A derivative-free trust region wavefront correction system and method with evaluation function switching

Abstract

A wavefront correction system and method with evaluation function switching and derivative-free trust region (RDR) is disclosed. This invention relates to the field of optimization and control technology for wavefront-free adaptive optics systems, specifically to a wavefront correction system and method with evaluation function switching and derivative-free trust region (RDR) optimization. The invention organically integrates a phased evaluation function strategy with a model-based derivative-free trust region optimization algorithm, achieving seamless switching of the evaluation function through observation data reuse. The method includes the following steps: simultaneously defining a first evaluation function and a second evaluation function; performing first-stage derivative-free trust region wavefront correction based on the evaluation function, while simultaneously constructing an observation data reuse buffer; switching to second-stage derivative-free trust region wavefront correction; selecting historical sampling points from the observation data reuse buffer, performing second-stage derivative-free trust region wavefront correction based on the observation data, and obtaining the optimal sampling point; updating the wavefront correction value based on the optimal sampling point.

Description

Technical Field

[0001] This invention relates to the field of optimization and control technology for wavefront-free adaptive optics systems, specifically to an evaluation function-switching derivative-free trust-domain wavefront correction system and method. Background Technology

[0002] Wavefront phase distortion caused by atmospheric turbulence is a key factor limiting the performance of free-space laser communication (FSOC) and astronomical imaging systems, and adaptive optics (AO) technology is a core approach to addressing this challenge. Traditional closed-loop AO systems rely on wavefront sensors for real-time detection and dynamic correction of wavefront distortion, but face engineering bottlenecks such as system complexity, difficult integration and maintenance, and high costs in low-cost, compact deployment scenarios. Sensorless adaptive optics (SLAO) eliminates the need for wavefront sensors, directly using observable performance indicators (evaluation functions) as feedback, and employs optimization algorithms to drive the wavefront corrector to achieve closed-loop control.

[0003] Since SLAO systems lack direct measurement of the wavefront phase, the evaluation function constitutes the core feedback information for sensing the system state and evaluating the correction effect during the closed-loop optimization process. Its mathematical characteristics, such as the distribution of local extrema, the smoothness of the response surface, and noise sensitivity, will significantly affect the convergence behavior, search efficiency, and final correction accuracy of the optimization algorithm. Therefore, the performance of SLAO systems largely depends on the co-design of the optimization algorithm and the evaluation function.

[0004] In terms of optimization algorithms, existing methods are mainly represented by Stochastic Parallel Gradient Descent (SPGD) and its improved algorithms. These methods estimate the gradient by applying random perturbations to the control parameters and combining this with changes in performance indicators. The gradient acquisition process relies on finite difference approximation, exhibiting strong randomness. Under low signal-to-noise ratio or strong turbulence conditions, performance indicators fluctuate significantly, easily leading to increased gradient estimation errors. This results in problems such as slow convergence speed, significant oscillations in the optimization process, and difficulty in meeting real-time closed-loop control requirements.

[0005] Regarding evaluation functions, existing SLAO systems mostly employ single performance metrics, such as the far-field Strehl ratio (SR), power-in-the-bucket (PIB), and second-order moment of light intensity. However, different evaluation functions exhibit significant differences at different stages of wavefront correction.

[0006] Among them, SR is highly sensitive to small changes in the wavefront and can effectively reflect the peak intensity in the far field, making it suitable for high-precision optimization. However, it is easily affected by noise under low signal-to-noise ratio or large aberration conditions, leading to instability in the optimization process. PIB has good smoothness and noise resistance, and can provide a stable global convergence trend, making it suitable for guiding the large dynamic range in the early stages of optimization. However, its insensitivity to spatial distribution details and peak changes limits the final correction accuracy. The second moment of light intensity characterizes the spatial diffusion of the spot energy distribution as a whole. By penalizing the energy distribution far from the center, it improves the compactness of the spot and suppresses the sidelobe structure, exhibiting good continuity and robustness. However, its response to changes in peak intensity is weak, and the gradient information it provides gradually weakens after entering the local optimization stage, thus limiting the system's ability to further converge to a high-quality focusing state.

