A servo controller based on combination of fuzzy logic and model predictive control

Abstract

The application discloses a kind of servo controller based on fuzzy logic and model predictive control combination, comprising: image acquisition module, for real-time capture image information and send to image processing and state estimation module;Image processing and state estimation module, for target detection and pre-processing to image information, extract target current position, and establish target-servo system's simplified kinematics model, calculate position error and speed error;Fuzzy control module, for dynamically adjusting the reference weight value of MPC according to position error and speed error;MPC optimization solution and control output module, for constructing MPC cost function according to the reference weight value of adjustment, optimal control amount is solved by quadratic programming, and control amount is converted into the angular acceleration instruction output executable by servo cloud platform module.The application realizes the adaptive control to target motion state, improves the stability and precision of system in complex target tracking task.

Description

Technical Field

[0001] This invention belongs to the field of automatic control and target tracking technology, specifically relating to a servo controller based on a combination of fuzzy logic and model predictive control, used to achieve high-precision real-time tracking of dynamic targets. Background Technology

[0002] Dynamic target tracking is a key technology in fields such as robotics, unmanned systems, and industrial automation. Servo control systems need to adjust actuators (such as gimbals and robotic arms) in real time to align with moving targets. Due to the nonlinear, abrupt, and unpredictable nature of target motion, traditional control methods (such as PID control) are prone to large tracking errors and system oscillations when dealing with sudden changes in target position and velocity. Furthermore, their response to external disturbances and noise is relatively passive, making it difficult to meet the requirements of high-precision tracking.

[0003] Model Predictive Control (MPC), as an advanced control method, has been introduced into the field of target tracking control. MPC predicts future outputs using a system model and solves the optimization problem online to obtain the optimal control input. However, traditional MPC controllers typically use fixed weight parameters, which limits their robustness and flexibility when dealing with highly complex or rapidly changing target motions. They cannot adaptively adjust the control strategy according to changes in the target's motion state, leading to a decline in tracking performance when the target moves at high speeds or experiences sudden trajectory changes.

[0004] Therefore, a high-performance servo controller that can adaptively adjust control parameters and adapt to complex target motion is needed. Summary of the Invention

[0005] To overcome the shortcomings of existing target tracking control methods, this invention proposes a real-time target tracking servo controller based on fuzzy model predictive control (F-MPC) to achieve adaptive control of the target motion state and improve the stability and accuracy of servo control in complex target tracking tasks.

[0006] The technical solution to achieve the purpose of this invention is as follows:

[0007] A servo controller based on a combination of fuzzy logic and model predictive control, comprising:

[0008] The image acquisition module is used to capture image information in real time and send it to the image processing and state estimation module;

[0009] The image processing and state estimation module is used to perform target detection and preprocessing on image information, extract the current position of the target, establish a simplified kinematic model of the target-servo system, and calculate the position error and velocity error.

[0010] The fuzzy control module is used to dynamically adjust the baseline weight value of MPC based on position error and velocity error;

[0011] The MPC optimization solution and control output module is used to construct the MPC cost function based on the adjusted baseline weight values, solve for the optimal control quantity through quadratic programming, and convert the control quantity into angular acceleration commands that can be executed by the servo gimbal module.

[0012] The significant advantages of this invention compared to existing technologies are:

[0013] (1) A weight online adaptive adjustment mechanism combining fuzzy control and MPC is proposed. Through the fuzzy control module, position error and velocity error are used as inputs, and weight adjustment coefficients are output in real time according to fuzzy rules to dynamically correct the state weight matrix and control weight matrix in the MPC cost function. Compared with traditional MPC with fixed weights, this invention enables the controller to adaptively cope with various complex motion modes such as step jumps, high-speed motion, and sinusoidal trajectories, significantly improving tracking accuracy and system robustness.

