Photoelectric stabilized platform composite anti-interference method based on improved active-disturbance-rejection control

Abstract

The invention belongs to the technical field of photoelectric stabilized platform intelligent control, and particularly relates to a photoelectric stabilized platform composite anti-interference method based on improved active-disturbance-rejection control, which realizes multi-band disturbance cooperative suppression through an improved extended state observer (ESO) and non-linear state error feedback (NLSEF) core architecture. For the problems of phase distortion and parameter sensitivity under traditional ADRC high-frequency disturbance, an ALMRBF-ESO observer is constructed by adopting a dynamic regularization constrained radial basis function neural network and an adaptive damping adjustment mechanism, gradient dispersion limitation of a traditional RBF in a high-frequency domain is broken through, and disturbance root-mean-square errors are reduced; by reconstructing a sliding mode gain equation of nonlinear tracking error feedback and implanting a parameter self-correction mechanism, dynamic response of a system is accelerated, overshoot is reduced at the same time, and the parameter drift rate of a controller is stabilized at 0.12. According to the method, on the basis of completely reserving ADRC robustness, the problem of collaborative optimization of ESO broadband disturbance observation precision and NLSEF dynamic tracking performance is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of optoelectronic stabilization platforms, and specifically to a composite disturbance rejection method for optoelectronic stabilization platforms based on an improved active disturbance rejection control. Background Technique

[0002] As the core carrier of modern precision optical systems, optoelectronic stabilization platforms play an irreplaceable role in fields such as aerospace remote sensing detection, attitude maintenance of intelligent ground mobile platforms, and high-precision industrial automation detection. The essence of this system is to isolate the interference of complex disturbances such as carrier vibration and external electromagnetic pulses on optical equipment through a multi-dimensional motion compensation mechanism. However, in practical applications, it faces three typical disturbance environments: low-frequency large-inertia random motion of the carrier such as ship roll and UAV attitude changes, medium- and high-frequency broadband electromechanical noise covering motor harmonics and bearing vibration transmission, and high-frequency micro-amplitude parameter drift phenomena such as temperature deformation and circuit stray capacitance effects. These disturbances exhibit multi-scale separation characteristics in the frequency spectrum, and are significantly time-varying and strongly coupled. Traditional single-frequency point compensation strategies lack a full-frequency band collaborative adjustment mechanism, resulting in deterioration of line-of-sight stability and attenuation of target acquisition accuracy.

[0003] In the existing control system, although the PID algorithm based on error integral suppression and its improved schemes dominate industrial applications, their linear adjustment mechanism has inherent contradictions in suppressing multi-frequency disturbances. The integral link is prone to causing the control signal to oscillate and diverge due to its sensitivity to high-frequency noise, while the phase lag characteristic of the differential module weakens the acceleration disturbance compensation ability. These two defects lead to a threshold interlocking effect between the overshoot and the dynamic error. The intelligent compensator using a radial basis function neural network attempts to improve the disturbance estimation accuracy through non-linear mapping, but the fixed gain characteristic of the hidden layer nodes cannot adapt to the disturbance frequency domain migration law, and there is still a phase compensation blind area under broadband composite disturbances. The control algorithm based on the sliding mode variable structure idea has strong robustness, but the fixed threshold design of its boundary layer will cause double-mode instability problems of high-frequency oscillation and phase margin decline under high-frequency disturbance excitation, and in severe cases, it will cause mechanical resonance of the actuator.

[0004] In view of the above technical limitations, the present invention breaks through the combination constraints of traditional control paradigms and constructs a collaborative architecture of nonlinear observation and multi-modal compensation with frequency-domain decoupling ability: by integrating an improved dynamic weight radial basis function neural network and an adaptive switching sliding mode control strategy, a spectrum decoupled composite disturbance rejection system is formed. At the observation level, a spectral coding activation dynamic weight matrix optimization method is adopted to realize the separation and extraction of multi-frequency disturbances of electromechanical noise harmonics and temperature drift parameters; at the compensation level, a nonlinear sliding mode control law with adaptive gain is designed, and the compensation gain curve is dynamically switched based on the characteristics of disturbance energy distribution to achieve energy consumption balance with overshoot suppression and high-frequency oscillation elimination. This method reduces the line-of-sight steady-state error to the order of 5.3 μrad under the wide-frequency disturbance condition of 155 Hz - 1.2 kHz, effectively solving the problem of control performance degradation caused by frequency-domain parameter coupling of traditional algorithms. Summary of the Invention