[0007] In summary, existing methods generally suffer from the following problems: on the one hand, optimization algorithms based on stochastic gradient estimation have insufficient convergence efficiency and stability in complex environments; on the other hand, a single evaluation function is difficult to simultaneously take into account both global exploration and local fine optimization capabilities, thus restricting the improvement of the overall performance of wavefrontless sensor-based adaptive optics systems. Summary of the Invention

[0008] To address the aforementioned problems, the purpose of this invention is to propose a wavefront correction system and method with a switching evaluation function and a derivative-free trust region. This system organically integrates a phased evaluation function strategy with a model-based derivative-free trust region optimization algorithm, achieving seamless switching of the evaluation function through observation data reuse.

[0009] The system includes: a laser source, a turbulence simulator, a wavefront corrector, a focusing lens, a camera, and a wavefront controller; The laser beam emitted by the laser source is introduced into the wavefront distortion by the turbulence simulator, forming a distorted signal that is incident on the wavefront corrector. The wavefront corrector corrects the distorted signal according to the current wavefront correction amount. The corrected beam is focused into a far-field spot by a focusing lens and projected onto a camera. The camera acquires the far-field light intensity distribution image in real time and sends it to the wavefront controller. Based on the far-field light intensity distribution image, the wavefront controller performs evaluation function switching derivative-free trust region wavefront correction and outputs the updated wavefront correction amount to control the wavefront corrector.

[0010] A method for derivative-free trust-region wavefront correction with evaluation function switching, the method being implemented based on the aforementioned system, and comprising the following steps: S1. Simultaneously define the first evaluation function. Second evaluation function ; S2. Initialize the control framework, optimize parameters and observation data reuse cache, and set the evaluation function switching threshold. ; S3, based on Perform the first-stage derivative-free trust region wavefront correction, and simultaneously, based on Build a cache for reused observation data; S4. When the observation data is in the reuse cache The value is continuous Secondary satisfaction ≥ Then switch to the second-stage derivativeless trust region wavefront correction; S5. In the observation data reuse cache, according to Sort the values ​​in descending order and select the first few. The historical sampling point set consists of several data points. ; S6, based on Perform the second-stage derivativeless trust region wavefront correction to obtain the second evaluation function. The optimal correction parameters are used to update the wavefront correction.

[0011] Furthermore, the first evaluation function Defined as: the second moment of the far-field spot image with a fixed field of view center; Second evaluation function Defined as: the Strell ratio of the far-field light spot; and The values ​​are all calculated from the same frame of far-field light intensity distribution image.

[0012] Furthermore, in step S2, the control framework is initialized, specifically by selecting... dimensional Zernike mode coefficient vector As an optimization control variable, and set initial value ; Constructing a pattern response matrix ,in, The number of wavefront corrector actuators; Wavefront correction amount The formula for calculation is:

[0013] Initialize the optimization parameters as follows: Initialize the trust threshold of the trust domain The trust region radius increases by a certain percentage. Trust region radius reduction ratio ; The initialization parameters for the first-stage TRDF include: the initial trust region radius for the first stage. Minimum Trust Region Radius in Phase 1 The first stage has the maximum number of function evaluations. ; The initialization parameters for the second-stage TRDF include: the initial trust region radius for the second stage. Second stage minimum trust region radius The second stage is the maximum number of function evaluations. ; Initialize the observation data reuse cache, specifically by setting the cache list. ,in, The index value of the sampling point. For the first Optimization of control variable coordinates for each sampling point For the second evaluation function At parameter points The value at that location.

[0014] Furthermore, the basis Perform the first-stage derivative-free trust region wavefront correction, and simultaneously, based on Constructing a reuse cache for observation data includes the following steps; S31, in Endogenous generation Optimization control variable coordinates for each sampling point , ; S32, Traverse sequentially indivual Simultaneously calculate the first evaluation function value Second evaluation function value ( ),Will and ( Store in the cache list At the same time and Store the initial interpolation point set ; S33, Set the iteration counter = 0, iterate through steps S34 to S36 until the switching conditions for the second-stage trust region optimization are met: S34, based on ,by Building a first-stage proxy model for trust domain centers ,in, As the current trust domain iteration center, = 0 ; S35, Based on the current first-stage trust region radius Solve for the current candidate sampling points and based on Drive the wavefront corrector (3) and acquire far-field images, and simultaneously calculate the first evaluation function value. Second evaluation function value ( ),Will and ( Store in the cache list At the same time and Update to interpolation point set ; S36, based on Calculate the ratio of the predicted value to the actual measured value. and according to Update the first-stage trust region radius , among which, when = 0, .

[0015] Furthermore, in step S35, the current candidate sampling point The formula for calculation is: ,in, The step size vector within the trust region. For the first The local quadratic interpolation model of the next iteration.