[0014] (2) A simplified kinematic model of the target-servo system is constructed, balancing accuracy and real-time performance. This invention establishes a state space vector and discrete state transition equations, which retains the key motion characteristics of the servo system while avoiding the solution burden caused by complex nonlinear models. Combined with the MPC cost function, real-time solutions can be completed within millisecond-level sampling periods, achieving a balance between high-precision tracking and real-time performance.

[0015] (3) The superior performance was demonstrated through dual verification using numerical simulation and Coppeliasim co-simulation. The F-MPC controller was comprehensively tested for step and sinusoidal signals in numerical simulation, and its actual tracking performance under linear and sinusoidal trajectories was verified in Coppeliasim. The results show that the F-MPC controller outperforms the traditional N-MPC in terms of response time, steady-state error, and maximum overshoot, effectively suppressing overshoot and oscillation.

[0016] (4) The F-MPC controller is suitable for various scenarios that require dynamic target tracking, such as robot vision tracking, UAV ground target tracking, and security monitoring. Attached Figure Description

[0017] Figure 1 This is a simplified kinematic model diagram of the servo system and the target in this invention;

[0018] Figure 2 This is a schematic diagram illustrating the definition of the target location in an image in this invention;

[0019] Figure 3 This is a control flowchart of the fuzzy MPC controller in this invention;

[0020] Figure 4 This is a membership function diagram of the position error in this invention;

[0021] Figure 5 This is a membership function diagram of the velocity error in this invention;

[0022] Figure 6 The position and velocity error weighting adjustment coefficients in this invention Membership function graph;

[0023] Figure 7 The control weight adjustment coefficient in this invention Membership function graph;

[0024] Figure 8 This is a comparison diagram of the tracking effect of a step signal with an amplitude of 1 in an embodiment of the present invention;

[0025] Figure 9 This is a comparison diagram of the tracking effect of a step signal with amplitude 3 in an embodiment of the present invention;

[0026] Figure 10 This is a comparison diagram of the tracking effect of a step signal with an amplitude of 5 in an embodiment of the present invention;

[0027] Figure 11 This is a comparison diagram of the tracking effect of a sinusoidal signal with a period of 10 seconds and an amplitude of 1 in an embodiment of the present invention;

[0028] Figure 12 This is a comparison diagram of the tracking effect of a sinusoidal signal with a period of 5 seconds and an amplitude of 1 in an embodiment of the present invention;

[0029] Figure 13 This is a comparison diagram of the tracking effect of a sinusoidal signal with a period of 5 seconds and an amplitude of 2 in an embodiment of the present invention;

[0030] Figure 14 This is a flowchart of the simulation verification process based on Coppeliasim in this invention;

[0031] Figure 15 This is a diagram showing the tracking error of the left and right boundaries of the identification frame for linear motion at 2 m / s in an embodiment of the present invention.

[0032] Figure 16 This is a diagram showing the tracking error of the upper and lower boundaries of the identification frame for linear motion at 2 m / s in an embodiment of the present invention.

[0033] Figure 17 This is a diagram showing the tracking error of the left and right boundaries of the identification frame for linear motion at 5 m / s in an embodiment of the present invention.

[0034] Figure 18 This is a diagram showing the tracking error of the upper and lower boundaries of the identification frame for linear motion at 5 m / s in an embodiment of the present invention.

[0035] Figure 19 This is a diagram showing the tracking error of the left and right boundaries of the identification frame for a sinusoidal motion at 2 m / s in an embodiment of the present invention.

[0036] Figure 20 This is a diagram showing the tracking error of the upper and lower boundaries of the identification frame for a sinusoidal motion at 2 m / s in an embodiment of the present invention.

[0037] Figure 21 This is a comparison diagram of the trajectory tracking effect of a 2 m / s sinusoidal motion in an embodiment of the present invention;

[0038] Figure 22 This is a diagram showing the tracking error of the left and right boundaries of the identification frame for a 5 m / s sinusoidal motion in an embodiment of the present invention.