[0005] (1) Technical problems to be solved

[0006] Aiming at the common technical defects of the existing methods, such as insufficient observation accuracy of high-frequency disturbances, coupling contradictions between disturbance rejection dynamic response and chattering suppression, and steady-state breakdown caused by parameter drift, the present invention proposes a composite disturbance rejection method for an optoelectronic stabilization platform based on improved active disturbance rejection control, breaking through the performance bottleneck of the traditional compensation framework.

[0007] (2) Technical solutions

[0008] The present invention specifically adopts the following technical solutions to achieve the above objectives:

[0009] A composite disturbance rejection method for an optoelectronic stabilization platform based on improved active disturbance rejection control, including a dual extended state observation module, a dynamic optimization feedback control module, and a collaborative disturbance rejection execution module, comprising the following steps;

[0010] S1. Deployment of an improved extended state observer:

[0011] The dual extended state observation module is to construct an improved extended state observer (ALMRBF-ESO) by integrating a dynamic regularization radial basis neural network and an adaptive damping factor adjustment mechanism within the framework of active disturbance rejection control, performing high-frequency phase compensation and low-frequency gain tuning on the multi-frequency disturbance signals received by the platform, and achieving the suppression of 10 Hz - 2 kHz disturbance observation errors;

[0012] S2. Implementation of a dynamic feedback control law:

[0013] The dynamic optimization feedback control module reconstructs the control law equation based on the sliding mode gain nonlinear tracking error feedback characteristic (SMNLSEF), dynamically adjusts the slope of the response curve through a variable structure exponential reaching law, and constructs a spectral energy functional constraint by integrating a dual error suppression function to achieve the phase-frequency error self-correction of the control quantity output under high-frequency disturbance excitation;

[0014] S3. Composite disturbance rejection signal synthesis:

[0015] Generate a composite control signal by fusing the dynamic weights of the ESO disturbance feedforward and the NLSEF feedback output, complete the broadband disturbance error convergence within 2 ms, reduce the system steady-state overshoot to 5.3%, and the parameter drift rate is lower than 0.12% / h.

[0016] Furthermore, the ALMRBF-ESO adopts a spectral coding reconstruction observation channel, maps the input disturbance signal to a preset frequency domain feature space through the RBF basis function, weights and separates the disturbance energy of different frequency bands in the hidden layer nodes, and combines the Jacobian matrix correction term

[0017] J reg = J + λ w W + λ c C + λ b B to iteratively optimize the parameter update direction to suppress the parameter drift problem caused by high-frequency disturbances.

[0018] Furthermore, the adaptive L2 regularization constraint of the ALMRBF-ESO optimization algorithm, its composite error function is defined as:

[0019]

[0020] In the formula, λ w , λ c , λ b are the regularization strength coefficients of the weight, kernel center, and width respectively, and balance the parameter update rate and global convergence stability through a dynamic normalization strategy.

[0021] Furthermore, the adaptive adjustment rule of the damping factor in the ALM-RBFESO optimization algorithm is Dynamically adjust the optimization stability according to the derivative value of the tracking error to ensure the convergence stability of the parameter update process under high-frequency disturbances.

[0022] Furthermore, the sliding mode control based on the sliding mode gain nonlinear tracking error feedback can be described as:

[0023]

[0024] Thus, the control law can be rewritten as:

[0025]

[0026] At the same time, in order to eliminate the chattering in the sliding mode control, the sgn function is replaced by the sat function;

[0027] where

[0028]

[0029] (3) Beneficial Effects

[0030] Compared with the prior art, the present invention provides a compound disturbance rejection method for an optoelectronic stabilization platform based on improved active disturbance rejection control, having the following beneficial effects:

[0031] Improved high-frequency disturbance estimation accuracy: The dynamic regularization constraint and dynamic damping co-optimization strategy of ALMRBF-ESO reduce the root mean square error (RMSE) under 2 kHz disturbance to 132.81, a 13.6% reduction compared to 153.71 of the traditional RBF method; Optimization of dynamic response speed: The sliding mode surface gain self-adaptation and boundary layer dynamic adjustment mechanism based on the sliding mode gain non-linear tracking error feedback shorten the step response rise time by 40%, while compressing the overshoot to ≤5.3%; Parameter drift suppression: Through spectrum coding reconstruction and L2 regularization gradient correction, the system frequency domain consistency index R2 is 0.9967, breaking through the bottleneck of high-frequency parameter sensitivity; High-frequency chattering suppression: The sliding mode surface design combined with the smooth switching of the sat function suppresses the high-frequency chattering of the sliding mode. Description of the Drawings

[0032] Figure 1 It is a schematic diagram of the overall architecture of the control system of the present invention;

[0033] Figure 2 It is the optimization flowchart of the ALMRBF-ESO network of the present invention;

[0034] Figure 3 It is the experimental result diagram of the disturbance rejection performance comparison of the present invention;

[0035] Figure 4 It is the experimental result diagram of the step response of the present invention. Detailed Embodiments

[0036] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0037] Embodiment

[0038] As Figures 1-4 shown, a compound disturbance rejection method for an optoelectronic stabilization platform based on improved active disturbance rejection control proposed in an embodiment of the present invention includes a disturbance estimation module, a sliding mode active disturbance rejection control module, and a compound control module, and includes the following steps.

[0039] S1. Deployment of the improved extended state observer:

[0040] (1) Principle derivation

[0041] The input values of the input layer nodes of the RBF neural network are generally the state variables of the system. Each hidden layer node contains a radial basis, that is, an activation function, and its output is

[0042]

[0043] where c j (t) is the center vector of the hidden layer node, ||x(t)-c j (t)|| is the Euclidean distance between the input vector and the center vector, b j is the width, and m represents the number of hidden layer nodes.

[0044] The output value of the output layer node is expressed as where w j is the weight factor of the hidden layer, h j (t) is the output value of the hidden layer, and its weighted sum gives the output value of the RBF neural network.

[0045] Replace the estimation of the unknown disturbance in the ESO part with the output of the RBF neural network, such as the ALMRBF-optimized ESO part in Figure 1 , to achieve ESO compensation. When the input of the neural network is the output of the neural network is where Substitute the output of the neural network into the original ESO formula to obtain the ESO based on the RBF neural network as

[0046] In the radial basis extended state observer (RBFESO), it is necessary to optimize the weight vector w, kernel center c j and kernel width b j of the RBF network to minimize the disturbance estimation error.

[0047] Let the true total disturbance of the controlled object be f(t), and the estimated value output by the RBF network be Then the instantaneous disturbance estimation error is defined as:

[0048]

[0049] where the input vector is composed of the tracking error e(t) and its derivative.

[0050] By minimizing the mean square error within the time window, the disturbance estimation accuracy is improved, that is

[0051] where, N is the length of the sampling window. When no regularization term is introduced, the objective function only focuses on the data fitting performance.

[0052] The Levenberg-Marquardt (LM) algorithm combines the advantages of the gradient descent method and the Gauss-Newton method, and its parameter update rule is Δθ = -(J T J + μI) -1 J T e, where: θ ∈ R 3m is the parameter vector to be optimized, including all network parameters w, c j , b j ; J ∈ R N×3m is the Jacobian matrix of the error with respect to the parameters; μ ≥ 0 is the damping factor; ∈ ∈ R N is the error vector within the window. The Jacobian matrix is composed of calculating the partial derivatives of the error with respect to each parameter: The parameter increment Δθ is jointly determined by the Jacobian matrix J, the damping factor μ, and the error vector ∈.

[0053] Update the parameters after each iteration: θ k+1 = θ k + Δθ; Adjust the damping factor according to the change of the error. If E k+1 < E k , accept the update and reduce the damping factor μ k+1 = μ k / 2 to enhance the Gauss-Newton characteristics; otherwise, reject the update, increase the damping factor μ k+1 = 2μ k , roll back the parameters to the previous state, and re-solve the increment.