[0016] Furthermore, in step S36, according to Calculate the ratio of the predicted value to the actual measured value. ; when Greater than the trust threshold of the trust domain At that time, increase the radius of the current trust region. and will Updated to ; Conversely, the radius of the current trust region is reduced. ,at the same time It remains unchanged.

[0017] Furthermore, in step S5, The range of values ​​for is: ; Based on historical sampling point set , with the greatest The sampling points corresponding to the values ​​are the centers of the trust regions, and the initial proxy model for the second stage is constructed. ,in, This is the index value of the historical sampling point.

[0018] Furthermore, with the maximum mentioned The sampling point corresponding to the value is the center of the initial trust region for the second-stage derivative-free trust region wavefront correction. Using the initial trust region radius, the second-stage derivative-free trust region wavefront correction is completed, the final trust region iteration center is output, and the optimal correction parameters are calculated based on the final trust region iteration center.

[0019] The method for further second-stage derivative-free trust region wavefront correction is consistent with steps S34 to S36.

[0020] The beneficial effects of the method described in this invention are as follows: (1) The synchronous dual evaluation function calculation and observation data reuse caching mechanism described in this invention. In each function evaluation of the first-stage Trust Region Differential-Free Optimization (TRDF), after the system acquires the far-field light intensity image, it simultaneously calculates the first evaluation function value (used to drive the current stage TRDF modeling and optimization) and the second evaluation function value (not used for optimization, only cached). The two evaluation functions are calculated based on the same frame of observation image, adding only negligible numerical calculation overhead and not introducing any additional physical sampling. At the same time, the coordinates of the control variables of the current sampling point and the second evaluation function value are stored together in the "observation data reuse cache" for direct use after switching.

[0021] (2) This invention is based on zero-sampling-cost surrogate model reconstruction using observation data reuse. When the switching criterion is met, the first-stage TRDF solution is terminated, and several historical sampling points (control variables + second evaluation function values) of the highest quality are extracted from the cache. These points are directly used as the initial interpolation point set for the second-stage TRDF to construct a surrogate model with the second evaluation function as the objective. This mechanism can reduce the possible modeling and sampling overhead caused by the switching of the evaluation function.

[0022] (3) The adaptive switching criterion based on performance triggering proposed in this invention. The switching criterion is based on the real-time judgment of the second evaluation function value of the synchronous cache, which can monitor the current system performance without additional sampling. When the latest second evaluation function value in the cache reaches the preset switching threshold and meets the stability condition, the switching is triggered immediately. The switching timing judgment is parallel to the first-stage TRDF optimization process, without introducing additional evaluation costs. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the system structure described in this invention; Figure 2 This is a flowchart of the method described in this invention; Among them, 1-laser source, 2-turbulence simulator, 3-wavefront corrector, 4-focusing lens, 5-camera, and 6-wavefront controller. Detailed Implementation

[0024] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1 This embodiment provides an evaluation function-switching derivative-free trust-domain wavefront correction system. The system is a wavefront-sensorless adaptive optics system, specifically including: a laser source 1, a turbulence simulator 2, a wavefront corrector 3, a focusing lens 4, a camera 5, and a wavefront controller 6. In this embodiment, the center wavelength of the laser source 1 is 632.8 nm, and the coherence length of the turbulence simulator 2 is... The wavefront corrector 3 employs a Solebre DMH40 series deformable mirror with 40 actuators, the focusing lens 4 has a focal length of 150mm, the camera 5 uses a Daheng Imaging MARS-251 series high-speed camera, and the wavefront controller 6 uses a high-performance computer and integrates the observation data multiplexing-driven evaluation function switching derivative-less trust-region wavefront correction method. It outputs the updated wavefront correction value to the wavefront corrector 3, forming a closed-loop control. The wavefront controller 6 includes an observation data multiplexing cache module for storing the coordinates of control variables and their corresponding second evaluation function values ​​for each historical sampling point.

[0026] like Figure 1 As shown, the beam emitted by the laser source 1 is introduced into the wavefront distortion by the turbulence simulator 2, forming a distortion signal and incident on the wavefront corrector 3. The wavefront corrector 3 corrects the distortion signal according to the current wavefront correction amount. The corrected beam is focused into a far-field spot by the focusing lens 4 and projected onto the camera 5. The camera 5 acquires the far-field light intensity distribution image in real time and sends it to the wavefront controller 6. Based on the far-field light intensity distribution image, the wavefront controller 6 performs evaluation function switching derivative-free trust region wavefront correction and outputs the updated wavefront correction amount to the wavefront corrector 3 to form a closed-loop control.