[0039] Figure 23 This is a diagram showing the tracking error of the upper and lower boundaries of the identification frame for a 5 m / s sinusoidal motion in an embodiment of the present invention.

[0040] Figure 24 This is a comparison diagram of the trajectory tracking effect of a 5 m / s sinusoidal motion in an embodiment of the present invention. Detailed Implementation

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0042] The present invention provides a real-time target tracking servo controller based on fuzzy model predictive control, comprising a controller hardware system and a controller software system.

[0043] The controller hardware system includes: an image acquisition module for capturing image information in real time and sending it to the embedded main control module; the embedded main control module for performing target detection on the captured image information, running the corresponding algorithm of the controller software system, and outputting control quantities; and a servo gimbal module, including an azimuth motor and a pitch motor, for driving the gimbal to rotate according to the control quantities output by the embedded main control module.

[0044] The controller software system running on the embedded main control module includes: an image processing and state estimation module, used for target detection and preprocessing of image information, extracting the current position of the target, establishing a simplified kinematic model of the target-servo system, and calculating position and velocity errors; a fuzzy control module, used for dynamically adjusting the weight parameters of the MPC based on the position and velocity errors; and an MPC optimization and control output module, used for constructing the MPC cost function based on the adjusted weight parameters, solving for the optimal control quantity through quadratic programming, and converting the control quantity into angular acceleration commands that can be executed by the servo gimbal module.

[0045] (a) Image Processing and State Estimation Module

[0046] To establish a simplified kinematic model of the target-servo system, a spatial rectangular coordinate system is established with the rotation center of the servo gimbal as the origin, and the Z-axis points towards the target. The complex motion of the target is simplified within the control cycle to a distance along the Z-axis from the servo gimbal. The motion of the meter plane relative to the gimbal, such as Figure 1 As shown. This controller only considers the target's displacement in the lateral and longitudinal directions, that is, the displacement in the X-axis and Y-axis directions.

[0047] The image plane coordinate system is defined with the bottom border as the X-axis and the left border as the Y-axis. The desired position (crosshair) is taken from the image center. This module first preprocesses the image acquired by the image acquisition module and identifies the target using a target detection algorithm. The target usually appears in the form of a bounding box, and the center of the bounding box is taken as the current position of the target. Figure 2 As shown.

[0048] The discretized state-space equation is shown in equation (1):

[0049] (1)

[0050] in, These are the state space vectors at time k+1 and k, respectively. Let A be the control input vector at time k (to be determined), A be the state transition matrix, and B be the control input matrix, specifically in the form of:

[0051] , ,

[0052] These are the acceleration values ​​of the crosshair along the x and y directions in the image plane, driven by a servo motor, with a sampling time interval of t.

[0053] The state-space vector is defined as shown in equation (2):

[0054] (2)

[0055] in, It is the distance from the target to the servo gimbal along the Z-axis; The desired coordinates of the target in the image plane; These are the current coordinates of the target's position in the image plane; It is the angular velocity of the servo gimbal around the Y-axis and X-axis; It is the actual velocity of the target in the x and y directions in the image plane; These are the positional errors of the target in the x and y directions in the image plane, respectively. These are the velocity errors of the target in the x and y directions in the image plane, respectively.

[0056] (ii) Fuzzy control module

[0057] Traditional MPC obtains the optimal control quantity by solving the cost function, but its fixed weight coefficients Q and R are difficult to adapt to complex and ever-changing target motion. Therefore, fuzzy control is introduced to adjust the weight coefficients in real time. Figure 3 As shown, the fuzzy control process includes four steps: fuzzification, fuzzy inference, defuzzification, and weight adjustment.

[0058] (1) Fuzzification: Transform the input variables into a fuzzy set and define its membership function;

[0059] (2) Fuzzy reasoning: Using a fuzzy rule base to perform logical reasoning on fuzzy input variables to generate fuzzy output;

[0060] (3) Defuzzification: converting the fuzzy output into a specific numerical form for actual control.