[0054] To suppress the overfitting and parameter drift caused by the susceptibility to high-frequency noise interference during the parameter optimization process of the traditional RBF-ESO and further improve the hyperparameter robustness of the disturbance observer, an optimization architecture ALMRBF-ESO that combines adaptive L2 regularization constraints and dynamic damping is introduced. The optimization process is shown in Figure 2 Steps.

[0055] First, superimpose the regularization penalty terms of the weight parameter and the kernel center / width on the optimization objective function of the original LM algorithm to construct a composite error function: where, λ w , λ c , λ b control the regularization strengths of the weight vector, the kernel center, and the width respectively. Subsequently, introduce a regularized gradient correction term into the Jacobian matrix to expand the original partial derivative of the error with respect to the parameters to: J reg = J + λ w W + λ c C + λb B, where \(W\in R\) N×m and \(C\in R\) N×2m and \(B\in R\) N×m The regular gradient matrices corresponding to the weight, kernel center, and width respectively, where the \(j\)-th column elements are \(2w\) j and \(2c\) j and \(2b\) j respectively.

[0056] Under this correction condition, the parameter update rule is adjusted to: \(\Delta\theta = - (J\) T J+\mu I+\lambda w I w +\lambda c I c +\lambda b I b ) -1 (J T \in+\Lambda\theta)\) where the regularization coefficient matrix \(\Lambda = diag[\lambda w _1 w ,\lambda c _1 c ,\lambda b _1 b . By multiplying with the parameter vector \(\theta\), each parameter component is forced to shrink towards the low-energy direction. Further, to meet the optimization stability requirements under time-varying perturbation intensities, a dynamic normalization adaptive damping factor is proposed: When the error differential instantaneously increases, \(\mu adapt non-linearly increases to ensure iterative convergence, while when the system enters the steady state, \(\mu adapt rapidly decays to improve the convergence speed of the Gauss-Newton method. This method significantly improves the physical adaptability of the observer to complex perturbations while ensuring the parameter learning accuracy through a multiple dynamic constraint mechanism. See the attached Figure 3 for experimental data comparison.

[0057] (2) Parameter initialization

[0058] Set the number of hidden layer nodes of the RBF network to 10, the initial distribution of the kernel center parameter to \(C_0=\{0.5, 1.2\}\), and the initial kernel width to \(B_0 = [100, 500, 1000]\); Regularization coefficients: weight regularization \(\beta_1 = 0.01\), kernel center regularization \(\beta_2 = 0.1\), kernel width regularization \(\beta_3 = 1\); Dynamic damping configuration: reference damping factor \(\beta base = 0.01\), error differential gain \(\alpha = 0.05\), adjusted in real-time according to the formula \(\mu=\beta base +\alpha\cdot|de / dt|\); Spectrum encoding execution: The input perturbation signal is mapped to the frequency domain hidden layer space through the spectrum encoding unit to suppress medium and high frequency interference.

[0059] S2. Implementation of dynamic feedback control law:

[0060] (1) Principle derivation

[0061] To further improve the response speed of the controller, this paper combines sliding mode control with active disturbance rejection control, and introduces the idea of sliding mode control into the non-linear error control law in active disturbance rejection control, such as Figure 1 the sliding mode-based non-linear error feedback control part in

[0062] The system output gives e = y d - y, where y is the output signal and y d # is the desired output signal. The sliding mode control can be described as where c > 0, thus Thus, the control law can be rewritten as At the same time, in order to eliminate the chattering in sliding mode control, the sgn function is replaced by the sat function. Where Δ is the boundary layer of sliding mode control.

[0063] Finally, replace the fal(e, α, δ) function in the non-linear state error feedback function with g(e) = -ηsgn(s) to obtain Finally, the new sliding mode-based non-linear state error feedback (SMNLSEF) is obtained

[0064]

[0065] (2) Sliding mode surface parameter setting:

[0066] The initial sliding mode gain K p = 350, the power coefficient c2 = 3; the adaptive sliding mode surface update formula: The boundary layer thickness: based on the disturbance energy spectral density E f Adjust the saturation interval in real time: δ = δ base ·exp(-ε2·|E f - E th |), where δ base = 0.025, ∈2 = 1.