[0027] Example 2 This embodiment further defines Embodiment 1. Based on the system described in Embodiment 1, this embodiment provides an evaluation function-switching derivative-free trust-region wavefront correction method. Within the unified mode trust-region derivative-free optimization framework, it introduces an evaluation function switching mechanism based on observation data reuse. By employing different evaluation functions in stages during the optimization process, it achieves a continuous transition from global convergence guidance to local fine-tuning in the wavefront-free adaptive optics system, thereby improving the system's convergence efficiency and final correction accuracy. Figure 2 As shown, the method includes the following steps: S1. Simultaneously define the first evaluation function. Second evaluation function ; The relevant operations in step S1 will be introduced with specific examples: Define the first evaluation function This is used to drive the first stage of TRDF optimization. In this embodiment, the second moment of the far-field spot image with a fixed field of view center is used as the first evaluation function. The smaller the function value, the more concentrated the spot energy and the better the wavefront distortion correction effect. Specifically, the first evaluation function is defined as:

[0028] in, As a preset fixed reference center, For pixel coordinates, for Intensity value at pixel and They are respectively shaft and The index value of the axis.

[0029] Define the second evaluation function This is used to drive the second-stage TRDF fine optimization. In this embodiment, the Strell ratio (SR) of the far-field spot is used as the second evaluation function. SR is defined as the ratio of the actual far-field peak intensity to the peak intensity under ideal diffraction-limited conditions, reflecting the degree to which the optical system approaches the diffraction limit after wavefront correction. Its value range is (0,1], and the closer the value is to 1, the better the correction effect. The formula for calculating the second evaluation function is:

[0030] in The peak intensity of the far-field light spot in the current frame. This is the calibration value of the far-field peak intensity when the system has no aberrations.

[0031] and All from the same frame of far-field light intensity image The calculations show that both involve only numerical processing of the acquired images, without requiring additional physical sampling or measurement operations. This characteristic is the physical basis of the observation data reuse mechanism of this invention.

[0032] S2. Initialize the control framework, optimize parameters and observation data reuse cache, and set the evaluation function switching threshold. ; The relevant operations in step S2 will be introduced with specific examples: Select an n-dimensional Zernike mode coefficient vector As an optimization control variable, specifically in this embodiment... This corresponds to Zernike polynomials of orders 3 to 37, after removing piston and tilt terms. Construct the mode response matrix. (M is the number of wavefront corrector actuators), satisfying , This is the wavefront corrector voltage control vector, i.e., the wavefront correction amount. Initialization mode control vector. .

[0033] Set trust threshold for trust domain The trust region radius increases by a certain percentage. Trust region radius reduction ratio In this embodiment, , , Set the first-stage TRDF parameters: initial trust region radius. Minimum trust region radius The first stage has the maximum number of function evaluations. In this embodiment , , .

[0034] Set the second-stage TRDF parameters: initial trust region radius Minimum trust region radius The second stage is the maximum number of function evaluations. In this embodiment , , .

[0035] Initialize observation data reuse cache: cache list Used to store the coordinates of control variables and their corresponding values ​​for each historical sampling point. Value, in the format of ,in, The index value of the sampling point. For the first Optimization of control variable coordinates for each sampling point For the second evaluation function At parameter points The value at that point is the far-field light intensity distribution image acquired from that point. Calculated.

[0036] S3, based on Perform the first-stage derivative-free trust region wavefront correction, and simultaneously, based on Build a cache for reused observation data; The relevant operations in step S3 will be introduced with specific examples: S31, in Endogenous generation Optimization control variable coordinates for each sampling point , In this embodiment, n=35, which is the Zernike dimension involved in the optimization control; For each sampling point The driving voltage is calculated and applied (wavefront correction value) to wavefront corrector 3, while the far-field light intensity distribution is acquired. ; S32, Traverse sequentially indivual Based on the synchronously acquired far-field light intensity distribution (observed image), the first evaluation function value is calculated synchronously. Second evaluation function value ( ),Will and ( Store in the cache list ( ), and at the same time and Store the initial interpolation point set , , The index value of the sampling point in the interpolation point set. For the interpolation point set, the first One optimization control variable, , The first evaluation function ,against Far-field light intensity distribution image acquired at [location] The calculation results; Complete traversal indivual Execute step S33; S33, Set the iteration counter = 0, repeat S34-S36 until the termination condition of the first stage trust region optimization is met: S34, based on ,by Building a first-stage proxy model for trust domain centers In particular, As the current trust domain iteration center, = 0 ;

[0037] in, For the proxy model, For the first The center point of the next trust region iteration is defined as the point up to the [number]th iteration. The optimal control vector obtained at the end of the first iteration. For gradient estimation, To approximate the Hessian matrix, all are uniquely determined through interpolation conditions and satisfy the following: .