[0061] (4) Weight adjustment: The weight matrix of MPC is dynamically adjusted based on the fuzzy inference results.

[0062] The input variable is the normalized position error. and speed error Position error is defined as the pixel difference between the actual and desired coordinates of the target in the image, normalized to the [0,1] interval; velocity error is defined as the difference between the target velocity and the crosshair velocity, with a normalization range set according to the actual scene. Output variables include position error weight adjustment coefficients. Speed ​​error weighting adjustment coefficient and control quantity weight adjustment coefficient .

[0063] To transform precise input quantities into fuzzy quantities suitable for fuzzy inference, the input variables need to be divided into multiple fuzzy levels, each corresponding to a fuzzy set and its membership function. Position error is divided into three fuzzy levels (level 1, 2, and 3), corresponding to low, medium, and high error levels, respectively; velocity error is divided into five fuzzy levels (levels 1 to 5), corresponding to five error levels from extremely low to extremely high; output weight adjustment coefficients... , and Similarly, it is divided into three fuzzy levels (Level 1, Level 2, and Level 3), corresponding to low, medium, and high adjustment ranges, respectively. All membership functions adopt a combination of triangular and trapezoidal functions. The membership function can be used to calculate the membership degree of the input value to each fuzzy level, providing a basis for subsequent fuzzy inference.

[0064] The fuzzy controller uses a fuzzy rule base to dynamically map input variables to output weights, adapting to changes in position and velocity during target motion. The fuzzy rule base is designed based on the practical needs of target tracking, combining experimental data and theoretical analysis to define... , right , and The logic of influence. When When the value is large, prioritize increasing. The weighting is adjusted to ensure the target is quickly aligned; when When the value is large, priority should be given to increasing it. The weighting enhances the system's dynamic response capability to rapidly changing targets; while and When both are relatively small, increase appropriately. The weighting of controls reduces the frequency of adjustments to control actions, ensuring the smoothness and stability of the system.

[0065] The designed fuzzy rule base is shown in Table 1. It employs Mamdani-type fuzzy inference, and defuzzification uses the centroid method. Level 3 and Taking level 2 as an example, its output Level 3 Level 2 Level 1.

[0066] Table 1 Fuzzy Rule Base

[0067]

[0068] Fuzzy control requires separate inputs for the x and y directions. and The processing flow is the same for the position error weight adjustment coefficient, velocity error weight adjustment coefficient, and control quantity weight adjustment coefficient corresponding to their respective directions, thus yielding the following weight adjustment formula:

[0069] (3)

[0070] These are the position error weighting adjustment coefficient, velocity error weighting adjustment coefficient, and control quantity weighting adjustment coefficient in the x-direction, respectively. These are the position error weighting adjustment coefficient, velocity error weighting adjustment coefficient, and control quantity weighting adjustment coefficient in the y-direction, respectively. For the updated The values ​​at the corresponding positions in the original Q and R matrices need to be replaced by the adjusted Q and R matrices. The adjusted Q and R matrices are then substituted back into the cost function of MPC (Equation (4)) for optimization in the next cycle.

[0071] (III) MPC Optimization Solution and Control Output Module

[0072] The constructed cost function is:

[0073] (4)

[0074] The cost function is used to design the control input to minimize position and velocity errors, making the system state approach the ideal tracking state (i.e., zero error). These are the state space vectors at steps i and N, respectively. Let represent the control input vector at time i, T represent the transpose of the vector, N be the prediction step size, Q be the state weight matrix used to measure the cost of position and velocity errors, R be the control weight matrix used to limit the cost of control input, and F be the terminal cost matrix.

[0075]

[0076]

[0077]

[0078] These represent the baseline weight values ​​for the position and velocity errors of the target in the image plane in the x-direction and y-direction, respectively, from step 0 to N-1. These represent the baseline weight values ​​for the position and velocity errors of the target in the image plane in the x-direction and the y-direction, respectively, at step N. These represent the baseline weight values ​​for the acceleration control quantities in the x and y directions, respectively. and The fuzzy control module needs to update the descendant matrix by equation (3) to update the Q and R matrices.