[0067] This method dynamically adjusts the slope of the response curve through the variable structure exponential reaching law (such as the slope change in the rising edge section in the appendix Figure 4 ), shortens the step response rise time by 40% (from 20 ms to 12 ms), and at the same time constructs a spectral energy functional constraint by combining the dual error suppression function to effectively suppress the chattering phenomenon under high-frequency disturbance excitation (such as the output waveform smoothness in the appendix Figure 4 ).

[0068] S3. Composite disturbance rejection signal synthesis:

[0069] After completing the settings of the extended state observer and the non-linear error feedback control in the active disturbance rejection control, parameter settings are made for other parts. Tracking differentiator (TD): cut-off frequency r0 = 500, filtering step size h0 = 0.001; ESO convergence: observer gain chain β3 = 18000, loop compensation factors k1 = 50, k2 = 675.

[0070] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A composite disturbance rejection method for an optoelectronic stabilization platform based on an improved active disturbance rejection control, including a double extended state observer module, a dynamic optimization feedback control module, and a collaborative disturbance rejection execution module, characterized in that: It includes the following steps: S1. Deployment of improved extended state observer: The dual extended state observation module constructs an improved extended state observer (ALMRBF-ESO) by fusing a dynamic regularized radial basis neural network and an adaptive damping factor adjustment mechanism within the active disturbance rejection control framework. It performs high-frequency phase compensation and low-frequency gain tuning on the multi-frequency disturbance signals received by the platform to achieve suppression of the disturbance observation error in the range of 10 Hz - 2 kHz; S2. Implementation of dynamic feedback control law: The dynamic optimization feedback control module reconstructs the control law equation based on the sliding mode gain non-linear tracking error feedback characteristic (SMNLSEF), dynamically adjusts the slope of the response curve through the variable structure exponential reaching law, and constructs a spectral energy functional constraint by fusing a dual error suppression function to achieve self-correction of the phase-frequency error of the control quantity output under high-frequency disturbance excitation; S3. Synthesis of composite disturbance rejection signal: A composite control signal is generated by fusing the dynamic weights of the ESO disturbance feedforward and the NLSEF feedback output, and the broadband disturbance error convergence is completed within 2 ms, reducing the system steady-state overshoot to 5.3% and the parameter drift rate to less than 0.12% / h.

2. A compound disturbance rejection method for an optoelectronic stabilization platform based on an improved active disturbance rejection control according to claim 1, characterized in that: In S1, the ALM-RBFESO adopts spectral coding to reconstruct the observation channel. The input disturbance signal is mapped to a preset frequency domain feature space through the RBF basis function, so that the disturbance energies in different frequency bands are weighted and separated in the hidden layer nodes, and combined with the Jacobian matrix correction term J reg = J + λ w W + λ c C + λ b The B iterative optimization parameter updates the direction to suppress the parameter drift problem caused by high-frequency disturbances.

3. A compound disturbance rejection method for an optoelectronic stabilization platform based on an improved active disturbance rejection control according to claim 1, characterized in that: In S1, the adaptive L2 regularization constraint of the ALMRBF-ESO optimization algorithm, and its composite error function is defined as: where λ w , λ c , λ b are the regularization intensity coefficients of the weight, kernel center, and width respectively, and balance the parameter update rate and global convergence stability through a dynamic normalization strategy.

4. A compound disturbance rejection method for an optoelectronic stabilization platform based on an improved active disturbance rejection control according to claim 1, characterized in that: The adaptive adjustment rule of the damping factor in the ALM-RBFESO optimization algorithm in S1 is Dynamically adjust the optimization stability according to the differential value of the tracking error to ensure the convergence stability of the parameter update process under high-frequency disturbances.

5. A compound disturbance rejection method for an optoelectronic stabilization platform based on an improved active disturbance rejection control according to claim 1, characterized in that: The sliding mode control of the SNNLSEF in S2 can be described as: Thus, the control law can be rewritten as: Meanwhile, in order to eliminate chattering in the sliding mode control, the sgn function is replaced by the sat function; Where

Images