[0038] S35, Based on the current first-stage trust region radius Initiate the trust region derivative-free optimization iteration to solve for the current candidate sampling point. and based on The wavefront corrector 3 is driven and far-field images are acquired, while the first evaluation function value is calculated simultaneously. Second evaluation function value ( ),Will and ( Store in the cache list ( ), and at the same time and Update to interpolation point set Specifically, replace with new dots. Find the point with the worst function value in the point set and store the replaced point set. ; Perform the following operations during each iteration: Solving the trust domain subproblem: Based on the current agent model In the current trust region radius Under constraints, solve the subproblem:

[0039] Candidate sampling points Special ,in, The step size vector within the trust region. As the current trust domain center, It is the optimal point obtained from the previous trust domain optimization step. For the first Local quadratic interpolation model for the next iteration Used to find the value of the independent variable that minimizes the objective function.

[0040] S36, Based on the first evaluation function value The ratio of the predicted decline to the actual decline calculated by the model. Based on this, update the trust region radius for the next iteration of the first phase. :

[0041] in, It refers to the performance index value obtained through actual physical measurements during the trust region iterative optimization process after applying a control vector to the control system. Specifically, this applies to the first stage of trust region iterative optimization. In the second stage of trust region iterative optimization .

[0042] when Greater than the trust threshold of the trust domain At that time, increase the radius of the current trust region. And Updated to ; Conversely, the radius of the current trust region is reduced. ,at the same time It remains unchanged.

[0043] Update the iteration counter: +1.

[0044] Preferably, in this embodiment , , .

[0045] S4. When the observation data is in the reuse cache The value is continuous Secondary satisfaction ≥ Then switch to the second-stage derivativeless trust region wavefront correction; The relevant operations in step S4 will be introduced with specific examples: Evaluation function switching criteria: Check the latest cached data. Value, if continuous Secondary satisfaction ≥ If the first phase function evaluation count reaches a certain threshold, a switch is triggered; otherwise, the first phase iteration continues (returning to S3). Regardless of whether a switch is triggered, the first phase of TRDF is terminated. Specifically, this embodiment sets the number of consecutive determinations. To avoid the risk of false triggering caused by noise in a single measurement, the threshold is switched. .

[0046] S5. In the observation data reuse cache, according to Sort the values ​​in descending order and select the first few. The historical sampling point set consists of several data points. ; The relevant operations in step S5 will be introduced with specific examples: Several historical sampling points are extracted from the cache for the initial construction of the proxy model in the second stage. The selection principle is as follows: priority is given to selecting... Points with larger SR values ​​(i.e., higher SR) are located near the high-performance region in the control space, have good interpolation geometry, and result in higher quality surrogate models; the number of points selected... satisfy Balancing modeling quality and computational cost; if the number of cache points is insufficient Then the current best point Additional sampling is performed at the initial point. Specifically, in this embodiment... This reduces the number of additional function evaluations required for initial TRDF modeling and improves overall optimization efficiency.

[0047] In the observation data reuse cache, select Points with larger values ​​are grouped into a historical sampling point set. , This is the index value of the historical sampling point; The selection in the observation data reuse cache The specific method for each historical sampling point is as follows: Historical data in the cache is processed according to the second evaluation function. Sort the values ​​in descending order and select the values ​​that appear at the top of the list. The historical sampling point set consists of several data points. .

[0048] The selection Points with larger values ​​refer to historical sampling points selected in the optimization space corresponding to areas with better link performance.

[0049] This selection method allows for the construction of a surrogate model for the second stage using high-quality data accumulated during the first-stage optimization process. This significantly reduces the number of iterations in the second stage while ensuring the model's prediction accuracy, achieving a smooth transition and efficient convergence during the evaluation function switching process.