[0079] By extending the prediction of the state in the next N steps, we define:

[0080] (5)

[0081] Matrix X is a stacked vector of state space vectors for the next N steps:

[0082] Matrix U is a stacked vector of control inputs for the next N steps:

[0083] Matrix M is composed of powers of the state transition matrix and is used to predict the influence of the initial state on future states:

[0084] Matrix C describes the effect of the control input on the future state:

[0085]

[0086] Rewriting the cost function of equation (4) in matrix form, and substituting equation (5) into the cost function, we get:

[0087]

[0088] In the formula, , const represents the constant part in the objective function, which can be ignored during optimization. The state cost of the first 0 to N-1 steps is represented by the Q matrix, and the state cost of the Nth step (the last step in the prediction time domain) is represented by the F matrix. It is the extended cost matrix of the control input:

[0089]

[0090]

[0091] Define the coefficient vector of the first-order term. The cost function is rewritten in quadratic programming form:

[0092] (6)

[0093] Solving the constrained optimization problem using quadratic programming yields the optimal control sequence U, and the first control input is selected. Execution for the current moment, Mapped to the angular acceleration values ​​of the two motors of the servo gimbal module The new angular velocity value is obtained by accumulating the values. Then, the angular velocity value is set as the target speed of the motor to update the attitude of the servo gimbal and complete one control cycle.

[0094] In practical applications, because the gimbal has a certain moment of inertia and the motor power is limited, the angular acceleration of the motor is... angular velocity All of these must satisfy certain constraints. Therefore, when finding the minimum value of the above quadratic form, the boundary values ​​of angular acceleration and angular velocity must also be included:

[0095] (7)

[0096] They are respectively The minimum and maximum values ​​that are allowed to be reached. They are respectively The minimum and maximum values ​​that are allowed to be reached.

[0097] Example 1

[0098] The described embodiments are only some embodiments of the present invention, and not all embodiments.

[0099] Example: A real-time target tracking servo controller based on fuzzy model predictive control

[0100] I. Hardware System Implementation

[0101] The controller hardware system in this embodiment includes:

[0102] (1) Image acquisition module: adopts CMOS image sensor, image sampling frame rate > 30 frames;

[0103] (2) Embedded main control module: Using an edge AI computing development platform, the 1080P image is cropped into a 400×400 pixel image, the target position is identified and extracted, and the F-MPC control algorithm is run.

[0104] (3) Servo gimbal module: Servo motors are used to form a two-degree-of-freedom gimbal (azimuth motor and pitch motor).

[0105] II. Software System Implementation

[0106] (I) System Modeling and State Estimation Module

[0107] like Figure 2 As shown, the target location is represented by the center of the recognition box. The image coordinates are defined as follows: the bottom border is the X-axis, with positive to the right; the left border is the Y-axis, with positive upward. The current coordinates of the target are... :

[0108] (8)

[0109] In the formula, These represent the coordinates of the top-left and bottom-right corners of the recognition box, respectively. (Crosshai coordinates) The width and height of the image are each half of the total number of pixels, i.e., (200, 200). Using the state space equations and state space vectors shown in equations (1) and (2), with a sampling time interval t of 0.04, we substitute them into matrices A and B.

[0110] (ii) Fuzzy control module

[0111] Input variable normalization: position error The ratio of the target pixel difference to the maximum pixel difference is used. In this example, the maximum pixel difference is set to 200, and the normalization range is [0,1]; speed error. The ratio of the target speed difference to the maximum speed difference is given in this example, where the maximum speed difference is set to 20 m / s and the normalization range is [0, 2000].

