and is used to adjust the feedback value of the posture control.

7. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 1 is characterized in that: The fin ray control strategy in step c) includes adjusting the swing amplitude and offset angle of the fin ray according to the pitch angle and roll angle deviation, and introducing a depth factor to adjust the output torque.

8. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 7 is characterized in that: The motion parameters of the fin ray are determined by calculating the amplitude base value and the deviation adjustment value, and the adjustment of the motion parameters of the fin ray includes the adjustment of the pitch angle and the roll angle.

9. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 1 is characterized in that: The residual MLP in step d) includes multiple residual blocks, the input parameters include depth and posture related variables, and the output parameters are adjustment coefficients for depth control and posture control.

10. The control method of the undulating fin robot based on residual MLP and active disturbance rejection controller according to claim 9 is characterized in that: The training of the residual MLP is achieved by optimizing the loss function, which is calculated based on the tracking error of depth and pose, and the parameters are updated using the gradient descent method.

Images