[0050] S6, based on Perform the second-stage derivativeless trust region wavefront correction to obtain the second evaluation function. The optimal correction parameters are used to update the wavefront correction.

[0051] The relevant operations in step S6 will be introduced with specific examples: Based on historical sampling point set , with the greatest The sampling points corresponding to the values ​​are used to construct the second-stage initial proxy model for the trust region center. The model construction process is consistent with the proxy model construction method in the first stage of step S4.

[0052] In cache The most valuable historical point To be the optimal starting point for the second stage of TRDF, with The initial trust region radius is used to complete the second-stage derivative-free trust region wavefront correction.

[0053] Based on the second-stage initial agent model Using the second evaluation function To optimize the objective, the function evaluation budget is... Internal TRDF iterative optimization is performed (second-stage derivative-free trust region wavefront correction).

[0054] The correction process is consistent with the trust region optimization process in step S4, including: solving the trust region sub-problem, updating the interpolation point set, calculating the ratio of actual improvement to predicted improvement, adjusting the trust region radius, and updating the surrogate model.

[0055] In each iteration, physical sampling is performed on candidate points, and calculation is performed. The model is updated and its values ​​are updated until the stopping condition is met.

[0056] Record the optimal Zernike coefficient vector for the convergence of the second-stage TRDF. , denoted as the optimal sampling point And based on this, the deformable mirror driving voltage (the optimal correction parameter) is calculated. It is then output to wavefront corrector 3 to complete wavefront correction under the current atmospheric turbulence conditions.

[0057] This invention achieves a continuous transition from global convergence guidance to local fine-grained optimization by introducing an evaluation function switching mechanism based on observation data reuse within a unified trust region derivative-free optimization framework.

[0058] According to the specific embodiments described above, this invention proposes an observation data reuse-driven evaluation function switching-based derivative-less trust region wavefront correction method. Within the unified mode trust region derivative-less optimization framework, it introduces an evaluation function switching mechanism based on observation data reuse. By employing different evaluation functions in stages during the optimization process, it achieves a continuous transition from global convergence guidance to local fine-tuning in a wavefront-less adaptive optics system. This method effectively improves system convergence efficiency and final correction accuracy while ensuring optimization continuity. Compared to traditional SPGD-type wavefront-less optimization methods, this invention overcomes the problems of slow convergence speed, susceptibility to local extrema, and high parameter sensitivity in strong turbulence and high-dimensional control scenarios. It significantly reduces the number of function evaluations required to achieve effective convergence while improving the system's adaptability to different turbulence intensities and overall control stability.

[0059] The trust-region-free derivative framework employed in this invention introduces a surrogate model as an intermediate structure, which introduces a technical contradiction to index switching that is fundamentally absent in SPGD-like methods: the construction and updating of the surrogate model depends on the continuous accumulation of evaluation function values ​​at each sampling point. After switching, the function values ​​of the new index at existing sampling points are unknown, making the surrogate model unusable. Forced switching would cause the model to fail, while resampling would introduce significant additional overhead. This invention fundamentally resolves this contradiction through synchronous dual-index calculation and a caching mechanism, achieving seamless reconstruction of the surrogate model rather than simply stacking methods.

[0060] It should be understood that the specific embodiments described herein are merely some examples of the present invention and are not intended to limit the present invention. Any modifications made in accordance with the technical concept proposed in this invention should be included within the scope of protection of this invention.

Claims

1. A wavefront correction method with evaluation function switching and no derivative trust region, characterized in that, The method includes the following steps: S1. Simultaneously define the first evaluation function. Second evaluation function ; S2. Initialize the control framework, optimize parameters and observation data reuse cache, and set the evaluation function switching threshold. ; S3, based on Perform the first-stage derivative-free trust region wavefront correction, and simultaneously, based on Build a cache for reused observation data; S4. When the observation data is in the reuse cache The value is continuous Secondary satisfaction ≥ Then switch to the second-stage derivativeless trust region wavefront correction; S5. In the observation data reuse cache, according to Sort the values ​​in descending order and select the first few. The historical sampling point set consists of several data points. ; S6, based on Perform the second-stage derivativeless trust region wavefront correction to obtain the second evaluation function. The optimal correction parameters are used to update the wavefront correction.

2. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 1, characterized in that, First evaluation function Defined as: the second moment of the far-field spot image with a fixed field of view center; Second evaluation function Defined as: the Strell ratio of the far-field light spot; and The values ​​are all calculated from the same frame of far-field light intensity distribution image.

3. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 2, characterized in that, In step S2, the control framework is initialized, specifically by selecting... dimensional Zernike mode coefficient vector As an optimization control variable, and set initial value ; Constructing a pattern response matrix ,in, The number of actuators for wavefront corrector (3); Wavefront correction amount The formula for calculation is: Initialize the optimization parameters as follows: Initialize the trust threshold of the trust domain The trust region radius increases by a certain percentage. Trust region radius reduction ratio ; The initialization parameters for the first-stage TRDF include: the initial trust region radius for the first stage. Minimum Trust Region Radius in Phase 1 The first stage has the maximum number of function evaluations. ; The initialization parameters for the second-stage TRDF include: the initial trust region radius for the second stage. Second stage minimum trust region radius The second stage is the maximum number of function evaluations. ; Initialize the observation data reuse cache, specifically by setting the cache list. ,in, The index value of the sampling point. For the first Optimization of control variable coordinates for each sampling point For the second evaluation function At parameter points The value at that location.

4. The evaluation function switching type derivativeless trust region wavefront correction method according to claim 3, characterized in that, The basis Perform the first-stage derivative-free trust region wavefront correction, and simultaneously, based on Constructing a reuse cache for observation data includes the following steps; S31, in Endogenous generation Optimization control variable coordinates for each sampling point , ; S32, Traverse sequentially indivual Simultaneously calculate the first evaluation function value Second evaluation function value ( ),Will and ( Store in the cache list At the same time and Store the initial interpolation point set ; S33, Set the iteration counter = 0, iterate through steps S34 to S36 until the switching conditions for the second-stage trust region optimization are met: S34, based on ,by Building a first-stage proxy model for trust domain centers in, As the current trust domain iteration center, = 0 ; S35, Based on the current first-stage trust region radius Solve for the current candidate sampling points and based on Drive the wavefront corrector (3) and acquire far-field images, and simultaneously calculate the first evaluation function value. Second evaluation function value ( ),Will and ( Store in the cache list At the same time and Update to interpolation point set ; S36, based on Calculate the ratio of the predicted value to the actual measured value. and according to Update the first-stage trust region radius , among which, when = 0, .

5. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 4, characterized in that, In step S35, the current candidate sampling point The formula for calculation is: ,in, The step size vector within the trust region. For the first The local quadratic interpolation model of the next iteration.

6. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 5, characterized in that, In step S36, according to Calculate the ratio of the predicted value to the actual measured value. ; when Greater than the trust threshold of the trust domain At that time, increase the radius of the current trust region. and will Updated to ; Conversely, the radius of the current trust region is reduced. ,at the same time It remains unchanged.

7. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 6, characterized in that, In step S5, The range of values ​​for is: ; Based on historical sampling point set , with the greatest The sampling points corresponding to the values ​​are the centers of the trust regions, and the initial proxy model for the second stage is constructed. ,in, This is the index value of the historical sampling point.

8. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 7, characterized in that, With the maximum The sampling point corresponding to the value is the center of the initial trust region for the second-stage derivative-free trust region wavefront correction. Using the initial trust region radius, the second-stage derivative-free trust region wavefront correction is completed, the final trust region iteration center is output, and the optimal correction parameters are calculated based on the final trust region iteration center.

9. The evaluation function-switching derivative-free trust region wavefront correction method according to claim 8, characterized in that, The method for wavefront correction without derivatives in the second stage is the same as steps S34 to S36.

10. The evaluation function-switching derivative-free trust region wavefront correction system according to claim 9, characterized in that, The system is used to implement the method of any one of claims 1-9, and the system includes: a laser source (1), a turbulence simulator (2), a wavefront corrector (3), a focusing lens (4), a camera (5), and a wavefront controller (6). The beam emitted by the laser source (1) is introduced into the wavefront distortion by the turbulence simulator (2), forming a distorted signal and incident on the wavefront corrector (3). The wavefront corrector (3) corrects the distorted signal according to the current wavefront correction amount. The corrected beam is focused into a far-field spot by the focusing lens (4) and projected onto the camera (5). The camera (5) acquires the far-field light intensity distribution image in real time and sends it to the wavefront controller (6). Based on the far-field light intensity distribution image, the wavefront controller (6) performs evaluation function switching derivative-free trust region wavefront correction and outputs the updated wavefront correction amount to control the wavefront corrector (3).

Images