[0112] Membership function design, such as Figure 4-7 As shown, the position error is divided into three fuzzy sets with ranges of [0,0.1], [0.1,0.4], and [0.4,1] for levels 1, 2, and 3, respectively, using triangular, trapezoidal, and trapezoidal membership functions; the velocity error is divided into five fuzzy sets with ranges of [0,50], [30,200], [150,500], [400,1000], and [800,2000] for levels 1, 2, 3, 4, and 5, all using trapezoidal membership functions. The output weights are divided into three fuzzy set ranges of level 1, level 2, and level 3, with ranges of [0.5, 0.9], [0.8, 1.2], and [1.1, 1.5], respectively, using triangular, trapezoidal, and triangular membership functions. The output weights are divided into three fuzzy sets of level 1, level 2, and level 3, with ranges of [0.1, 0.3], [0.2, 0.6], and [0.6, 1], respectively, using triangular, trapezoidal, and triangular membership functions.

[0113] The fuzzy rule base is shown in Table 1. Defuzzification is performed using the centroid method. The rule "..." is used as an example. Level 3 and Taking "Level 2" as an example, its output They are classified as Level 3 (adjustment coefficient 1.2–1.5) and Level 2 (adjustment coefficient 0.8–1.2). Level 1 (adjustment coefficient 0.1-0.3). The weight adjustment formula is equation (3), which is obtained by defuzzification. Multiplying the weights by the baseline weights yields the new weight matrices Q and R.

[0114] (III) MPC Optimization Solution Module

[0115] Using the cost function shown in equation (4), this embodiment sets the prediction step size N to 5, and the baseline value of the state weight matrix is:

[0116] , ,

[0117] After the fuzzy control module updates the Q and R matrices, the optimal control sequence U is obtained by solving equation (6) using quadratic programming, and the first control input is taken. Used for execution at the current moment.

[0118] III. Simulation Verification

[0119] (I) Numerical simulation verification

[0120] This embodiment sets the total duration to 20 seconds, the sampling interval to 0.04 seconds, the prediction step size N=5, and the upper limit of control to 0.1 for numerical simulation. Step signals with amplitudes of 1, 3, and 5, and sinusoidal signals with periods of 5 and 10 seconds and amplitudes of 1 and 2, are tracked respectively. The results are as follows: Figure 8-13 As shown in Table 2.

[0121] Table 2 Comparison of Numerical Simulation Results

[0122]

[0123] As shown in Table 2, under large-amplitude step signals (amplitudes 3 and 5), F-MPC not only completely eliminates the overshoot present in traditional MPC, but also reduces the response time by more than 30%. In the most challenging high-frequency large-amplitude sinusoidal signal (period 5s / amplitude 2) tracking, the mean square error of F-MPC is reduced by 98.3% compared to traditional N-MPC, demonstrating extremely strong nonlinear tracking capability.

[0124] (II) Coppeliasim co-simulation verification

[0125] Simulation process as follows Figure 14 As shown, Coppeliasim is used as the server and Python as the client, with data interaction achieved through the API. In the simulation environment, a 3D model of the servo gimbal system is imported, and virtual camera parameters are set. The target moves back and forth along a straight trajectory and a sinusoidal trajectory at speeds of 2 m / s and 5 m / s respectively. The total length of the straight trajectory is 40 meters, and the distance between the beginning and end of the sinusoidal trajectory is 50 meters, with an amplitude of 10 meters and a cycle length of 12.5 meters. The results are as follows. Figure 15-24 As shown:

[0126] 1. During linear motion at 2 m / s, pixel errors decay rapidly and tend to stabilize, while steady-state errors remain at a low level. Figure 15-16 );

[0127] 2. In linear motion at 5 m / s, despite a large initial error, F-MPC can still effectively reduce steady-state error and avoid overshoot and oscillation. Figure 17-18 );

[0128] 3. During sinusoidal motion at 2 m / s, the pixel error exhibits periodic fluctuations, but it maintains a low error at the extreme points of the trajectory. Figures 19-20 ), trajectory tracking, for example Figure 21 As shown;

[0129] 4. During sinusoidal motion at 5 m / s, the F-MPC's collimator trajectory matches the target trajectory well. Although there is a slight lag at trajectory inflection points, it can quickly approach the target trajectory and restore consistency. Figure 22-24 ).

Claims

1. A servo controller based on a combination of fuzzy logic and model predictive control, characterized in that, include: The image acquisition module is used to capture image information in real time and send it to the image processing and state estimation module; The image processing and state estimation module is used to perform target detection and preprocessing on image information, extract the current position of the target, establish a simplified kinematic model of the target-servo system, and calculate the position error and velocity error. The fuzzy control module is used to dynamically adjust the baseline weight value of MPC based on position error and velocity error; The MPC optimization solution and control output module is used to construct the MPC cost function based on the adjusted baseline weight values, solve for the optimal control quantity through quadratic programming, and convert the control quantity into angular acceleration commands that can be executed by the servo gimbal module.

2. The servo controller according to claim 1, characterized in that, The optimal control quantity is the first control input of the optimal control sequence U obtained by solving a quadratic programming problem, where the quadratic programming form is: The coefficient vector of the first term ; ; In the formula matrix ; is the extended cost matrix of the control weight matrix R; C is the matrix describing the influence of the control input on the future state; matrix M is composed of powers of the state transition matrix. It is the control input vector at time k; It is the state-space vector at time k; matrix The state cost for the first 0 to N-1 steps is represented by matrix Q, and the state cost for the Nth step is represented by matrix F. Q is the state weight matrix, whose elements are the baseline weight values ​​of the target's position error and velocity error in the x-direction and y-direction in the image plane during steps 0 to N-1. F is the terminal cost matrix, whose elements are the baseline weight values ​​of the target's position error and velocity error in the x-direction and y-direction in the image plane during step N. The elements of the control weight matrix R are the baseline weight values ​​of the acceleration control quantities in the x and y directions.

3. The servo controller according to claim 1, characterized in that, The input variables of the fuzzy control module are the normalized position and velocity errors in the x and y directions, respectively. The output variables are the position error weight adjustment coefficient, velocity error weight adjustment coefficient, and control quantity weight adjustment coefficient for each of the two directions. The reference weight values ​​are adjusted using the following formula: In the formula, These are the position error weighting adjustment coefficient, velocity error weighting adjustment coefficient, and control quantity weighting adjustment coefficient in the x-direction, respectively. These are the position error weighting adjustment coefficient, velocity error weighting adjustment coefficient, and control quantity weighting adjustment coefficient in the y-direction, respectively. For the updated , These represent the baseline weight values ​​for the target's position and velocity errors in the x-direction and y-direction of the image plane at steps 0 to N-1, respectively. These represent the baseline weight values ​​for the acceleration control quantities in the x and y directions, respectively.

4. The servo controller according to claim 1, characterized in that, The fuzzy control module achieves dynamic mapping between input and output variables through a fuzzy rule base. Input variables are divided into multiple fuzzy levels, each level corresponding to a fuzzy set and its membership function. Position error is divided into three fuzzy levels, velocity error into five fuzzy levels, and output variables into three fuzzy levels. Mamdani-type fuzzy inference is adopted, and centroid method is used for defuzzification.

5. The servo controller according to claim 2, characterized in that, The state space vector at time k is: The control input vector at time k is: in These represent the acceleration values ​​of the crosshair along the x and y directions in the image plane under the drive of the servo motor. It is the distance from the target to the servo gimbal along the Z-axis; The desired coordinates of the target in the image plane; These are the current coordinates of the target's position in the image plane; It is the angular velocity of the servo gimbal around the Y-axis and X-axis; It is the actual velocity of the target in the x and y directions in the image plane.

6. The servo controller according to claim 1, characterized in that, The image acquisition module uses a CMOS image sensor.

Images